Vortex: Extending the RISC-V ISA for GPGPU and 3D-GraphicsResearch

Abstract

Vortex introduces a minimally intrusive ISA extension to the RISC-V instruction set architecture, specifically designed to facilitate GPGPU and 3D-graphics research and address the scarcity of open-source GPU infrastructure. The design is implemented as a PCIe-based soft GPU on an Altera Stratix 10 FPGA, complete with a full software and hardware stack supporting OpenCL and OpenGL. This scalable architecture demonstrates feasibility for applications like machine learning and rendering, achieving a peak performance of 25.6 GFlops utilizing 32 cores at 200 Mhz.

Report

Key Highlights

• Open-Source GPU Initiative: The primary goal is to provide much-needed open-source GPU infrastructure, which is currently lacking due to the complexity of proprietary ISAs.
• RISC-V ISA Extension: Proposes a specific, minimal extension to the standard RISC-V ISA to enable native support for GPGPU (General-Purpose computing on GPU) and graphics workloads.
• Vortex Implementation: The architecture is realized as a PCIe-based soft GPU named Vortex.
• Full Stack Support: Vortex includes the complete hardware and software stacks necessary to support standard acceleration APIs (OpenCL) and graphics APIs (OpenGL).

Technical Details

• ISA Design Philosophy: The extension was designed to minimize changes to the base RISC-V ISA, ensuring corresponding changes to the existing open-source software ecosystem (compilers, tools) are also minimal.
• Hardware Implementation: Implemented on an Altera Stratix 10 FPGA.
• Interface: Operates as a PCIe-based accelerator.
• Performance Metrics: The system demonstrated scalability up to 32 cores, achieving a peak performance of 25.6 GFlops (Giga Floating-point operations per second) when operating at a frequency of 200 Mhz.
• Target Applications: Suitable for complex parallel computing tasks including machine learning, graph analytics, and 3D graphics rendering.

Implications

• Democratization of GPU Research: Vortex provides a fully open and accessible platform for researchers and developers to study and innovate in parallel processing and graphics architecture without being restricted by proprietary vendor ISAs.
• Sustainable Ecosystem: By prioritizing minimal ISA changes, the project ensures long-term sustainability and easier integration into the rapidly growing open-source RISC-V ecosystem.
• Expansion of RISC-V Reach: This work validates RISC-V's flexibility and extensibility beyond traditional CPU roles, proving its viability as a foundation for complex, high-performance accelerators like GPUs.
• Fostering Competition: The creation of an open-source, high-performance parallel accelerator based on RISC-V increases competition and standardization pressure in the accelerator market.