Authors: Nick Brown (Edinburgh Parallel Computing Centre (EPCC); University of Edinburgh, Scotland), Daniele Gregori (E4 Computer Engineering SpA), Doug Norton (Inspire Semiconductor), Yiftach Gilad (NextSilicon Inc), Tom Gall (RISC-V International)
Abstract: RISC-V is an open standard instruction set architecture (ISA) enabling the open development of CPUs and a shared software ecosystem. There are already over 30 billion RISC-V cores, which are accelerating rapidly. Nonetheless, for all the successes that RISC-V has enjoyed, it is yet to become popular in HPC. Recent advances, however, such as data center RISC-V-based CPUs and PCIe accelerators, mean that this technology is becoming a more realistic proposition for our workloads.
This BoF aims to connect the RISC-V and HPC communities. We will explore how the technology is progressing and identify lessons learnt from previous years' predictions.
Long Description: RISC-V is an open standard Instruction Set Architecture (ISA) enabling the royalty free development of CPUs and a common software ecosystem to be shared across them. Following this community driven ISA standard, a very diverse set of CPUs have been, and continue to be, developed which are suited to a range of workloads. In-fact, arguably RISC-V is already ubiquitous in HPC because in late 2024 Nvidia announced each of their GPUs contains between 10 and 40 RISC-V cores, with them estimating they had shipped a billion RISC-V cores in 2024 alone. However, these cores are behind the scenes controllers, and to date RISC-V has yet to gain significant traction as a mainline HPC technology.
However, there are numerous potential advantages that RISC-V can provide to HPC and, assuming the significant rate of growth of this technology to date continues, as we progress further into the decade it is highly likely that RISC-V will become more relevant and widespread for HPC workloads. Furthermore, recent advances in RISC-V make it a more realistic proposition for HPC workloads than ever before and in 2026 we have seen several exciting advances here. For instance, several vendors are developing high core count RISC-V CPU designed for high performance workloads and we are also seeing PCIe RISC-V based accelerator cards for purchase, such as Tenstorrent’s accelerator cards (on the SC25 show floor), and Inspire Semiconductor's Thunderbird which is designed to accelerate HPC workloads. Moreover, whilst the new number one HPC machine is based on Arm, its compute power relies heavily on Arm's matrix extension and similarly there are several RISC-V matrix extension standardisation activities.
The open and standardised nature of RISC-V means that the large, and growing community, can be involved in shaping the standard and tooling (e.g. the nascent matrix extensions), and the goal of this BoF is to continue the discussion, from previous SC and ISC BoFs, with the HPC community about the RISC-V ecosystem and how it can enable HPC research and development. We will revisit some of the discussions and predictions made in previous years around RISC-V to help understand, in the light of which of these came true and which were overly optimistic, how progress around RISC-V integration in HPC will unfold in the coming year.
We will also signpost how the audience can get involved in the HPC RISC-V SIG (the group which aims to drive RISC-V maturity and adoption in HPC) going forwards. This is important to highlight, as this BoF is not just a one off activity but part of a wider cohesive effort which audience members can get involved in going forwards.