Patent Application
20230325670
Once deployed in the field, Deep Neural Networks (DNNs) run on devices with widely different compute capabilities and whose computational load varies over time. Pruning, early exits, and dynamic DNNs are techniques aimed at running the DNNs efficiently, but these techniques have drawbacks: pruned models cannot handle time-varying load conditions; early exits are not easily integrated into complex architectures; the execution time of dynamic DNNs cannot be foreseen; and none of these solutions respect the legacy of pre-trained DNNs. There is a need for addressing these issues and/or other issues associated with the prior art.
Embodiments of the present disclosure relate to augmenting legacy neural network models for flexible inference. Systems and methods are disclosed that dynamically configure and execute an augmented neural network in real-time according to performance constraints while also maintaining the legacy neural network execution path. A neural network model that has been trained for a task is augmented by pairing a low-compute “shallow” phase with each legacy phase. The legacy phases of the neural network model are held constant (e.g., unchanged) while the shallow phases are trained. During inference, one or more of the shallow phases can be selectively executed in place of the corresponding legacy phase. Compared with the legacy phases, the shallow phases are typically less accurate, but have reduced latency and consume less power. Therefore, processing using one or more of the shallow phases in place of one or more of the legacy phases enables the augmented neural network to dynamically adapt to changes in the execution environment (e.g., processing load or performance requirement).
Conventional techniques for reducing latency and energy consumption of neural networks include modifying the neural network (e.g., pruning) and do not preserve the original neural network. For dynamic neural networks, different versions of the neural network are constructed and trained for specific tasks and the original neural network is not preserved. Switching between the different neural networks is typically not possible in real-time due to the overhead needed to initialize each neural network with the corresponding weights. In contrast, the augmented neural network model does not require training for the specific tasks and can be dynamically reconfigured in real-time using “switches”. Conventional slimmable networks have configurable activation channel widths and also fail to preserve the legacy neural network. Specifically, training different slimmable widths changes the weights for the full network configuration.
In an embodiment, the method includes receiving inputs to an augmented neural network, where the augmented neural network comprises a pre-trained legacy neural network including legacy processing phases and associated legacy parameters and at least one of the legacy processing phases is paired with a shallow processing phase and associated shallow parameters. A processing path through the augmented neural network model is dynamically modified, according to constraints, to provide processing paths that each include one or more of the at least one shallow processing phases to process the inputs and produce outputs. The augmented neural network is trained by adjusting the shallow parameters without changing the legacy parameters to match a distribution of the output compared with a legacy distribution of legacy outputs produced by processing the inputs using only the legacy processing phases.
The present systems and methods for augmenting legacy neural network models for flexible inference are described in detail below with reference to the attached drawing figures, wherein:
Systems and methods are disclosed that augment legacy neural network models for flexible inference. A neural network model that has been trained for a task is augmented with one or more low-compute “shallow” phases that are paired with pre-trained legacy phases. The legacy phases of the neural network model are held constant (e.g., legacy parameters are unchanged) while the shallow phases are trained. During inference, one or more of the shallow phases can be selectively executed in place of the corresponding legacy phase. Compared with the legacy phases, the shallow phases are typically less accurate, but have reduced latency and consume less power. Therefore, processing using one or more of the shallow phases in place of one or more of the legacy phases enables the augmented neural network to dynamically adapt to changes in the execution environment (e.g., processing load or performance requirement). The augmented neural network may be dynamically configured and executed in real-time according to constraints while the legacy neural network execution path is preserved.
Deep neural networks (DNNs) deliver state-of-the-art results in a wide range of applications. Very often, however, they are characterized by a high inference cost; when pushed to their limit, inefficient DNNs with large latency waste computational resources for little performance gains. The reasons for these inefficiencies are varied. Large or deep DNNs require high compute at inference time. Efficient, static architectures such as residual connections, pixel shuffle, or pruning partially alleviate the inefficiencies, but low utilization of computational resources is still possible when a DNN overthinks about easy cases. Furthermore, deploying DNNs onto a wide range of devices adds a layer of complexity, as different computing systems require diverse optimal implementations of the same DNN. Even the computational load on the same device changes over time. These issues can be alleviated by affording a DNN the ability to vary its architecture on-the-fly, based on current system state. Thus, providing the capability of dynamic configuration is highly desirable for practical implementation and field deployment.
Conventional dynamic DNNs are models that incorporate a gating policy function to route an input through paths with varying compute complexity. Different paths specialize during training to handle different classes of data, making dynamic DNNs highly accurate and efficient. At the same time, the inclusion of the routing policy into the DNN prevents direct control by a user over the computational cost for a given input. Anytime prediction uses early exits that allow selecting among a limited set of compute/accuracy compromises, but cannot be easily adapted to non-trivial architectures, like U-nets. All-for-one DNNs are initially trained to be flexible, i.e., such that sub-networks can be run while maintaining high accuracy. The All-for-one DNNs greatly simplify the shipping of a model to different devices, but the DNN eventually deployed consists of one of the sub-networks, further fine-tuned for the target device. Therefore, the deployed All-for-one DNN has a static architecture that prevents run time adaptation to available system resources. All of these conventional networks require training a new network without preserving an original model. Overall, none of the conventional techniques work with all DNN architectures, provide full real time control of the DNN configuration, and preserve the legacy network.
In contrast, LeAF (Legacy Augmented for Flexible inference) is a paradigm to transform pre-trained, static DNNs into flexible ones, while also preserving the ability to run the legacy neural network model. Neural network models augmented according to LeAF disentangle the gating policy problem from the computational aspects, enabling selection of the optimal accuracy/compute cost compromise on-the-fly. The LeAF paradigm trains a flexible augmented neural network model whose architecture can be changed on-the-fly according to a constraint imposed by the state of the compute device used to execute the augmented neural network model. One or more performance constraints may be provided as an input to dynamically select a routing configuration of a processing path through the augmented neural network model. In an embodiment, the routing policy can be explicitly controlled by a user and changed on-the-fly to meet constraints of the compute device. Through dynamic configuration, the augmented neural network model may adapt to real-time changes in the performance constraints.
The cost of execution of the augmented neural network model may be reduced by using more shallow phases during high-load scenarios while taking advantage of low-load scenarios to use legacy phases to perform computations. The approach can also be used to shift execution resources to more critical inference tasks based on environmental conditions. For example, for a system deployed within a vehicle, execution resources for a pedestrian detection system can be reduced when travelling on an expressway and vehicle detection processing can be increased to ensure collision avoidance systems operate at peak capability. In an embodiment, runtime is a resource constraint in real-time systems. In an embodiment, performance constraints include a metric, such as inferencing execution time (latency), floating-point operations per second (FLOPS), or energy consumption and a target value for the metric.
In the context of the following description, modifications implemented via augmentation reduce computations performed by the augmented neural network model compared with the original neural network model. In an embodiment, the additional shallow phases increase the storage space consumed for the augmented neural network model by 13.8% compared with the original (legacy) neural network model. In an embodiment, the augmented neural network model has an accuracy of 76.1% to 64.8% compared with the accuracy of the original (legacy) neural network model while requiring 4 to 0.68 GFLOPs.
More illustrative information will now be set forth regarding various optional architectures and features with which the foregoing framework may be implemented, per the desires of the user. It should be strongly noted that the following information is set forth for illustrative purposes and should not be construed as limiting in any manner. Any of the following features may be optionally incorporated with or without the exclusion of other features described.
The augmented legacy neural network system 100 includes a routing configuration generator 105 and an augmented neural network 125 that includes switches 120, and N shallow phases 115 paired with N legacy phases 110. A neural network model comprising the legacy phases 110-0 through 110-N+1 that has been trained for a task is augmented with the shallow phases 115-1 through 115-N. Parameters of the shallow phases 115 (shallow parameters) are then trained while parameters of the legacy phases 110 (legacy parameters) are held constant (e.g., frozen). The shallow parameters and legacy parameters are applied to the input tensor by the corresponding shallow and legacy processing phases to produce the output.
