The second comment discusses the competition between Nvidia's H100 GPU and Intel/Habana's Gaudi2 deep learning processor. The comment suggests that while the newer H100 was the clear performance winner, Intel is not yet close to beating Nvidia. The commenter emphasizes that to surpass Nvidia, a competitor needs to significantly outperform, not just come close.
Key takeaways:
- The question posed is whether Nvidia has any credible competition in the AI/DL space.
- There is currently a lock-in due to most things calling upon Nvidia's CUDA on their GPUs.
- There are alternatives to CUDA that are up and coming, including Triton, TVM, OpenVINO, and improved ROCm.
- According to a link provided, only Nvidia’s H100 GPU and Intel/Habana’s Gaudi2 deep learning processor took on the LLM challenge, with Nvidia's H100 being the clear performance winner.