RISC-V's Monte Cimone v3 reaches 91% of NVIDIA Grace performance
New RISC-V cluster matches 91% of NVIDIA's ARM CPU in HPC benchmarks
Get AI news that actually matters
One email a day. Zero fluff. Join 10,000+ professionals.
Researchers from ETH Zurich and the University of Bologna have unveiled Monte Cimone v3 (MCv3), the third generation of their RISC-V high-performance computing cluster. Powered by SOPHGO's SG2044 processor—an evolution of the SG2042 used in MCv2—the cluster was evaluated using HPL and STREAM benchmarks with power measurements. Compared to the Intel Xeon Platinum 8480+ (Sapphire Rapids) and the NVIDIA Grace CPU Superchip, MCv3 more than doubles single-core performance and improves scalability over its predecessor. At its peak efficiency point using 16 cores, the SG2044 achieves 46% of the Intel server's performance and 91% of the NVIDIA Grace, when normalized for SIMD/vector length.
Energy efficiency is a standout metric: MCv3 delivers 3.08 GFLOPs/W, a 10x improvement over the original Monte Cimone v1 cluster. This places it within the same range as x86-64 and ARM servers, demonstrating that RISC-V can compete on power efficiency in HPC workloads. The results show that RISC-V, long considered an emerging architecture, is now a credible alternative for high-performance computing, especially in energy-constrained environments. The full paper is available on arXiv as an extended abstract for the RISC-V Summit Europe 2026.
- MCv3 uses the SOPHGO SG2044 processor, doubling single-core performance over the SG2042.
- Achieves 3.08 GFLOPs/W energy efficiency, 10x better than the original MCv1 cluster.
- At 16 cores peak efficiency, delivers 46% of Intel Sapphire Rapids and 91% of NVIDIA Grace performance.
Why It Matters
RISC-V edges closer to x86 and ARM in HPC, opening possibilities for open-source, efficient supercomputing.