The routing configuration generator 105 receives constraints and determines a routing configuration based on the constraints. In an embodiment, the constraints are user selected or user specified. In an embodiment, the routing policy implemented by the routing configuration generator 105 is independent from the task performed by the augmented neural network 125. The routing configuration (external routing policy) may be any user-defined function of the state of the execution environment. The switches 120 are dynamically configured according to the routing configuration to control the execution path through the augmented neural network 125 for processing input tensors to produce outputs. As shown in
In an embodiment, at least one legacy phase 110 is augmented with a shallow phase 115, but in other embodiments several or all legacy phases are augmented. The augmentations that are used to implement a shallow phase 115 corresponding to a legacy phase 110 may include removal of a layer, changing a layer to perform an identity operation, reduction of a number of channels input to a layer (equivalent to zeroing the weight planes for a given input), reduction of embedded categories output by a layer (equivalent to removing the weights associated with the target dimension), reduction of a sampling scale factor of activation height and width between layers, and/or removal of an output from a layer (equivalent to removing an entire weight from a layer). Performing an identity operation can be implemented by replacing the computation a layer performs with a data transfer or modifying the data reference used by the following layers.
In an embodiment, compared with the legacy phase 110 the corresponding shallow phase 115 only processes a reduced number of channels. In an embodiment, the reduced number of channels is predetermined and may be different for each shallow phase 115. In an embodiment, inputs and outputs of the shallow phase 115 are compatible (impedance matched) with the inputs and outputs of the corresponding legacy phase 110. Compared with the legacy phases 110, the shallow phases 115 are typically less accurate, but have reduced latency and consume less power. Therefore, processing using one or more of the shallow phases 115 in place of one or more of the legacy phases 110 enables the augmented legacy neural network system 100 to dynamically adapt to changes in the execution environment (e.g., processing load or performance metric). In contrast with conventional techniques, a user may define the routing policy, the user may directly provide the constraints, and the legacy processing path is preserved.
At step 135, parameters of the shallow phases 115 are trained for all possible routing configurations during a first training pass. Parameters of the legacy phases 110 remain unchanged during step 135. In an embodiment, parameters of the shallow phases 115 are trained for a portion of all possible routing configurations. A goal of training the augmented neural network 125 is for the output of the augmented neural network 125 to have the same statistical distribution as the original legacy neural network. In this way, any additional module that receives the output of the augmented neural network 125 as an input, potentially does not need to be retrained, as the distribution of the data output for the augmented neural network 125 is the same for a processing path configured via routing through either only the legacy phases 115 or a combination of the legacy phases 110 and at least one of the shallow phases 115. In an embodiment, training can also enforce the same distribution between the legacy neural network configuration and other neural network configurations at the output of each legacy phase 110 or corresponding shallow phase 115. An additional distribution cost term, such as Kullbach-Leibler divergence, may be used between the output of the legacy neural network (or legacy phase 110) and the output of the augmented legacy neural network (or shallow phase 115).
At step 140, the augmented legacy neural network system 100 performs inference for a baseline set of routing configurations. During step 140, input tensors from a dataset including the input tensors and corresponding expected outputs is processed through the augmented neural network 125 using the baseline set to predict outputs. In an embodiment, the batch sizes of the inputs are fixed during the first training pass and step 140. The predicted outputs are compared with the legacy outputs for the same input tensors to characterize the loss in accuracy for each routing configuration and batch size. At step 145, a characterization loss function is computed to characterize the loss in accuracy for the baseline set of routing configurations.
At step 150, a subset of the configurations is selected based on the accuracies. A subset of the routing configurations (Pareto set) is selected that satisfies a threshold level of accuracy. In an embodiment, the selection of the Pareto set uses system performance and batch size as part of the selection criteria. The augmented neural network 125 is then characterized for the subset of routing configurations to map the constraints to the routing configurations.
At step 155, a second training step fine tunes the augmented neural network 125 (via forward and backward propagation) for the subset of the routing configurations and batch sizes. Importantly, only the shallow parameters are adjusted (via backpropagation) based on the characterization loss function for the subset of the routing configurations. For all of the steps shown in method 130, the legacy parameters are preserved. During inference, one or more constraints may be provided as an input to dynamically select the routing configuration for the augmented neural network 125. Through dynamic configuration of the execution path through the augmented neural network 125, the augmented legacy neural network system 100 may adapt to real-time changes in the constraints. However, the legacy (before augmentation) neural network model is preserved within the augmented neural network 125.
The training and characterization unit 210 may use the method 130 to train and characterize the augmented neural network 125 for the subset of routing configurations and map the constraints to the routing configurations. Given N shallow phases 115, the number of possible execution paths is 2N, including the original legacy execution path. During execution, any policy function τ=π(s) can be used to generate the routing configurations and control the set of active shallow phases 115, where s is the current state of the device executing the augmented legacy neural network system 100. In an embodiment, s comprises the current processing load and or target accuracy. In an embodiment, τ is a binary vector—therefore, reconfiguration overhead is minimal, allowing routing configuration updates in real time for any input tensor.
The training and characterization unit 210 freezes the legacy parameters θ while the shallow parameters ω are trained. Holding the legacy parameters constant significantly decreases the overhead of training the augmented neural network 125 which has multiple selectable processing paths, while also preserving the ability to execute the original legacy neural network. To preserve accuracy of the original legacy neural network, statistics in batch normalization layers may also be frozen while the shallow parameters are trained. In an embodiment, the training and characterization unit 210 trains the shallow parameters in two steps, the first (baseline) training step and the second fine-tuning training step. Before training the shallow parameters, in an embodiment, the weights co of the shallow phases are initialized by sampling from the original weights θ to improve stability during training.
The training and characterization unit 210 receives a dataset that includes input tensors and expected outputs. In an embodiment, during the first training step, the dataset is processed for all possible routing configurations and/or the subset of the routing configurations to characterize the augmented neural network model 125. In an embodiment, instead of applying all routing configurations, the routing configurations are randomly sampled to provide a portion of all of the possible routing configurations for the first training step. In an embodiment, the shallow parameters are trained through minimization of a loss function:
J
tot(ω;θ,X,y)=Σi=0|τ|J(ω;τi,θ,X,y)·αi, Eq. (1)
where J(ω; τi, θ, X, y) is the traditional loss (e.g., cross entropy) computed for the execution profile τi and αi is a multiplicative factor corresponding to the fraction of time a path is expected to run. The first training step uses all 2N routing configurations of the augmented neural network 125 with αi=1/(2N)∀i. In an embodiment, αi=1/|τ| (so that any path has the same probability of being executed), but other choices can be made to give more importance (and consequently increase the accuracy) of a specific configuration τi.
Although equation (1) leads to a usable, flexible neural network, a better result may be achieved by adjusting the loss to compute an adaptive loss as follows:
where [J(ω; τi, θ, X, y)] is the sample average computed over the last 100 training iterations. The rationale behind the adaptive loss is that different neural network configurations are characterized by different levels of noise. While the loss in equation (1) tries to achieve the same level of accuracy for any routing configuration, the adaptive loss in equation (2) normalizes each routing configuration loss by its expected value and thus in practice targets the same signal to noise ratio for any routing configuration, which is theoretically correct in the context of maximum likelihood optimization. Therefore, in the forward pass, for each batch, all routing configurations are applied and the loss in equation (2) is accumulated before an optimization step is performed.
Once the first training step is complete, the training and characterization unit 210 may characterize the augmented neural network 125 by measuring the accuracy and computing a cost for each routing configuration, including a target batch size. In an embodiment, the cost of compute may be derived analytically in the case of FLOPs, or experimentally in the case of latency, energy, or power consumption. Including routing configurations which achieve a sub-optimal accuracy/FLOPs ratio in the cost function of equation (2) is detrimental to the final accuracy. Therefore, in an embodiment, the routing configurations in the Pareto set are identified and for the routing configurations that do not reside in the Pareto set, αi=0 for a fine-tuning training set. The Pareto set routing configurations may be stored by the training and characterization unit 210 with the associated final accuracy and costs in a look-up table that can be accessed by the routing configuration generator 105 to implement the policy gating function τ=π(s).
In an embodiment, during characterization, inferencing latency (e.g., execution time) for the augmented neural network model 110 is measured and the measured latencies are binned according to the target execution times specified by the constraints. Similarly, the characterization process may also measure additional performance metrics and the measured values are also binned according to the target metric values. When the measurements are completed, routing configurations are selected for the specific target metrics associated with each bin. The configuration settings having the highest accuracy that meets the target performance metric are identified for each bin. The identified routing configurations may be stored in a configuration table for access by the routing configuration generator 105.
Each constraint may be defined as a type of constraint or metric and a target value for the metric. The routing configuration generator 105 allows the augmented legacy neural network system 100 to support multiple types of metrics such as energy and runtime using a single configuration table. In an embodiment, the configuration table comprises a lookup table structure that outputs configuration settings based on a precomputed range for each metric that bounds the metric value. In other words, the configuration table stores information defining metric bins corresponding to the configuration settings and that contain the metric values.
