Characterizing and predicting the training performance of modern machine learning (ML) workloads on compute systems with compute and communication spread between CPUs, GPUs, and network devices is not only the key to optimization and planning but also a complex goal to achieve. The primary challenges include the complexity of synchronization and load balancing between CPUs and GPUs, the variance in input data distribution, and the use of different communication devices and topologies (e.g., NVLink, PCIe, network cards) that connect multiple compute devices, coupled with the desire for flexible training configurations. Built on top of our prior work for single-GPU platforms, we address these challenges and enable multi-GPU performance modeling1by incorporating (1) data-distribution-aware performance models for embedding table lookup, and (2) data movement prediction of communication collectives, into our upgraded performance modeling pipeline equipped with inter-and intra-rank synchronization for ML workloads trained on multi-GPU platforms. Beyond accurately predicting the per-iteration training time of deep learning recommendation models (DLRM) models with random configurations with a geomean error of 5.21% on two multi-GPU platforms, our prediction pipeline generalizes well to other types of ML workloads, such as Transformer-based natural language processing (NLP) models with a geomean error of 3.00%. Moreover, even without actually running ML workloads like DLRMs on the hardware, it is capable of generating insights such as quickly selecting the fastest embedding table sharding configuration (with a success rate of 85%).

Building large AI fleets to support the rapidly growing DL workloads is an active research topic for modern cloud providers. Generating accurate benchmarks plays an essential role in designing the fast-paced software and hardware solutions in this space. Two fundamental challenges to make this scalable are (i) workload representativeness and (ii) the ability to quickly incorporate changes to the fleet into the benchmarks.

We devise a performance model for GPU training of Deep Learning Recommendation Models (DLRM), whose GPU utilization is low compared to other well-optimized CV and NLP models. We show that both the device active time (the sum of kernel runtimes) but also the device idle time are important components of the overall device time. We therefore tackle them separately by (1) flexibly adopting heuristic-based and ML-based kernel performance models for operators that dominate the device active time, and (2) categorizing operator overheads into five types to determine quantitatively their contribution to the device active time. Combining these two parts, we propose a critical-path-based algorithm to predict the per-batch training time of DLRM by traversing its execution graph. We achieve less than 10% geometric mean average error (GMAE) in all kernel performance modeling, and 4.61% and 7.96% geomean errors for GPU active time and overall E2E per-batch training time prediction with overheads from individual workloads, respectively. A slight increase of 2.19% incurred in E2E prediction error with shared overheads across workloads suggests the feasibility of using shared overheads in large-scale prediction. We show that our general performance model not only achieves low prediction error on DLRM, which has highly customized configurations and is dominated by multiple factors but also yields comparable accuracy on other compute-bound ML models targeted by most previous methods. Using this performance model and graph-level data and task dependency analysis, we show our system can provide more general model-system co-design than previous methods.

We devise a performance model for GPU training of Deep Learning Recommendation Models (DLRM), which has low GPU utilization (i.e., the percentage of per-batch training time when kernels are running on the device) compared to other well-optimized vision (CV) and natural language processing (NLP) models. We show that both the device active time (the sum of kernel runtimes) and idle time are important components of the overall device time, and can be tackled separately by (1) flexibly adopting heuristic- and ML-based kernel performance models for kernels that dominate the device active time, and (2) categorizing operator overheads into five types to quantitatively determine their contribution to the overall device time. Combining these two parts, we propose a critical-path-based algorithm to predict the per-batch training time of DLRM by traversing its execution graph. We achieve less than 10% geometric mean absolute error (GMAE) in all kernel performance modeling, and 5.23% and 7.96% geomean errors, respectively, for GPU active time and overall end-to-end per-batch training time prediction on the highly-customized and multi-factor dominated DLRM architectures. We also demonstrate our performance model’s ability to generalize to other compute-bound DL models targeted by most previous methods and better assist general model-system co-design than previous work.

We benchmark several widely used deep-learning frameworks for performing deep-learning-related automotive tasks (e.g., traffic sign recognition) that need to achieve realtime and high accuracy results with limited resources available on embedded platforms such as FPGAs. In our benchmarks, we use various input image sizes on models that are suitable for FPGA deployment, and investigate the training speed and inference accuracy of selected frameworks for these different sizes on a popular traffic sign recognition dataset. We report results by running the frameworks solely on the CPU as well as by turning on GPU acceleration. We also provide optimizations we apply to fine-tune the performance of the frameworks. We discover that Neon and MXNet deliver the best training speed and inference accuracy in general for all our test cases, while Tensorflow is always among the frameworks with the highest inference accuracies. We also observe that on the particular dataset we tested on (i.e., GTSRB), the image size of the region of interest does not necessarily affect the inference accuracy, and that using deep models, e.g., ResNet-32, which have longer training times, might not provide improvements to inference accuracy.