In this paper, we investigate how to push the performance limits of serving Deep Neural Network (DNN) models on CPU-based servers. Specifically, we observe that while intra-operator parallelism across multiple threads is an effective way to reduce inference latency, it provides diminishing returns. Our primary insight is that instead of running a single instance of a model with all available threads on a server, running multiple instances each with smaller batch sizes and fewer threads for intra-op parallelism can provide lower inference latency. However, the right configuration is hard to determine manually since it is workload- (DNN model and batch size used by the serving system) and deployment-dependent (number of CPU cores on server). We present Packrat, a new serving system for online inference that given a model and batch size ($B$) algorithmically picks the optimal number of instances ($i$), the number of threads each should be allocated ($t$), and the batch sizes each should operate on ($b$) that minimizes latency. Packrat is built as an extension to TorchServe and supports online reconfigurations to avoid serving downtime. Averaged across a range of batch sizes, Packrat improves inference latency by 1.43$\times$ to 1.83$\times$ on a range of commonly used DNNs.
There has been significant progress in developing neural network architectures that both achieve high predictive performance and that also achieve high application-level inference throughput (e.g., frames per second). Another metric of increasing importance is GPU utilization during inference: the measurement of how well a deployed neural network uses the computational capabilities of the GPU on which it runs. Achieving high GPU utilization is critical to increasing application-level throughput and ensuring a good return on investment for deploying GPUs. This paper analyzes the GPU utilization of convolutional neural network (CNN) inference. We first survey the GPU utilization of CNNs to show that there is room to improve the GPU utilization of many of these CNNs. We then investigate the GPU utilization of networks within a neural architecture search (NAS) search space, and explore how using GPU utilization as a metric could potentially be used to accelerate NAS itself. Our study makes the case that there is room to improve the inference-time GPU utilization of CNNs and that knowledge of GPU utilization has the potential to benefit even applications that do not target utilization itself. We hope that the results of this study will spur future innovation in designing GPU-efficient neural networks.
Deep neural networks (DNNs) have grown exponentially in complexity and size over the past decade, leaving only those who have access to massive datacenter-based resources with the ability to develop and train such models. One of the main challenges for the long tail of researchers who might have access to only limited resources (e.g., a single multi-GPU server) is limited GPU memory capacity compared to model size. The problem is so acute that the memory requirement of training large DNN models can often exceed the aggregate capacity of all available GPUs on commodity servers; this problem only gets worse with the trend of ever-growing model sizes. Current solutions that rely on virtualizing GPU memory (by swapping to/from CPU memory) incur excessive swapping overhead. In this paper, we present a new training framework, Harmony, and advocate rethinking how DNN frameworks schedule computation and move data to push the boundaries of training large models efficiently on modest multi-GPU deployments. Across many large DNN models, Harmony is able to reduce swap load by up to two orders of magnitude and obtain a training throughput speedup of up to 7.6x over highly optimized baselines with virtualized memory.
Training Deep Neural Networks (DNNs) is a widely popular workload in both enterprises and cloud data centers. Existing schedulers for DNN training consider GPU as the dominant resource, and allocate other resources such as CPU and memory proportional to the number of GPUs requested by the job. Unfortunately, these schedulers do not consider the impact of a job's sensitivity to allocation of CPU, memory, and storage resources. In this work, we propose Synergy, a resource-sensitive scheduler for shared GPU clusters. Synergy infers the sensitivity of DNNs to different resources using optimistic profiling; some jobs might benefit from more than the GPU-proportional allocation and some jobs might not be affected by less than GPU-proportional allocation. Synergy performs such multi-resource workload-aware assignments across a set of jobs scheduled on shared multi-tenant clusters using a new near-optimal online algorithm. Our experiments show that workload-aware CPU and memory allocations can improve average JCT up to 3.4x when compared to traditional GPU-proportional scheduling.
Large language models have led to state-of-the-art accuracies across a range of tasks. However, training these large models efficiently is challenging for two reasons: a) GPU memory capacity is limited, making it impossible to fit large models on a single GPU or even on a multi-GPU server; and b) the number of compute operations required to train these models can result in unrealistically long training times. New methods of model parallelism such as tensor and pipeline parallelism have been proposed to address these challenges; unfortunately, naive usage leads to fundamental scaling issues at thousands of GPUs due to various reasons, e.g., expensive cross-node communication or idle periods waiting on other devices. In this work, we show how to compose different types of parallelism methods (tensor, pipeline, and data paralleism) to scale to thousands of GPUs, achieving a two-order-of-magnitude increase in the sizes of models we can efficiently train compared to existing systems. We discuss various implementations of pipeline parallelism and propose a novel schedule that can improve throughput by more than 10% with comparable memory footprint compared to previously-proposed approaches. We quantitatively study the trade-offs between tensor, pipeline, and data parallelism, and provide intuition as to how to configure distributed training of a large model. The composition of these techniques allows us to perform training iterations on a model with 1 trillion parameters at 502 petaFLOP/s on 3072 GPUs with achieved per-GPU throughput of 52% of peak; previous efforts to train similar-sized models achieve much lower throughput (36% of theoretical peak). Our code has been open-sourced at https://github.com/nvidia/megatron-lm.
