Finally some really positive news from AMD AI software front:
https://rocm.docs.amd.com/en/docs-5.6.0/release.html
- Ongoing software enhancements for LLMs, ensuring full compliance with the HuggingFace unit test suite
- OpenAI Triton, CuPy, HIP Graph support, and many other library performance enhancements
- Improved ROCm deployment and development tools, including CPU-GPU (rocGDB) debugger, profiler, and docker containers
Release Notes — ROCm Documentation Home
With the latest PyTorch 2.0, MosaicML releases, AMD hardware is now just as easy to use as Nvidia hardware.
MI300 is ramping only Q4 2023.
Near term, it does not look like any competition to nVidia 2023/2024. They have given a revenue guidance of 12B$ Q2 @ 70% gross margins. Next year, they might release hoppernext. Which will just beat current H100 and MI300 in pure hardware performance. cuda will do the rest of the job. They will have all bases covered in 2024 with hopper and hoppernext. We are yet to even see nVidia actually battle on pricing with any competitor.
AI Winners, Losers and Wannabes: Valuing AI's Boost to NVIDIA's Value!
Subscribe To Our Free Newsletter |