In an embodiment, the routing configurations may control switches that partially disable and enable each legacy phase 110, replacing a subset of the legacy phase 110 with a corresponding subset of a shallow phase rather than replacing the entire legacy phase 110 with the shallow phase 115.
The phase output 330 comprises outputs of the legacy phase common 310 and either output of either the legacy phase subset 325 or shallow phase subset 315. The output of the legacy phase common 310 and either output of either the legacy phase subset 325 or shallow phase subset 315 may be concatenated to provide the phase output 330 as an input to the next processing phase.
At step 365, inputs to an augmented neural network are received, where the augmented neural network comprises a pre-trained legacy neural network including legacy processing phases and associated legacy parameters and at least one of the legacy processing phases is paired with a shallow processing phase and associated shallow parameters. In an embodiment, the augmented neural network is the augmented legacy neural network system 100. In an embodiment, one or more augmented processing phases in the augmented neural network 125 are replaced with the hybrid phase 300. In an embodiment, the shallow parameters are initialized based on at least a portion of the legacy parameters. In an embodiment, the portion uses the legacy parameter values (bits per-channel or entire channels) or samples values from a distribution informed by the legacy parameters. In an embodiment, each shallow phase that is paired with a legacy phase processes a reduced precision compared with the legacy processing phase. In an embodiment, an amount of precision reduction may be configured by the routing configurations for one or more shallow phases either as a group or individually. In an embodiment, the reduction comprises a reduced number of channels or reduced precision for calculations.
At step 370, a processing path through the augmented neural network model is dynamically modified, according to constraints, to include one or more of the at least one shallow processing phases to process the inputs and produce outputs. In an embodiment, the augmented neural network is trained by adjusting the shallow parameters without changing the legacy parameters to match a distribution of the outputs compared with a legacy distribution of legacy outputs produced by processing the inputs using only the legacy processing phases. In an embodiment, the processing path is dynamically modified to maximize accuracy of the outputs for each variation of the constraints.
In an embodiment, the constraints comprise a metric and a value of the metric. In an embodiment, the metric is inference latency or energy consumption. In an embodiment, the metric is at least one of batch size, accuracy, and floating-point operations per second. In an embodiment, batch sizes of the inputs are varied to produce the outputs.
In an embodiment, the processing path is selected from a set of processing paths through the augmented neural network model that satisfies a performance criterion. In an embodiment, the set of processing paths define a Pareto set for at least one metric (e.g., batch size, FLOPs, etc.). In an embodiment, the Pareto set is associated with a particular batch size or FLOPs. In an embodiment, only the shallow parameters that are included in the set of processing paths are adjusted to match the distribution of the outputs processed through the set of processing paths and the legacy distribution. In an embodiment, each combination of the constraints is associated with one of the processing paths in the set of processing paths.
In an embodiment, batch sizes of additional inputs processed by the set of processing paths are dynamically modified to produce additional outputs. In an embodiment, the shallow parameters are adjusted without changing the legacy parameters to match an additional distribution between the additional outputs and an additional legacy distribution for additional legacy outputs produced by processing the additional inputs only using the legacy processing phases. In an embodiment, the processing path is defined by a routing configuration that, for each legacy processing phase, enables either the legacy processing phase or the shallow phase that is paired with the legacy processing phase. In an embodiment, the processing path is defined by a routing configuration that enables either the legacy processing phase or a portion of the legacy processing phase and the shallow phase that is paired with the legacy processing phase.
The augmented legacy neural network system 100 improves the overall neural network performance on a training set or balances a combination of generating the legacy output and improving the model performance while also preserving the legacy neural network model. In contrast, conventional techniques for reducing latency and energy consumption of neural networks include modifying the neural network (e.g., pruning) and do not preserve the original neural network. For dynamic neural networks, different versions of the neural network are constructed and trained for specific tasks and the original neural network is not preserved. Switching between the different neural networks is typically not possible in real-time due to the overhead needed to initialize each neural network with the corresponding weights. In contrast, the augmented neural network 125 does not require training for the specific tasks and can be dynamically reconfigured in real-time using the switches 120. Conventional slimmable networks have configurable activation channel widths and also fail to preserve the legacy neural network. Specifically, training different slimmable widths changes the weights for the full neural network configuration.
DNNs with variable architectures allow many possible configurations, but few of the configurations belong to the Pareto set in the accuracy/compute cost plane. The augmented legacy neural network system 100 benefits from improved accuracy resulting from being fine-tuned only for routing configurations in the Pareto set. The selection and the execution of the compute path in flexible DNNs is disentangled to increase versatility of the augmented neural network 125 compared with conventional solutions. This turns out to be a critical feature for the deployment of many real-time systems, where the currently available system resources determine the operating point to use. For example, a mostly idle system could use a highly accurate but costly DNN configuration, and switch to a lower performing, less costly one in case of high load scenario.
Analysis shows the importance of the batch size on the performance of DNNs with variable architectures. The fact that beyond a target device for execution of the DNN a target batch size should also be considered adds an additional axis to the problem of creating a flexible DNN. The LeAF design philosophy offers a possible solution: after the first training step, a second training step may be performed to extract and fine-tune a set of configurations, each optimized for a different batch size, and the routing configuration may still be changed on-the-fly to optimize performance for varying batch sizes.
A unique feature of the augmented legacy neural network system 100 is the integration and preservation of a legacy pre-trained DNN. This allows deploying flexible DNNs in the field while also ensuring that resources put into developing the legacy neural network are leveraged. Overall, there is a tension between accuracy, compute cost, flexibility and legacy preservation. After the first training step, the augmented legacy neural network system 100 offers the maximum flexibility (high number of executable configurations). By leveraging the available information on the target device (including the target batch size), the number of executable configurations may be significantly reduced (to the size of the Pareto set) and the average accuracy of the augmented legacy neural network system 100 may be increased. Flexibility is reduced by the second training step because the set of available, optimized routing configurations is now targeted towards the constraints used in the Pareto set (e.g., minimum latency). Compared with conventional dynamic neural network techniques, the augmented legacy neural network system 100 offers more flexibility and preserves the legacy DNN model in the executable set.
In an embodiment, the PPU 400 is a multi-threaded processor that is implemented on one or more integrated circuit devices. The PPU 400 is a latency hiding architecture designed to process many threads in parallel. A thread (e.g., a thread of execution) is an instantiation of a set of instructions configured to be executed by the PPU 400. In an embodiment, the PPU 400 is a graphics processing unit (GPU) configured to implement a graphics rendering pipeline for processing three-dimensional (3D) graphics data in order to generate two-dimensional (2D) image data for display on a display device. In other embodiments, the PPU 400 may be utilized for performing general-purpose computations. While one exemplary parallel processor is provided herein for illustrative purposes, it should be strongly noted that such processor is set forth for illustrative purposes only, and that any processor may be employed to supplement and/or substitute for the same.
One or more PPUs 400 may be configured to accelerate thousands of High Performance Computing (HPC), data center, cloud computing, and machine learning applications. The PPU 400 may be configured to accelerate numerous deep learning systems and applications for autonomous vehicles, simulation, computational graphics such as ray or path tracing, deep learning, high-accuracy speech, image, and text recognition systems, intelligent video analytics, molecular simulations, drug discovery, disease diagnosis, weather forecasting, big data analytics, astronomy, molecular dynamics simulation, financial modeling, robotics, factory automation, real-time language translation, online search optimizations, and personalized user recommendations, and the like.
As shown in
The NVLink 410 interconnect enables systems to scale and include one or more PPUs 400 combined with one or more CPUs, supports cache coherence between the PPUs 400 and CPUs, and CPU mastering. Data and/or commands may be transmitted by the NVLink 410 through the hub 430 to/from other units of the PPU 400 such as one or more copy engines, a video encoder, a video decoder, a power management unit, etc. (not explicitly shown). The NVLink 410 is described in more detail in conjunction with
The I/O unit 405 is configured to transmit and receive communications (e.g., commands, data, etc.) from a host processor (not shown) over the interconnect 402. The I/O unit 405 may communicate with the host processor directly via the interconnect 402 or through one or more intermediate devices such as a memory bridge. In an embodiment, the I/O unit 405 may communicate with one or more other processors, such as one or more the PPUs 400 via the interconnect 402. In an embodiment, the I/O unit 405 implements a Peripheral Component Interconnect Express (PCIe) interface for communications over a PCIe bus and the interconnect 402 is a PCIe bus. In alternative embodiments, the I/O unit 405 may implement other types of well-known interfaces for communicating with external devices.
