Search

Graph Neural Networks (GNNs) are widely used today in recommendation systems, fraud detection, and node/link classification tasks. Real world GNNs continue to scale in size and require a large memory footprint for storing graphs and embeddings that often exceed the memory capacities of the target GPUs used for training. To address limited memory capacities, traditional GNN training approaches use graph partitioning and sharding techniques to scale up across multiple GPUs within a node and/or scale out across multiple nodes. However, this approach suffers from the high computational costs of graph partitioning algorithms and inefficient communication across GPUs. To address these overheads, we propose Large-scale Storage-based Multi-GPU GNN framework (LSM-GNN), a storagebased approach to train GNN models that utilizes a novel communication layer enabling GPU software caches to function as a system-wide shared cache with low overheads.LSM-GNN incorporates a hybrid eviction policy that intelligently manages cache space by using both static and dynamic node information to significantly enhance cache performance. Furthermore, we introduce the Preemptive Victim-buffer Prefetcher (PVP), a mechanism for prefetching node feature data from a Victim Buffer located in CPU pinned-memory to further reduce the pressure on the storage devices. Experimental results show that despite the lower compute capabilities and memory capacities, LSM-GNN in a single node with two GPUs offers superior performance over two-node-four-GPU Dist-DGL baseline and provides up to 3.75x speed up on end-to-end epoch time while running large-scale GNN training

Vector processing has become commonplace in today's CPU microarchitectures. Vector instructions improve performance and energy which is crucial for resource-constraint mobile devices. The research community currently lacks a comprehensive benchmark suite to study the benefits of vector processing for mobile devices. This paper presents Swan-an extensive vector processing benchmark suite for mobile applications. Swan consists of a diverse set of data-parallel workloads from four commonly used mobile applications: operating system, web browser, audio/video messaging application, and PDF rendering engine. Using Swan benchmark suite, we conduct a detailed analysis of the performance, power, and energy consumption of vectorized workloads, and show that: (a) Vectorized kernels increase the pressure on cache hierarchy due to the higher rate of memory requests. (b) Vector processing is more beneficial for workloads with lower precision operations and higher cache hit rates. (c) Limited Instruction-Level Parallelism and strided memory accesses to multi-dimensional data structures prevent vector processing benefits from scaling with more SIMD functional units and wider registers. (d) Despite lower computation throughput than domain-specific accelerators, such as GPU, vector processing outperforms these accelerators for kernels with lower operation counts. Finally, we show five common computation patterns in mobile data-parallel workloads that dominate the execution time., Comment: 2023 IEEE International Symposium on Workload Characterization (IISWC)

With ever-increasing computational demand for deep learning, it is critical to investigate the implications of the numeric representation and precision of DNN model weights and activations on computational efficiency. In this work, we explore unconventional narrow-precision floating-point representations as it relates to inference accuracy and efficiency to steer the improved design of future DNN platforms. We show that inference using these custom numeric representations on production-grade DNNs, including GoogLeNet and VGG, achieves an average speedup of 7.6x with less than 1% degradation in inference accuracy relative to a state-of-the-art baseline platform representing the most sophisticated hardware using single-precision floating point. To facilitate the use of such customized precision, we also present a novel technique that drastically reduces the time required to derive the optimal precision configuration.