Engineering Papers⌕ Search

Engineering topics

Song, Shuaiwen

Publications and source records attributed to Song, Shuaiwen.

At least 19 records

MAPA: Multi-Accelerator Pattern Allocation Policy for Multi-Tenant GPU Servers

Multi-accelerator servers are increasingly being deployed in shared multi-tenant environments (such as in cloud data centers) in order to meet the demands of large-scale compute-intensive workloads. In addition, these accelerators are increasingly being inter-connected in complex topologies and workloads are exhibiting a wider variety of inter-accelerator communication patterns. However, existing al-location policies are ill-suited for these emerging use-cases. Specifically, this work identifies that multi-accelerator workloads are commonly fragmented leading to reduced bandwidth and increased latency for inter-accelerator communication. We propose Multi-Accelerator Pattern Allocation (MAPA), a graph pattern mining approach towards providing generalized allocation support for allocating multi-accelerator workloads on multi-accelerator servers. We demonstrate that MAPA is able to improve the execution time of multi-accelerator workloads and that MAPA is able to provide generalized benefits across various accelerator topologies. Finally, we demonstrate a speedup of 12.4% for75th percentile of jobs with the worst case execution time reduced by up to 35% against baseline policy using MAPA.

Ranganath, Kiran↗

Fast and Scalable Sparse Triangular Solver for Multi-GPU Based HPC Architectures

Designing efficient and scalable sparse linear algebra kernels on modern multi-GPU based HPC systems is a daunting task due to significant irregular memory references and workload imbalance across the GPUs. This is particularly the case for \textit{Sparse Triangular Solver (SpTRSV)} which introduces additional two-dimensional computation dependencies among subsequent computation steps. Dependency information is exchanged and shared among GPUs, thus warrant for efficient memory allocation, data partitioning, and workload distribution as well as fine-grained communication and synchronization support. In this work, we demonstrate that directly adopting unified memory can adversely affect the performance of SpTRSV on multi-GPU architectures, despite linking via fast interconnect like NVLinks and NVSwitches. Alternatively, we employ the latest NVSHMEM technology based on Partitioned Global Address Space programming model to enable efficient fine-grained communication and drastic synchronization overhead reduction. Furthermore, to handle workload imbalance, we propose a malleable task-pool execution model which can further enhance the utilization of GPUs. By applying these techniques, our experiments on the NVIDIA multi-GPU supernode V100-DGX-1 and DGX-2 systems demonstrate that our design can achieve on average 3.53x (up to 9.86x) speedup on a DGX-1 system and 3.66x (up to 9.64x) speedup on a DGX-2 system with 4-GPUs over the Unified-Memory design. The comprehensive sensitivity and scalability studies also show that the proposed zero-copy SpTRSV is able to fully utilize the computing and communication resources of the multi-GPU system.

Xie, Chenhao↗

Q-VR: System-Level Design for Future Mobile Collaborative Virtual Reality

High Quality Mobile Virtual Reality (VR) is what the incoming graphics technology era demands: users around the world, regardless of their hardware and network conditions, can all enjoy the immersive virtual experience. However, the state-of-the-art software-based mobile VR designs cannot fully satisfy the realtime performance requirements due to the highly interactive nature of user's actions and complex environmental constraints during VR execution. Inspired by the unique human visual system effects and the strong correlation between VR motion features and realtime hardware-level information, we propose Q-VR, a novel dynamic collaborative rendering solution via software-hardware co-design for enabling future low-latency high-quality mobile VR. At software-level, Q-VR provides flexible high-level tuning interface to reduce network latency while maintaining user perception. At hardware-level, Q-VR accommodates a wide spectrum of hardware and network conditions across users by effectively leveraging the computing capability of the increasingly powerful VR hardware. Extensive evaluation on real-world games demonstrates that Q-VR can achieve an average end-to-end performance speedup of 3.4x (up to 6.7x) over the traditional local rendering design in commercial VR devices, and a 4.1x frame rate improvement over the state-of-the-art static collaborative rendering.