The I/O unit 405 decodes packets received via the interconnect 402. In an embodiment, the packets represent commands configured to cause the PPU 400 to perform various operations. The I/O unit 405 transmits the decoded commands to various other units of the PPU 400 as the commands may specify. For example, some commands may be transmitted to the front end unit 415. Other commands may be transmitted to the hub 430 or other units of the PPU 400 such as one or more copy engines, a video encoder, a video decoder, a power management unit, etc. (not explicitly shown). In other words, the I/O unit 405 is configured to route communications between and among the various logical units of the PPU 400.
In an embodiment, a program executed by the host processor encodes a command stream in a buffer that provides workloads to the PPU 400 for processing. A workload may comprise several instructions and data to be processed by those instructions. The buffer is a region in a memory that is accessible (e.g., read/write) by both the host processor and the PPU 400. For example, the I/O unit 405 may be configured to access the buffer in a system memory connected to the interconnect 402 via memory requests transmitted over the interconnect 402. In an embodiment, the host processor writes the command stream to the buffer and then transmits a pointer to the start of the command stream to the PPU 400. The front end unit 415 receives pointers to one or more command streams. The front end unit 415 manages the one or more streams, reading commands from the streams and forwarding commands to the various units of the PPU 400.
The front end unit 415 is coupled to a scheduler unit 420 that configures the various GPCs 450 to process tasks defined by the one or more streams. The scheduler unit 420 is configured to track state information related to the various tasks managed by the scheduler unit 420. The state may indicate which GPC 450 a task is assigned to, whether the task is active or inactive, a priority level associated with the task, and so forth. The scheduler unit 420 manages the execution of a plurality of tasks on the one or more GPCs 450.
The scheduler unit 420 is coupled to a work distribution unit 425 that is configured to dispatch tasks for execution on the GPCs 450. The work distribution unit 425 may track a number of scheduled tasks received from the scheduler unit 420. In an embodiment, the work distribution unit 425 manages a pending task pool and an active task pool for each of the GPCs 450. As a GPC 450 finishes the execution of a task, that task is evicted from the active task pool for the GPC 450 and one of the other tasks from the pending task pool is selected and scheduled for execution on the GPC 450. If an active task has been idle on the GPC 450, such as while waiting for a data dependency to be resolved, then the active task may be evicted from the GPC 450 and returned to the pending task pool while another task in the pending task pool is selected and scheduled for execution on the GPC 450.
In an embodiment, a host processor executes a driver kernel that implements an application programming interface (API) that enables one or more applications executing on the host processor to schedule operations for execution on the PPU 400. In an embodiment, multiple compute applications are simultaneously executed by the PPU 400 and the PPU 400 provides isolation, quality of service (QoS), and independent address spaces for the multiple compute applications. An application may generate instructions (e.g., API calls) that cause the driver kernel to generate one or more tasks for execution by the PPU 400. The driver kernel outputs tasks to one or more streams being processed by the PPU 400. Each task may comprise one or more groups of related threads, referred to herein as a warp. In an embodiment, a warp comprises 32 related threads that may be executed in parallel. Cooperating threads may refer to a plurality of threads including instructions to perform the task and that may exchange data through shared memory. The tasks may be allocated to one or more processing units within a GPC 450 and instructions are scheduled for execution by at least one warp.
The work distribution unit 425 communicates with the one or more GPCs 450 via XBar 470. The XBar 470 is an interconnect network that couples many of the units of the PPU 400 to other units of the PPU 400. For example, the XBar 470 may be configured to couple the work distribution unit 425 to a particular GPC 450. Although not shown explicitly, one or more other units of the PPU 400 may also be connected to the XBar 470 via the hub 430.
The tasks are managed by the scheduler unit 420 and dispatched to a GPC 450 by the work distribution unit 425. The GPC 450 is configured to process the task and generate results. The results may be consumed by other tasks within the GPC 450, routed to a different GPC 450 via the XBar 470, or stored in the memory 404. The results can be written to the memory 404 via the memory partition units 480, which implement a memory interface for reading and writing data to/from the memory 404. The results can be transmitted to another PPU 400 or CPU via the NVLink 410. In an embodiment, the PPU 400 includes a number U of memory partition units 480 that is equal to the number of separate and distinct memory devices of the memory 404 coupled to the PPU 400. Each GPC 450 may include a memory management unit to provide translation of virtual addresses into physical addresses, memory protection, and arbitration of memory requests. In an embodiment, the memory management unit provides one or more translation lookaside buffers (TLBs) for performing translation of virtual addresses into physical addresses in the memory 404.
In an embodiment, the memory partition unit 480 includes a Raster Operations (ROP) unit, a level two (L2) cache, and a memory interface that is coupled to the memory 404. The memory interface may implement 32, 64, 128, 1024-bit data buses, or the like, for high-speed data transfer. The PPU 400 may be connected to up to Y memory devices, such as high bandwidth memory stacks or graphics double-data-rate, version 5, synchronous dynamic random access memory, or other types of persistent storage. In an embodiment, the memory interface implements an HBM2 memory interface and Y equals half U. In an embodiment, the HBM2 memory stacks are located on the same physical package as the PPU 400, providing substantial power and area savings compared with conventional GDDR5 SDRAM systems. In an embodiment, each HBM2 stack includes four memory dies and Y equals 4, with each HBM2 stack including two 128-bit channels per die for a total of 8 channels and a data bus width of 1024 bits.
In an embodiment, the memory 404 supports Single-Error Correcting Double-Error Detecting (SECDED) Error Correction Code (ECC) to protect data. ECC provides higher reliability for compute applications that are sensitive to data corruption. Reliability is especially important in large-scale cluster computing environments where PPUs 400 process very large datasets and/or run applications for extended periods.
In an embodiment, the PPU 400 implements a multi-level memory hierarchy. In an embodiment, the memory partition unit 480 supports a unified memory to provide a single unified virtual address space for CPU and PPU 400 memory, enabling data sharing between virtual memory systems. In an embodiment the frequency of accesses by a PPU 400 to memory located on other processors is traced to ensure that memory pages are moved to the physical memory of the PPU 400 that is accessing the pages more frequently. In an embodiment, the NVLink 410 supports address translation services allowing the PPU 400 to directly access a CPU's page tables and providing full access to CPU memory by the PPU 400.
In an embodiment, copy engines transfer data between multiple PPUs 400 or between PPUs 400 and CPUs. The copy engines can generate page faults for addresses that are not mapped into the page tables. The memory partition unit 480 can then service the page faults, mapping the addresses into the page table, after which the copy engine can perform the transfer. In a conventional system, memory is pinned (e.g., non-pageable) for multiple copy engine operations between multiple processors, substantially reducing the available memory. With hardware page faulting, addresses can be passed to the copy engines without worrying if the memory pages are resident, and the copy process is transparent.
Data from the memory 404 or other system memory may be fetched by the memory partition unit 480 and stored in the L2 cache 460, which is located on-chip and is shared between the various GPCs 450. As shown, each memory partition unit 480 includes a portion of the L2 cache associated with a corresponding memory 404. Lower level caches may then be implemented in various units within the GPCs 450. For example, each of the processing units within a GPC 450 may implement a level one (L1) cache. The L1 cache is private memory that is dedicated to a particular processing unit. The L2 cache 460 is coupled to the memory interface 470 and the XBar 470 and data from the L2 cache may be fetched and stored in each of the L1 caches for processing.
In an embodiment, the processing units within each GPC 450 implement a SIMD (Single-Instruction, Multiple-Data) architecture where each thread in a group of threads (e.g., a warp) is configured to process a different set of data based on the same set of instructions. All threads in the group of threads execute the same instructions. In another embodiment, the processing unit implements a SIMT (Single-Instruction, Multiple Thread) architecture where each thread in a group of threads is configured to process a different set of data based on the same set of instructions, but where individual threads in the group of threads are allowed to diverge during execution. In an embodiment, a program counter, call stack, and execution state is maintained for each warp, enabling concurrency between warps and serial execution within warps when threads within the warp diverge. In another embodiment, a program counter, call stack, and execution state is maintained for each individual thread, enabling equal concurrency between all threads, within and between warps. When execution state is maintained for each individual thread, threads executing the same instructions may be converged and executed in parallel for maximum efficiency.
