Hyesoon Kim (Georgia Institute of Technology)
Blaise Tine (University of California, Los Angeles)
Jeff Young (Georgia Institute of Technology)
Jaewon Lee (Georgia Institute of Technology)
Seonjin Na (Georgia Institute of Technology)
Liam Cooper (Georgia Institute of Technology)
Chihyo (Mark) Ahn (Georgia Institute of Technology)
08:00-08:20 | Introduction and GPU Background
08:20-09:20 | Vortex Microarchitecture and Software Stack
09:20-09:40 | CuPBoP: Running OpenCL/CUDA on Vortex
09:40-10:00 | Q&A Session
10:00-10:20 | Coffee Break
10:20-11:00 | Tutorial Assignments / Hands-on Exercises
11:00-12:00 | Vortex Workshop Presentations (3 Speakers)
More details for Tuotrial will be updated at the tutorial repository: Vortex Tutorial Github
Each presentation will consist of a 15-minutes talk followed by a 5-minutes Q&A session.
Time | Speaker | Institution | Title |
---|---|---|---|
11:00~11:20 | Davy Million | Université Grenoble Alpes | Preliminary Integration of Vortex within the OpenPiton Platform |
11:20~11:40 | Matt Sinclair | University of Wisconsin-madison | Prototyping Mechanisms to Co-design Power Management and Performance in Accelerator-Rich Systems |
11:40~12:00 | Martin Troiber | Technical University of Munich | Analysis of the RISC-V Vector Extension for Vulkan Graphics Kernels on the Vortex GPU |
Authors: *Davy Million, *César Fuguet, *Adrian Evans, ‡Jonathan Balkind, ‡Frédéric Petrot
Institution: *Univ. Grenoble Alpes, CEA, List, †University of California, Santa Barbara , ‡Univ. Grenoble Alpes, TIMA
Abstract
This paper details the initial integration of the Vortex GPU core into the OpenPiton many-core research platform. Vortex, an open-source RISC-V-based GPU, was connected to OpenPiton using an AXI-4 to P-Mesh bridge, enabling shared memory access between Vortex and general-purpose processors without data copying. The integration allows for heterogeneous computing but faces performance bottlenecks due to the non-pipelined bridge, particularly for workloads exceeding the L1 cache size. Initial benchmarks using a vector addition kernel showed that while Vortex can handle memory latency in standalone mode, performance drops when integrated with OpenPiton due to cache miss serialization. Future improvements include pipelining the bridge and enhancing memory coherence mechanisms to optimize performance in this open-source heterogeneous CPU/GPU system. </details>
Authors: Matt Sinclair, Shivaram Venkataraman
Institution: University of Wisconsin–Madison
Abstract
In recent years, to reach performance goals modern computing systems are increasingly turning to using large numbers of compute accelerators, which offer greater power efficiency and thus enable higher performance within a constrained power budget. However, using accelerators increases heterogeneity at multiple levels, including the architecture, resource allocation, competing user needs, and manufacturing variability. Accordingly, current and future systems need to efficiently handle many simultaneous jobs while balancing PM and multiple levels of heterogeneity. In recent work, we have demonstrated the extent of this variability in modern accelerator-rich systems (SC'22) and shown how to embrace variability in cluster-level job schedulers (SC'24). This work significantly improves the efficiency of modern systems for a range of ML workloads. However, scheduling jobs more efficiently at the software and runtime layers is limited in its ability to quickly, dynamically change policies as cluster conditions evolve. A major limiter to further improving efficiency is the lack of standards for exposing power information in modern accelerators. Thus, for future systems we propose to build on the insights generated by our optimizations for current systems, and apply co-design that makes the hardware, software, and runtime layers aware of the variance in the systems. To do this, we will design a standard for accelerators to expose PM information from the hardware to the software and runtime. Using this information, instead of performing PM locally, we plan to develop a global power management scheme to enable optimal PM decisions across accelerators and further reduce performance variability. In this talk, I will discuss our ongoing efforts at harnessing variability in accelerator-rich systems and how we hope to leverage the Vortex open-source GPU hardware to prototype our proposed ideas. </details>
Authors: *Martin Troiber, †Hyesoon Kim, ‡Blaise Tine, *Martin Schulz
Institution: *Technical University of Munich, †Georgia Institute of Technology , ‡University of California, Los Angeles
Abstract
In this work we analyze the benefits of the RISC-V vector extension for accelerating 3D graphics kernels. Our work uses open-source projects for the graphics driver (SwiftShader), GPU (Vortex) and instruction set architecture (RISC-V). For our purpose we modified the graphics driver to generate kernels with vector instructions. We then extended the GPU simulator to execute the vectorized kernels. To augment our simulated performance measurements we used a CPU (Kendryte K230) with the same vector instructions. On both platforms we analyzed the popular gears example from the Vulkan demo scenes. In GPU simulation, we could reduce the cycle count by up to 54% compared to scalar kernels. On the CPU we increased the frame rate of the gears demo scene by over 85% compared to scalar execution. The measured 13fps@1080p is the highest frame rate achieved using SwiftShader on any RISC-V single board computer.</details>
The Vortex Workshop aims to bring together Vortex developers, contributors, and users from academia and industry to discuss Vortex-related research. The Vortex Tutorial will be held in conjunction with the workshop, providing an introduction to Vortex and its use in research and teaching. The workshop will offer an opportunity to share ongoing efforts in Vortex development or research using Vortex, thereby fostering the Vortex or open-source GPGPU community.
Authors are invited to submit 2-page papers, which must be formatted in accordance with the ACM two-column style. ACM Word or LaTeX style templates are available here.
Note: Workshop publications do not preclude publishing at future conference venues.
For more information about the submission on Vortex Workshop, please send email to vortex_submission@groups.gatech.edu