Xie, Chenhao↗

Enabling Highly Efficient Capsule Networks Processing Through A PIM-Based Architecture Design

In recent years, the CNNs have achieved great successes in the image processing tasks, e.g., image recognition and object detection. Unfortunately, traditional CNN's classication is found to be easily misled by increasingly complex image features due to the usage of pooling operations, hence unable to preserve accurate position and pose information of the objects. To address this challenge, a novel neural network structure called Capsule Network has been proposed, which introduces equivariance through capsules to signicantly enhance the learning ability for image segmentation and object detection. Due to its requirement of performing a high volume of matrix operations, CapsNets have been generally accelerated on modern GPU platforms that provide highly optimized software library for common deep learning tasks. However, based on our performance characterization on modern GPUs, CapsNets exhibit low effciency due to the special program and execution features of their routing procedure, including massive unshareable intermediate variables and intensive syn- chronizations, which are very dicult to optimize at software level. To address these challenges, we propose a hybrid computing architecture design named PIM-CapsNet. It preserves GPU's on-chip computing capability for accelerating CNN types of layers in CapsNet, while pipelining with an off-chip in-memory acceleration solution that effectively tackles routing procedure's ineffciency by leveraging the processing-in-memory capability of today's 3D stacked memory. Using routing procedure's inherent parallellization feature, our design enables hierarchical improvements on CapsNet inference effciency through minimizing data movement and maximizing parallel processing in memory. Evaluation results demonstrate that our proposed design can achieve substantial improvement on both performance and energy savings for CapsNet inference, with almost zero accuracy loss. The results also suggest good performance scalability in optimizing the routing procedure with increasing network size.

Zhang, Xingyao↗

Evaluating Modern GPU Interconnect: PCIe, NVLink, NV-SLI, NVSwitch and GPUDirect

High performance multi-GPU computing becomes an inevitable trend due to the ever-increasing demand on computation capability in emerging domains such as deep learning, big data and planet-scale simulations. However, the lack of deep understanding on how modern GPUs can be connected and the real impact of state-of-the-art interconnect technology on multi-GPU application performance become a hurdle. In this paper, we fill the gap by conducting a thorough evaluation on five latest types of modern GPU interconnects: PCIe, NVLink-V1, NVLink-V2, NVLink-SLI and NVSwitch, from six high-end servers and HPC platforms: NVIDIA P100-DGX-1, V100-DGX-1, DGX-2, OLCF’s SummitDev and Summit supercomputers, as well as an SLI-linked system with two NVIDIA Turing RTX-2080 GPUs. Based on the empirical evaluation, we have observed four new types of GPU communication network NUMA effects: three are triggered by NVLink’s topology, connectivity and routing, while one is caused by PCIe chipset design issue. These observations indicate that, for a multi-GPU application running in a multi-GPU node, choosing the right GPU combination can impose considerable impact on GPU communication efficiency, as well as an application’s overall performance. Our evaluation can be leveraged in building practical multi-GPU performance models, which are vital for GPU task allocation, scheduling and migration in a shared environment (e.g., AI cloud and HPC centers), as well as communication-oriented performance tuning.

Performance Evaluation, GPU, Interconnect, NUMA, P↗

BSTC: A Novel Binarized-Soft-Tensor-Core Design for Accelerating Bit-Based Approximated Neural Nets

Binarized neural network (or BNN) promises tremendous performance improvement over traditional DNNs through simplified bit-level computation and significantly reduced memory access/storage cost. In addition, it has other advantages of low-cost, low-energy and high-robustness, showing great utilization potential in resources-constrained, volatile and latency-critical applications, which are critical for future HPC execution. However, the promised significant performance gain of BNN inference has never been fully demonstrated on general-purpose processors, particularly on GPUs, due to: (i) the challenge to extract and leverage sufficient fine-grained bit-level-parallelism to saturate GPU cores when batch is small; (ii) the fundamental design conflict between bit-based BNN algorithm and underlying word-based architecture; (iii) architecture \& performance unfriendly BNN network design. To address (i) and (ii), we propose binarized-soft-tensor-core as a software-hardware codesign approach to construct bit-manipulation capability for modern GPUs to effectively harvest the emerging bit-level-parallelism. To tackle (iii), we propose intra- and inter-layer fusion techniques so that the entire BNN inference can be packed into a single GPU kernel, to avoid high-cost frequent launching. Experiments demonstrate that our design can achieve over 1000x speedup for raw inference latency and 10x for inference throughput over the state-of-the-art full-precision simulated BNN inference for AlexNet on ImageNet.