Cooperative Groups is a programming model for organizing groups of communicating threads that allows developers to express the granularity at which threads are communicating, enabling the expression of richer, more efficient parallel decompositions. Cooperative launch APIs support synchronization amongst thread blocks for the execution of parallel algorithms. Conventional programming models provide a single, simple construct for synchronizing cooperating threads: a barrier across all threads of a thread block (e.g., the syncthreads( ) function). However, programmers would often like to define groups of threads at smaller than thread block granularities and synchronize within the defined groups to enable greater performance, design flexibility, and software reuse in the form of collective group-wide function interfaces.
Cooperative Groups enables programmers to define groups of threads explicitly at sub-block (e.g., as small as a single thread) and multi-block granularities, and to perform collective operations such as synchronization on the threads in a cooperative group. The programming model supports clean composition across software boundaries, so that libraries and utility functions can synchronize safely within their local context without having to make assumptions about convergence. Cooperative Groups primitives enable new patterns of cooperative parallelism, including producer-consumer parallelism, opportunistic parallelism, and global synchronization across an entire grid of thread blocks.
Each processing unit includes a large number (e.g., 128, etc.) of distinct processing cores (e.g., functional units) that may be fully-pipelined, single-precision, double-precision, and/or mixed precision and include a floating point arithmetic logic unit and an integer arithmetic logic unit. In an embodiment, the floating point arithmetic logic units implement the IEEE 754-2008 standard for floating point arithmetic. In an embodiment, the cores include 64 single-precision (32-bit) floating point cores, 64 integer cores, 32 double-precision (64-bit) floating point cores, and 8 tensor cores.
Tensor cores configured to perform matrix operations. In particular, the tensor cores are configured to perform deep learning matrix arithmetic, such as GEMM (matrix-matrix multiplication) for convolution operations during neural network training and inferencing. In an embodiment, each tensor core operates on a 4×4 matrix and performs a matrix multiply and accumulate operation D=A′B+C, where A, B, C, and D are 4×4 matrices.
In an embodiment, the matrix multiply inputs A and B may be integer, fixed-point, or floating point matrices, while the accumulation matrices C and D may be integer, fixed-point, or floating point matrices of equal or higher bitwidths. In an embodiment, tensor cores operate on one, four, or eight bit integer input data with 32-bit integer accumulation. The 8-bit integer matrix multiply requires 1024 operations and results in a full precision product that is then accumulated using 32-bit integer addition with the other intermediate products for a 8×8×16 matrix multiply. In an embodiment, tensor Cores operate on 16-bit floating point input data with 32-bit floating point accumulation. The 16-bit floating point multiply requires 64 operations and results in a full precision product that is then accumulated using 32-bit floating point addition with the other intermediate products for a 4×4×4 matrix multiply. In practice, Tensor Cores are used to perform much larger two-dimensional or higher dimensional matrix operations, built up from these smaller elements. An API, such as CUDA 9 C++ API, exposes specialized matrix load, matrix multiply and accumulate, and matrix store operations to efficiently use Tensor Cores from a CUDA-C++ program. At the CUDA level, the warp-level interface assumes 16×16 size matrices spanning all 32 threads of the warp.
Each processing unit may also comprise M special function units (SFUs) that perform special functions (e.g., attribute evaluation, reciprocal square root, and the like). In an embodiment, the SFUs may include a tree traversal unit configured to traverse a hierarchical tree data structure. In an embodiment, the SFUs may include texture unit configured to perform texture map filtering operations. In an embodiment, the texture units are configured to load texture maps (e.g., a 2D array of texels) from the memory 404 and sample the texture maps to produce sampled texture values for use in shader programs executed by the processing unit. In an embodiment, the texture maps are stored in shared memory that may comprise or include an L1 cache. The texture units implement texture operations such as filtering operations using mip-maps (e.g., texture maps of varying levels of detail). In an embodiment, each processing unit includes two texture units.
Each processing unit also comprises N load store units (LSUs) that implement load and store operations between the shared memory and the register file. Each processing unit includes an interconnect network that connects each of the cores to the register file and the LSU to the register file, shared memory. In an embodiment, the interconnect network is a crossbar that can be configured to connect any of the cores to any of the registers in the register file and connect the LSUs to the register file and memory locations in shared memory.
The shared memory is an array of on-chip memory that allows for data storage and communication between the processing units and between threads within a processing unit. In an embodiment, the shared memory comprises 128 KB of storage capacity and is in the path from each of the processing units to the memory partition unit 480. The shared memory can be used to cache reads and writes. One or more of the shared memory, L1 cache, L2 cache, and memory 404 are backing stores.
Combining data cache and shared memory functionality into a single memory block provides the best overall performance for both types of memory accesses. The capacity is usable as a cache by programs that do not use shared memory. For example, if shared memory is configured to use half of the capacity, texture and load/store operations can use the remaining capacity. Integration within the shared memory enables the shared memory to function as a high-throughput conduit for streaming data while simultaneously providing high-bandwidth and low-latency access to frequently reused data.
When configured for general purpose parallel computation, a simpler configuration can be used compared with graphics processing. Specifically, fixed function graphics processing units, are bypassed, creating a much simpler programming model. In the general purpose parallel computation configuration, the work distribution unit 425 assigns and distributes blocks of threads directly to the processing units within the GPCs 450. Threads execute the same program, using a unique thread ID in the calculation to ensure each thread generates unique results, using the processing unit(s) to execute the program and perform calculations, shared memory to communicate between threads, and the LSU to read and write global memory through the shared memory and the memory partition unit 480. When configured for general purpose parallel computation, the processing units can also write commands that the scheduler unit 420 can use to launch new work on the processing units.
The PPUs 400 may each include, and/or be configured to perform functions of, one or more processing cores and/or components thereof, such as Tensor Cores (TCs), Tensor Processing Units (TPUs), Pixel Visual Cores (PVCs), Ray Tracing (RT) Cores, Vision Processing Units (VPUs), Graphics Processing Clusters (GPCs), Texture Processing Clusters (TPCs), Streaming Multiprocessors (SMs), Tree Traversal Units (TTUs), Artificial Intelligence Accelerators (AIAs), Deep Learning Accelerators (DLAs), Arithmetic-Logic Units (ALUs), Application-Specific Integrated Circuits (ASICs), Floating Point Units (FPUs), input/output (I/O) elements, peripheral component interconnect (PCI) or peripheral component interconnect express (PCIe) elements, and/or the like.
The PPU 400 may be included in a desktop computer, a laptop computer, a tablet computer, servers, supercomputers, a smart-phone (e.g., a wireless, hand-held device), personal digital assistant (PDA), a digital camera, a vehicle, a head mounted display, a hand-held electronic device, and the like. In an embodiment, the PPU 400 is embodied on a single semiconductor substrate. In another embodiment, the PPU 400 is included in a system-on-a-chip (SoC) along with one or more other devices such as additional PPUs 400, the memory 404, a reduced instruction set computer (RISC) CPU, a memory management unit (MMU), a digital-to-analog converter (DAC), and the like.
In an embodiment, the PPU 400 may be included on a graphics card that includes one or more memory devices. The graphics card may be configured to interface with a PCIe slot on a motherboard of a desktop computer. In yet another embodiment, the PPU 400 may be an integrated graphics processing unit (iGPU) or parallel processor included in the chipset of the motherboard. In yet another embodiment, the PPU 400 may be realized in reconfigurable hardware. In yet another embodiment, parts of the PPU 400 may be realized in reconfigurable hardware.
Systems with multiple GPUs and CPUs are used in a variety of industries as developers expose and leverage more parallelism in applications such as artificial intelligence computing. High-performance GPU-accelerated systems with tens to many thousands of compute nodes are deployed in data centers, research facilities, and supercomputers to solve ever larger problems. As the number of processing devices within the high-performance systems increases, the communication and data transfer mechanisms need to scale to support the increased bandwidth.
The NVLink 410 provides high-speed communication links between each of the PPUs 400. Although a particular number of NVLink 410 and interconnect 402 connections are illustrated in
In another embodiment (not shown), the NVLink 410 provides one or more high-speed communication links between each of the PPUs 400 and the CPU 530 and the switch 510 interfaces between the interconnect 402 and each of the PPUs 400. The PPUs 400, memories 404, and interconnect 402 may be situated on a single semiconductor platform to form a parallel processing module 525. In yet another embodiment (not shown), the interconnect 402 provides one or more communication links between each of the PPUs 400 and the CPU 530 and the switch 510 interfaces between each of the PPUs 400 using the NVLink 410 to provide one or more high-speed communication links between the PPUs 400. In another embodiment (not shown), the NVLink 410 provides one or more high-speed communication links between the PPUs 400 and the CPU 530 through the switch 510. In yet another embodiment (not shown), the interconnect 402 provides one or more communication links between each of the PPUs 400 directly. One or more of the NVLink 410 high-speed communication links may be implemented as a physical NVLink interconnect or either an on-chip or on-die interconnect using the same protocol as the NVLink 410.
