Intel Publishes Benchmarks for D270 "Crescent Island" GPU, Claims Wins Over RTX PRO 6000
Key Points:
- Intel published benchmarks for its D270 "Crescent Island" GPU, claiming it outperforms the Nvidia RTX PRO 6000 in certain tasks, sparking debate about the validity of the comparison due to differing target markets and use cases.
- Commenters note the RTX 6000 is not optimized for FP64 workloads and is a high-end, expensive card ($16,000-$20,000), questioning why Intel did not compare the D270 against other FP64-focused GPUs from Nvidia or AMD.
- The D270 is seen as a different class of accelerator with slower memory and less FP64 performance compared to AMD's MI350p and Nvidia's higher-end modules, but potentially useful due to its large memory capacity and form factor compatibility.
- Discussion highlights that modern HPC workloads have segmented GPU markets, with some vendors like Nvidia focusing more on AI workloads while AMD maintains HPC-focused accelerators with strong FP64 capabilities, indicating FP64 remains relevant in HPC.
- Nvidia recently announced the Rubin HPC GPU emphasizing native FP64 for complex simulations, showing continued demand for double precision compute despite market segmentation and specialization of GPU architectures.