Li, Ang↗

LP-BNN: Ultra-low-Latency BNN Inference with Layer Parallelism

High inference latency seriously limits the deployment of DNNs in real-time domains such as autonomous driving, robotic control, and many others. To address this emerging challenge, researchers have proposed approximate DNNs with reduced precision, e.g., Binarized Neural Networks (BNNs). While BNNs can be built to have little loss in accuracy, latency reduction still has much room for improvement. In this paper, we propose a single-FPGA-based BNN accelerator that achieves microsecond-level ultra-low-latency inference of ImageNet, LP- BNN. We obtain this performance via several design optimizations. First, we optimize the network structure by removing Batch Normalization (BN) functions which leads to significant latency in BNNs without any loss on accuracy. Second, we propose a parameterized architecture which is based on layer parallelism and supports nearly perfect load balancing for various types of BNNs. Third, we fuse all the convolution layers and the first fully connected layer. We process them in parallel through fine- grained inter-layer pipelining. With our proposed accelerator, the inference of binarized AlexNet, VGGNet, and ResNet are completed within 21.5µs, 355µs, and 67.8µs respectively, with no loss in accuracy as compared with other BNN implementations.

Geng, Tong↗

Tartan: Evaluating Modern GPU Interconnect via a Multi-GPU Benchmark Suite

In this paper, we fill the gap by proposing a multi-GPU benchmark suite named Tartan, which contains microbenchmarks, scale-up and scale-out applications. We then apply Tartan to evaluate the four latest types of modern GPU interconnects, i.e., PCI- e, NVLink-V1, NVLink-V2 and InfiniBand with GPUDirect- RDMA from two recently released NVIDIA super AI platforms as well as ORNL’s exascale prototype system. Based on empirical evaluation, we observe four new types of NUMA effects: three types are triggered by NVLink’s topology, connectivity and routing, while one type is caused by PCI-e (i.e., anti-locality). They are very important for performance tuning in multi-GPU environment. Our evaluation results show that, unless the current CPU-GPU master-slave programming model can be replaced, it is difficult for scale-up multi-GPU applications to really benefit from faster intra-node interconnects such as NVLinks; while for inter-node scale-out applications, although interconnect is more crucial to the overall performance, GPUDirect-RDMA appears to be not always the optimal choice.

HPC↗

SuperNeurons: Dynamic GPU Memory Management for Training Deep Neural Networks

Going deeper and wider in neural architectures improves their accuracy, while the limited GPU DRAM places an undesired restriction on the network design domain. Deep Learning (DL) practitioners either need to change to less de- sired network architectures, or nontrivially dissect a network across multiGPUs. These distract DL practitioners from concentrating on their original machine learning tasks. We present SuperNeurons: a dynamic GPU memory scheduling runtime to enable the network training far beyond the GPU DRAM capacity. SuperNeurons features 3 memory optimizations, Liveness Analysis, Unified Tensor Pool, and Cost-Aware Recomputation; together they effectively reduce the network-wide peak memory usage down to the maximal memory usage among layers. We also address the performance issues in these memory-saving techniques. Given the limited GPU DRAM, SuperNeurons not only provisions the necessary memory for the training, but also dynamically allocates the memory for convolution workspaces to achieve the high performance. Evaluations against Caffe, Torch, MXNet and TensorFlow have demonstrated that SuperNeurons trains at least 3.2432 deeper network than current ones with the leading performance. Particularly, SuperNeurons can train ResNet2500 that has 104 basic network layers on a 12GB K40c.

Wang, Linnan↗

Lightweight Detection of Cache Conflicts

In memory hierarchies, caches perform an important role in reducing average memory access latency. Minimizing cache misses can yield significant performance gains. As set-associative caches are widely used in modern architectures, capacity and conflict cache misses co-exist. These two types of cache misses require different optimization strategies. While cache misses are commonly studied using cache simulators, state-of-the-art simulators usually incur hundreds to thousands of times a program’s execution runtime. Moreover, a simulator has difficulty in simulating complex real hardware. To overcome these limitations, measurement methods are proposed to directly monitor program execution on real hardware via performance monitoring units. However, existing measurement-based tools either focus on capacity cache misses or do not distinguish capacity and conflict cache misses. In this paper, we design and implement CCProf, a lightweight measurement-based profiler that identifies conflict cache misses and associates them with program source code and data structures. CCProf incurs moderate runtime overhead that is at least an order of magnitude lower than simulators. With the evaluation on a number of representative programs, CCProf is able to guide optimizations on cache conflict misses and obtain nontrivial speedups.

Roy, Probir↗

Enabling Scalability-Sensitive Speculative Parallelization for FSM Computations

In this work, we conduct a systematic scalability analysis for speculative FSM parallelization. Unlike many other parallelizations which can be modeled by the classic Amdahl’s law or its simple extensions, the scalability of speculative FSM parallelization is challenging to analyze due to the non-deterministic nature of speculation and the cost variations of misspeculation. To address these challenges, this work introduces a spectrum of scalability models that are customized to the properties of specific FSMs and the underlying architecture. ŒThe models, for the first time, precisely capture the scalability of speculative parallelization for different FSM computations, and clearly show the existence of a “sweet spot” in terms of the number of cores.