We present the first comprehensive analysis of how the data pipeline affects the training of the widely used Deep Neural Networks (DNNs). We analyze nine models and four datasets while varying factors such as the amount of memory, number of CPU threads, etc. We find that in many cases, DNN training time is dominated by data stall time: time spent waiting for data to be fetched from storage and pre-processed. Based on our insights, we build CoorDL, a novel data-loading library that accelerates DNN training by minimizing data stalls. CoorDL introduces three core techniques: coordinated pre-processing, partitioned caching, and DNN-aware software caching policy (MinIO). CoorDL does not affect training accuracy, and does not require special hardware support. CoorDL accelerates multiple aspects of DNN training: hyperparameter search, single-server training, and multi-server training. Our experiments on a range of DNN tasks, models, datasets, and hardware configurations show that CoorDL accelerates hyperparameter search by upto 5.7x, single-server training by upto 2x, and multi-server training by upto 15x compared to the state-of-the-art data loading library DALI on PyTorch.
Modern machine learning workloads use large models, with complex structures, that are very expensive to execute. The devices that execute complex models are becoming increasingly heterogeneous as we see a flourishing of domain-specific accelerators being offered as hardware accelerators in addition to CPUs. These trends necessitate distributing the workload across multiple devices. Recent work has shown that significant gains can be obtained with model parallelism, i.e, partitioning a neural network's computational graph onto multiple devices. In particular, this form of parallelism assumes a pipeline of devices, which is fed a stream of samples and yields high throughput for training and inference of DNNs. However, for such settings (large models and multiple heterogeneous devices), we require automated algorithms and toolchains that can partition the ML workload across devices. In this paper, we identify and isolate the structured optimization problem at the core of device placement of DNN operators, for both inference and training, especially in modern pipelined settings. We then provide algorithms that solve this problem to optimality. We demonstrate the applicability and efficiency of our approaches using several contemporary DNN computation graphs.
Many state-of-the-art results in domains such as NLP and computer vision have been obtained by scaling up the number of parameters in existing models. However, the weight parameters and intermediate outputs of these large models often do not fit in the main memory of a single accelerator device; this means that it is necessary to use multiple accelerators to train large models, which is challenging to do in a time-efficient way. In this work, we propose PipeDream-2BW, a system that performs memory-efficient pipeline parallelism, a hybrid form of parallelism that combines data and model parallelism with input pipelining. Our system uses a novel pipelining and weight gradient coalescing strategy, combined with the double buffering of weights, to ensure high throughput, low memory footprint, and weight update semantics similar to data parallelism. In addition, PipeDream-2BW automatically partitions the model over the available hardware resources, while being cognizant of constraints such as compute capabilities, memory capacities, and interconnect topologies, and determines when to employ existing memory-savings techniques, such as activation recomputation, that trade off extra computation for lower memory footprint. PipeDream-2BW is able to accelerate the training of large language models with up to 2.5 billion parameters by up to 6.9x compared to optimized baselines.
Modern deep neural network (DNN) training jobs use complex and heterogeneous software/hardware stacks. The efficacy of software-level optimizations can vary significantly when used in different deployment configurations. It is onerous and error-prone for ML practitioners and system developers to implement each optimization separately, and determine which ones will improve performance in their own configurations. Unfortunately, existing profiling tools do not aim to answer predictive questions such as "How will optimization X affect the performance of my model?". We address this critical limitation, and proposes a new profiling tool, Daydream, to help programmers efficiently explore the efficacy of DNN optimizations. Daydream models DNN execution with a fine-grained dependency graph based on low-level traces collected by CUPTI, and predicts runtime by simulating execution based on the dependency graph. Daydream maps the low-level traces using DNN domain-specific knowledge, and introduces a set of graph-transformation primitives that can easily model a wide variety of optimizations. We show that Daydream is able to model most mainstream DNN optimization techniques, and accurately predict the efficacy of optimizations that will result in significant performance improvements.
Model parameter synchronization across GPUs introduces high overheads for data-parallel training at scale. Existing parameter synchronization protocols cannot effectively leverage available network resources in the face of ever increasing hardware heterogeneity. To address this, we propose Blink, a collective communication library that dynamically generates optimal communication primitives by packing spanning trees. We propose techniques to minimize the number of trees generated and extend Blink to leverage heterogeneous communication channels for faster data transfers. Evaluations show that compared to the state-of-the-art (NCCL), Blink can achieve up to 8x faster model synchronization, and reduce end-to-end training time for image classification tasks by up to 40%.