In the context of the present description, a single semiconductor platform may refer to a sole unitary semiconductor-based integrated circuit fabricated on a die or chip. It should be noted that the term single semiconductor platform may also refer to multi-chip modules with increased connectivity which simulate on-chip operation and make substantial improvements over utilizing a conventional bus implementation. Of course, the various circuits or devices may also be situated separately or in various combinations of semiconductor platforms per the desires of the user. Alternately, the parallel processing module 525 may be implemented as a circuit board substrate and each of the PPUs 400 and/or memories 404 may be packaged devices. In an embodiment, the CPU 530, switch 510, and the parallel processing module 525 are situated on a single semiconductor platform.
In an embodiment, the signaling rate of each NVLink 410 is 20 to 25 Gigabits/second and each PPU 400 includes six NVLink 410 interfaces (as shown in
In an embodiment, the NVLink 410 allows direct load/store/atomic access from the CPU 530 to each PPU's 400 memory 404. In an embodiment, the NVLink 410 supports coherency operations, allowing data read from the memories 404 to be stored in the cache hierarchy of the CPU 530, reducing cache access latency for the CPU 530. In an embodiment, the NVLink 410 includes support for Address Translation Services (ATS), allowing the PPU 400 to directly access page tables within the CPU 530. One or more of the NVLinks 410 may also be configured to operate in a low-power mode.
Although the various blocks of
The system 565 also includes a main memory 540. Control logic (software) and data are stored in the main memory 540 which may take the form of a variety of computer-readable media. The computer-readable media may be any available media that may be accessed by the system 565. The computer-readable media may include both volatile and nonvolatile media, and removable and non-removable media. By way of example, and not limitation, the computer-readable media may comprise computer-storage media and communication media.
The computer-storage media may include both volatile and nonvolatile media and/or removable and non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules, and/or other data types. For example, the main memory 540 may store computer-readable instructions (e.g., that represent a program(s) and/or a program element(s), such as an operating system. Computer-storage media may include, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which may be used to store the desired information and which may be accessed by system 565. As used herein, computer storage media does not comprise signals per se.
The computer storage media may embody computer-readable instructions, data structures, program modules, and/or other data types in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media. The term “modulated data signal” may refer to a signal that has one or more of its characteristics set or changed in such a manner as to encode information in the signal. By way of example, and not limitation, the computer storage media may include wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared and other wireless media. Combinations of any of the above should also be included within the scope of computer-readable media.
Computer programs, when executed, enable the system 565 to perform various functions. The CPU(s) 530 may be configured to execute at least some of the computer-readable instructions to control one or more components of the system 565 to perform one or more of the methods and/or processes described herein, such as the method 130 shown in
In addition to or alternatively from the CPU(s) 530, the parallel processing module 525 may be configured to execute at least some of the computer-readable instructions to control one or more components of the system 565 to perform one or more of the methods and/or processes described herein, such as the method 130 shown in
The system 565 also includes input device(s) 560, the parallel processing system 525, and display device(s) 545. The display device(s) 545 may include a display (e.g., a monitor, a touch screen, a television screen, a heads-up-display (HUD), other display types, or a combination thereof), speakers, and/or other presentation components. The display device(s) 545 may receive data from other components (e.g., the parallel processing system 525, the CPU(s) 530, etc.), and output the data (e.g., as an image, video, sound, etc.).
The network interface 535 may enable the system 565 to be logically coupled to other devices including the input devices 560, the display device(s) 545, and/or other components, some of which may be built in to (e.g., integrated in) the system 565. Illustrative input devices 560 include a microphone, mouse, keyboard, joystick, game pad, game controller, satellite dish, scanner, printer, wireless device, etc. The input devices 560 may provide a natural user interface (NUI) that processes air gestures, voice, or other physiological inputs generated by a user. In some instances, inputs may be transmitted to an appropriate network element for further processing. An NUI may implement any combination of speech recognition, stylus recognition, facial recognition, biometric recognition, gesture recognition both on screen and adjacent to the screen, air gestures, head and eye tracking, and touch recognition (as described in more detail below) associated with a display of the system 565. The system 565 may be include depth cameras, such as stereoscopic camera systems, infrared camera systems, RGB camera systems, touchscreen technology, and combinations of these, for gesture detection and recognition. Additionally, the system 565 may include accelerometers or gyroscopes (e.g., as part of an inertia measurement unit (IMU)) that enable detection of motion. In some examples, the output of the accelerometers or gyroscopes may be used by the system 565 to render immersive augmented reality or virtual reality.
Further, the system 565 may be coupled to a network (e.g., a telecommunications network, local area network (LAN), wireless network, wide area network (WAN) such as the Internet, peer-to-peer network, cable network, or the like) through a network interface 535 for communication purposes. The system 565 may be included within a distributed network and/or cloud computing environment.
The network interface 535 may include one or more receivers, transmitters, and/or transceivers that enable the system 565 to communicate with other computing devices via an electronic communication network, included wired and/or wireless communications. The network interface 535 may be implemented as a network interface controller (NIC) that includes one or more data processing units (DPUs) to perform operations such as (for example and without limitation) packet parsing and accelerating network processing and communication. The network interface 535 may include components and functionality to enable communication over any of a number of different networks, such as wireless networks (e.g., Wi-Fi, Z-Wave, Bluetooth, Bluetooth LE, ZigBee, etc.), wired networks (e.g., communicating over Ethernet or InfiniBand), low-power wide-area networks (e.g., LoRaWAN, SigFox, etc.), and/or the Internet.
The system 565 may also include a secondary storage (not shown). The secondary storage includes, for example, a hard disk drive and/or a removable storage drive, representing a floppy disk drive, a magnetic tape drive, a compact disk drive, digital versatile disk (DVD) drive, recording device, universal serial bus (USB) flash memory. The removable storage drive reads from and/or writes to a removable storage unit in a well-known manner. The system 565 may also include a hard-wired power supply, a battery power supply, or a combination thereof (not shown). The power supply may provide power to the system 565 to enable the components of the system 565 to operate.
Each of the foregoing modules and/or devices may even be situated on a single semiconductor platform to form the system 565. Alternately, the various modules may also be situated separately or in various combinations of semiconductor platforms per the desires of the user. While various embodiments have been described above, it should be understood that they have been presented by way of example only, and not limitation. Thus, the breadth and scope of a preferred embodiment should not be limited by any of the above-described exemplary embodiments, but should be defined only in accordance with the following claims and their equivalents.
Network environments suitable for use in implementing embodiments of the disclosure may include one or more client devices, servers, network attached storage (NAS), other backend devices, and/or other device types. The client devices, servers, and/or other device types (e.g., each device) may be implemented on one or more instances of the processing system 500 of
Components of a network environment may communicate with each other via a network(s), which may be wired, wireless, or both. The network may include multiple networks, or a network of networks. By way of example, the network may include one or more Wide Area Networks (WANs), one or more Local Area Networks (LANs), one or more public networks such as the Internet and/or a public switched telephone network (PSTN), and/or one or more private networks. Where the network includes a wireless telecommunications network, components such as a base station, a communications tower, or even access points (as well as other components) may provide wireless connectivity.
Compatible network environments may include one or more peer-to-peer network environments—in which case a server may not be included in a network environment—and one or more client-server network environments—in which case one or more servers may be included in a network environment. In peer-to-peer network environments, functionality described herein with respect to a server(s) may be implemented on any number of client devices.
In at least one embodiment, a network environment may include one or more cloud-based network environments, a distributed computing environment, a combination thereof, etc. A cloud-based network environment may include a framework layer, a job scheduler, a resource manager, and a distributed file system implemented on one or more of servers, which may include one or more core network servers and/or edge servers. A framework layer may include a framework to support software of a software layer and/or one or more application(s) of an application layer. The software or application(s) may respectively include web-based service software or applications. In embodiments, one or more of the client devices may use the web-based service software or applications (e.g., by accessing the service software and/or applications via one or more application programming interfaces (APIs)). The framework layer may be, but is not limited to, a type of free and open-source software web application framework such as that may use a distributed file system for large-scale data processing (e.g., “big data”).