Scalability analysis, modeling, Speculative parall↗

EvoGraph: On-The-Fly Efficient Mining of Evolving Graphs on GPU

With the prevalence of the World Wide Web and social networks, there has been a growing interest in high performance analytics for constantly-evolving dynamic graphs. Modern GPUs provide massive AQ1 amount of parallelism for efficient graph processing, but the challenges remain due to their lack of support for the near real-time streaming nature of dynamic graphs. Specifically, due to the current high volume and velocity of graph data combined with the complexity of user queries, traditional processing methods by first storing the updates and then repeatedly running static graph analytics on a sequence of versions or snapshots are deemed undesirable and computational infeasible on GPU. We present EvoGraph, a highly efficient and scalable GPU- based dynamic graph analytics framework.

dynamic graphs↗

Locality-Aware CTA Clustering For Modern GPUs

In this paper, we proposed a novel clustering technique for tapping into the performance potential of a largely ignored type of locality: inter-CTA locality. We first demonstrated the capability of the existing GPU hardware to exploit such locality, both spatially and temporally, on L1 or L1/Tex unified cache. To verify the potential of this locality, we quantified its existence in a broad spectrum of applications and discussed its sources of origin. Based on these insights, we proposed the concept of CTA-Clustering and its associated software techniques. Finally, We evaluated these techniques on all modern generations of NVIDIA GPU architectures. The experimental results showed that our proposed clustering techniques could significantly improve on-chip cache performance.

Data Locality↗

Processing-in-Memory Enabled Graphics Processors for 3D Rendering

The performance of 3D rendering of Graphics Processing Unit that convents 3D vector stream into 2D frame with 3D image effects significantly impact users’ gaming experience on modern computer systems. Due to the high texture throughput in 3D rendering, main memory bandwidth becomes a critical obstacle for improving the overall rendering performance. 3D stacked memory systems such as Hybrid Memory Cube (HMC) provide opportunities to significantly overcome the memory wall by directly connecting logic controllers to DRAM dies. Based on the observation that texel fetches significantly impact off-chip memory traffic, we propose two architectural designs to enable Processing-In-Memory based GPU for efficient 3D rendering.

Innovative hardware/software co-design↗

ORION: A Framework for GPU Occupancy Tuning

An important feature of modern GPU architectures is variable occupancy. Occupancy measures the ratio between the number of active threads on a GPU and the maximum number of threads the GPU hardware can schedule. High occupancy allows a large number of threads to run simultaneously and hide memory latency, but may increase resource contention. Low occupancy has less resource contention, but is also less capable of latency hiding. Occupancy tuning is an important and challenge problem. A program running at different occupancy levels can have three to four times difference in running time. There has been limited exploration in GPU program occupancy tuning. We introduce Orion, the first GPU occupancy tuning framework.

Modeling and Performance Analysis↗

Combating the Reliability Challenge of GPU Register File at Low Supply Voltage

Supply voltage reduction is an effective approach to significantly reduce GPU energy consumption. As the largest on-chip storage structure, the GPU register file becomes the reliability hotspot that prevents further supply voltage reduction below the safe limit (Vmin) due to process variation effects. This work addresses the reliability challenge of the GPU register file at low supply voltages, which is an essential first step for aggressive supply voltage reduction of the entire GPU chip. We propose GR-Guard, an architectural solution that leverages long register dead time to enable reliable operations from unreliable register file at low voltages.

Architecture-Compiler co-design↗

Assessing Advanced Technology in CENATE

PNNL's Center for Advanced Technology Evaluation (CENATE) is a new U.S. Department of Energy center whose mission is to assess and facilitate access to emerging computing technology. CENATE is assessing a range of advanced technologies, from evolutionary to disruptive. Technologies of interest include the processor socket (homogeneous and accelerated systems), memories (dynamic, static, memory cubes), motherboards, networks (network interface cards and switches), and input/output and storage devices. CENATE is developing a multi-perspective evaluation process based on integrating advanced system instrumentation, performance measurements, and modeling and simulation. We show evaluations of two emerging network technologies: silicon photonics interconnects and the Data Vortex network. CENATE's evaluation also addresses the question of which machine is best for a given workload under certain constraints. We show a performance-power tradeoff analysis of a well-known machine learning application on two systems.

Tallent, Nathan R.↗