A cloud-based network environment may provide cloud computing and/or cloud storage that carries out any combination of computing and/or data storage functions described herein (or one or more portions thereof). Any of these various functions may be distributed over multiple locations from central or core servers (e.g., of one or more data centers that may be distributed across a state, a region, a country, the globe, etc.). If a connection to a user (e.g., a client device) is relatively close to an edge server(s), a core server(s) may designate at least a portion of the functionality to the edge server(s). A cloud-based network environment may be private (e.g., limited to a single organization), may be public (e.g., available to many organizations), and/or a combination thereof (e.g., a hybrid cloud environment).
The client device(s) may include at least some of the components, features, and functionality of the example processing system 500 of
Deep neural networks (DNNs) developed on processors, such as the PPU 400 have been used for diverse use cases, from self-driving cars to faster drug development, from automatic image captioning in online image databases to smart real-time language translation in video chat applications. Deep learning is a technique that models the neural learning process of the human brain, continually learning, continually getting smarter, and delivering more accurate results more quickly over time. A child is initially taught by an adult to correctly identify and classify various shapes, eventually being able to identify shapes without any coaching. Similarly, a deep learning or neural learning system needs to be trained in object recognition and classification for it get smarter and more efficient at identifying basic objects, occluded objects, etc., while also assigning context to objects.
At the simplest level, neurons in the human brain look at various inputs that are received, importance levels are assigned to each of these inputs, and output is passed on to other neurons to act upon. An artificial neuron or perceptron is the most basic model of a neural network. In one example, a perceptron may receive one or more inputs that represent various features of an object that the perceptron is being trained to recognize and classify, and each of these features is assigned a certain weight based on the importance of that feature in defining the shape of an object.
A deep neural network (DNN) model includes multiple layers of many connected nodes (e.g., perceptrons, Boltzmann machines, radial basis functions, convolutional layers, etc.) that can be trained with enormous amounts of input data to quickly solve complex problems with high accuracy. In one example, a first layer of the DNN model breaks down an input image of an automobile into various sections and looks for basic patterns such as lines and angles. The second layer assembles the lines to look for higher level patterns such as wheels, windshields, and mirrors. The next layer identifies the type of vehicle, and the final few layers generate a label for the input image, identifying the model of a specific automobile brand.
Once the DNN is trained, the DNN can be deployed and used to identify and classify objects or patterns in a process known as inference. Examples of inference (the process through which a DNN extracts useful information from a given input) include identifying handwritten numbers on checks deposited into ATM machines, identifying images of friends in photos, delivering movie recommendations to over fifty million users, identifying and classifying different types of automobiles, pedestrians, and road hazards in driverless cars, or translating human speech in real-time.
During training, data flows through the DNN in a forward propagation phase until a prediction is produced that indicates a label corresponding to the input. If the neural network does not correctly label the input, then errors between the correct label and the predicted label are analyzed, and the weights are adjusted for each feature during a backward propagation phase until the DNN correctly labels the input and other inputs in a training dataset. Training complex neural networks requires massive amounts of parallel computing performance, including floating-point multiplications and additions that are supported by the PPU 400. Inferencing is less compute-intensive than training, being a latency-sensitive process where a trained neural network is applied to new inputs it has not seen before to classify images, detect emotions, identify recommendations, recognize and translate speech, and generally infer new information.
Neural networks rely heavily on matrix math operations, and complex multi-layered networks require tremendous amounts of floating-point performance and bandwidth for both efficiency and speed. With thousands of processing cores, optimized for matrix math operations, and delivering tens to hundreds of TFLOPS of performance, the PPU 400 is a computing platform capable of delivering performance required for deep neural network-based artificial intelligence and machine learning applications.
Furthermore, images generated applying one or more of the techniques disclosed herein may be used to train, test, or certify DNNs used to recognize objects and environments in the real world. Such images may include scenes of roadways, factories, buildings, urban settings, rural settings, humans, animals, and any other physical object or real-world setting. Such images may be used to train, test, or certify DNNs that are employed in machines or robots to manipulate, handle, or modify physical objects in the real world. Furthermore, such images may be used to train, test, or certify DNNs that are employed in autonomous vehicles to navigate and move the vehicles through the real world. Additionally, images generated applying one or more of the techniques disclosed herein may be used to convey information to users of such machines, robots, and vehicles.
In at least one embodiment, requests are able to be submitted across at least one network 504 to be received by a provider environment 506. In at least one embodiment, a client device may be any appropriate electronic and/or computing devices enabling a user to generate and send such requests, such as, but not limited to, desktop computers, notebook computers, computer servers, smartphones, tablet computers, gaming consoles (portable or otherwise), computer processors, computing logic, and set-top boxes. Network(s) 504 can include any appropriate network for transmitting a request or other such data, as may include Internet, an intranet, an Ethernet, a cellular network, a local area network (LAN), a wide area network (WAN), a personal area network (PAN), an ad hoc network of direct wireless connections among peers, and so on.
In at least one embodiment, requests can be received at an interface layer 508, which can forward data to a training and inference manager 532, in this example. The training and inference manager 532 can be a system or service including hardware and software for managing requests and service corresponding data or content, in at least one embodiment, the training and inference manager 532 can receive a request to train a neural network, and can provide data for a request to a training module 512. In at least one embodiment, training module 512 can select an appropriate model or neural network to be used, if not specified by the request, and can train a model using relevant training data. In at least one embodiment, training data can be a batch of data stored in a training data repository 514, received from client device 502, or obtained from a third party provider 524. In at least one embodiment, training module 512 can be responsible for training data. A neural network can be any appropriate network, such as a recurrent neural network (RNN) or convolutional neural network (CNN). Once a neural network is trained and successfully evaluated, a trained neural network can be stored in a model repository 516, for example, that may store different models or networks for users, applications, or services, etc. In at least one embodiment, there may be multiple models for a single application or entity, as may be utilized based on a number of different factors.
In at least one embodiment, at a subsequent point in time, a request may be received from client device 502 (or another such device) for content (e.g., path determinations) or data that is at least partially determined or impacted by a trained neural network. This request can include, for example, input data to be processed using a neural network to obtain one or more inferences or other output values, classifications, or predictions, or for at least one embodiment, input data can be received by interface layer 508 and directed to inference module 518, although a different system or service can be used as well. In at least one embodiment, inference module 518 can obtain an appropriate trained network, such as a trained deep neural network (DNN) as discussed herein, from model repository 516 if not already stored locally to inference module 518. Inference module 518 can provide data as input to a trained network, which can then generate one or more inferences as output. This may include, for example, a classification of an instance of input data. In at least one embodiment, inferences can then be transmitted to client device 502 for display or other communication to a user. In at least one embodiment, context data for a user may also be stored to a user context data repository 522, which may include data about a user which may be useful as input to a network in generating inferences, or determining data to return to a user after obtaining instances. In at least one embodiment, relevant data, which may include at least some of input or inference data, may also be stored to a local database 534 for processing future requests. In at least one embodiment, a user can use account information or other information to access resources or functionality of a provider environment. In at least one embodiment, if permitted and available, user data may also be collected and used to further train models, in order to provide more accurate inferences for future requests. In at least one embodiment, requests may be received through a user interface to a machine learning application 526 executing on client device 502, and results displayed through a same interface. A client device can include resources such as a processor 528 and memory 562 for generating a request and processing results or a response, as well as at least one data storage element 552 for storing data for machine learning application 526.
In at least one embodiment a processor 528 (or a processor of training module 512 or inference module 518) will be a central processing unit (CPU). As mentioned, however, resources in such environments can utilize GPUs to process data for at least certain types of requests. With thousands of cores, GPUs, such as PPU 300 are designed to handle substantial parallel workloads and, therefore, have become popular in deep learning for training neural networks and generating predictions. While use of GPUs for offline builds has enabled faster training of larger and more complex models, generating predictions offline implies that either request-time input features cannot be used or predictions must be generated for all permutations of features and stored in a lookup table to serve real-time requests. If a deep learning framework supports a CPU-mode and a model is small and simple enough to perform a feed-forward on a CPU with a reasonable latency, then a service on a CPU instance could host a model. In this case, training can be done offline on a GPU and inference done in real-time on a CPU. If a CPU approach is not viable, then a service can run on a GPU instance. Because GPUs have different performance and cost characteristics than CPUs, however, running a service that offloads a runtime algorithm to a GPU can require it to be designed differently from a CPU based service.
In at least one embodiment, video data can be provided from client device 502 for enhancement in provider environment 506. In at least one embodiment, video data can be processed for enhancement on client device 502. In at least one embodiment, video data may be streamed from a third party content provider 524 and enhanced by third party content provider 524, provider environment 506, or client device 502. In at least one embodiment, video data can be provided from client device 502 for use as training data in provider environment 506.
In at least one embodiment, supervised and/or unsupervised training can be performed by the client device 502 and/or the provider environment 506. In at least one embodiment, a set of training data 514 (e.g., classified or labeled data) is provided as input to function as training data. In at least one embodiment, training data can include instances of at least one type of object for which a neural network is to be trained, as well as information that identifies that type of object. In at least one embodiment, training data might include a set of images that each includes a representation of a type of object, where each image also includes, or is associated with, a label, metadata, classification, or other piece of information identifying a type of object represented in a respective image. Various other types of data may be used as training data as well, as may include text data, audio data, video data, and so on. In at least one embodiment, training data 514 is provided as training input to a training module 512. In at least one embodiment, training module 512 can be a system or service that includes hardware and software, such as one or more computing devices executing a training application, for training a neural network (or other model or algorithm, etc.). In at least one embodiment, training module 512 receives an instruction or request indicating a type of model to be used for training, in at least one embodiment, a model can be any appropriate statistical model, network, or algorithm useful for such purposes, as may include an artificial neural network, deep learning algorithm, learning classifier, Bayesian network, and so on. In at least one embodiment, training module 512 can select an initial model, or other untrained model, from an appropriate repository 516 and utilize training data 514 to train a model, thereby generating a trained model (e.g., trained deep neural network) that can be used to classify similar types of data, or generate other such inferences. In at least one embodiment where training data is not used, an appropriate initial model can still be selected for training on input data per training module 512.
In at least one embodiment, a model can be trained in a number of different ways, as may depend in part upon a type of model selected. In at least one embodiment, a machine learning algorithm can be provided with a set of training data, where a model is a model artifact created by a training process. In at least one embodiment, each instance of training data contains a correct answer (e.g., classification), which can be referred to as a target or target attribute. In at least one embodiment, a learning algorithm finds patterns in training data that map input data attributes to a target, an answer to be predicted, and a machine learning model is output that captures these patterns. In at least one embodiment, a machine learning model can then be used to obtain predictions on new data for which a target is not specified.
In at least one embodiment, training and inference manager 532 can select from a set of machine learning models including binary classification, multiclass classification, generative, and regression models. In at least one embodiment, a type of model to be used can depend at least in part upon a type of target to be predicted.
Images generated applying one or more of the techniques disclosed herein may be displayed on a monitor or other display device. In some embodiments, the display device may be coupled directly to the system or processor generating or rendering the images. In other embodiments, the display device may be coupled indirectly to the system or processor such as via a network. Examples of such networks include the Internet, mobile telecommunications networks, a WIFI network, as well as any other wired and/or wireless networking system. When the display device is indirectly coupled, the images generated by the system or processor may be streamed over the network to the display device. Such streaming allows, for example, video games or other applications, which render images, to be executed on a server, a data center, or in a cloud-based computing environment and the rendered images to be transmitted and displayed on one or more user devices (such as a computer, video game console, smartphone, other mobile device, etc.) that are physically separate from the server or data center. Hence, the techniques disclosed herein can be applied to enhance the images that are streamed and to enhance services that stream images such as NVIDIA GeForce Now (GFN), Google Stadia, and the like.
In an embodiment, the streaming system 605 is a game streaming system and the sever(s) 604 are game server(s). In the system 605, for a game session, the client device(s) 604 may only receive input data in response to inputs to the input device(s) 626, transmit the input data to the server(s) 603, receive encoded display data from the server(s) 603, and display the display data on the display 624. As such, the more computationally intense computing and processing is offloaded to the server(s) 603 (e.g., rendering—in particular ray or path tracing—for graphical output of the game session is executed by the GPU(s) 615 of the server(s) 603). In other words, the game session is streamed to the client device(s) 604 from the server(s) 603, thereby reducing the requirements of the client device(s) 604 for graphics processing and rendering.
For example, with respect to an instantiation of a game session, a client device 604 may be displaying a frame of the game session on the display 624 based on receiving the display data from the server(s) 603. The client device 604 may receive an input to one of the input device(s) 626 and generate input data in response. The client device 604 may transmit the input data to the server(s) 603 via the communication interface 621 and over the network(s) 606 (e.g., the Internet), and the server(s) 603 may receive the input data via the communication interface 618. The CPU(s) 608 may receive the input data, process the input data, and transmit data to the GPU(s) 615 that causes the GPU(s) 615 to generate a rendering of the game session. For example, the input data may be representative of a movement of a character of the user in a game, firing a weapon, reloading, passing a ball, turning a vehicle, etc. The rendering component 612 may render the game session (e.g., representative of the result of the input data) and the render capture component 614 may capture the rendering of the game session as display data (e.g., as image data capturing the rendered frame of the game session). The rendering of the game session may include ray or path-traced lighting and/or shadow effects, computed using one or more parallel processing units—such as GPUs, which may further employ the use of one or more dedicated hardware accelerators or processing cores to perform ray or path-tracing techniques—of the server(s) 603. The encoder 616 may then encode the display data to generate encoded display data and the encoded display data may be transmitted to the client device 604 over the network(s) 606 via the communication interface 618. The client device 604 may receive the encoded display data via the communication interface 621 and the decoder 622 may decode the encoded display data to generate the display data. The client device 604 may then display the display data via the display 624.
It is noted that the techniques described herein may be embodied in executable instructions stored in a computer readable medium for use by or in connection with a processor-based instruction execution machine, system, apparatus, or device. It will be appreciated by those skilled in the art that, for some embodiments, various types of computer-readable media can be included for storing data. As used herein, a “computer-readable medium” includes one or more of any suitable media for storing the executable instructions of a computer program such that the instruction execution machine, system, apparatus, or device may read (or fetch) the instructions from the computer-readable medium and execute the instructions for carrying out the described embodiments. Suitable storage formats include one or more of an electronic, magnetic, optical, and electromagnetic format. A non-exhaustive list of conventional exemplary computer-readable medium includes: a portable computer diskette; a random-access memory (RAM); a read-only memory (ROM); an erasable programmable read only memory (EPROM); a flash memory device; and optical storage devices, including a portable compact disc (CD), a portable digital video disc (DVD), and the like.
It should be understood that the arrangement of components illustrated in the attached Figures are for illustrative purposes and that other arrangements are possible. For example, one or more of the elements described herein may be realized, in whole or in part, as an electronic hardware component. Other elements may be implemented in software, hardware, or a combination of software and hardware. Moreover, some or all of these other elements may be combined, some may be omitted altogether, and additional components may be added while still achieving the functionality described herein. Thus, the subject matter described herein may be embodied in many different variations, and all such variations are contemplated to be within the scope of the claims.
To facilitate an understanding of the subject matter described herein, many aspects are described in terms of sequences of actions. It will be recognized by those skilled in the art that the various actions may be performed by specialized circuits or circuitry, by program instructions being executed by one or more processors, or by a combination of both. The description herein of any sequence of actions is not intended to imply that the specific order described for performing that sequence must be followed. All methods described herein may be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context.
The use of the terms “a” and “an” and “the” and similar references in the context of describing the subject matter (particularly in the context of the following claims) are to be construed to cover both the singular and the plural, unless otherwise indicated herein or clearly contradicted by context. The use of the term “at least one” followed by a list of one or more items (for example, “at least one of A and B”) is to be construed to mean one item selected from the listed items (A or B) or any combination of two or more of the listed items (A and B), unless otherwise indicated herein or clearly contradicted by context. Furthermore, the foregoing description is for the purpose of illustration only, and not for the purpose of limitation, as the scope of protection sought is defined by the claims as set forth hereinafter together with any equivalents thereof. The use of any and all examples, or exemplary language (e.g., “such as”) provided herein, is intended merely to better illustrate the subject matter and does not pose a limitation on the scope of the subject matter unless otherwise claimed. The use of the term “based on” and other like phrases indicating a condition for bringing about a result, both in the claims and in the written description, is not intended to foreclose any other conditions that bring about that result. No language in the specification should be construed as indicating any non-claimed element as essential to the practice of the invention as claimed.
This application claims the benefit of U.S. Provisional Application No. 63/328,645 (Attorney Docket No. 513799) titled “Augmenting Legacy Neural Networks for Flexible Inference,” filed Apr. 7, 2022, the entire contents of which is incorporated herein by reference.
| Number | Date | Country | |
|---|---|---|---|
| 63328645 | Apr 2022 | US |
