AMD’s MI455X brings 432GB of HBM4 per GPU and day-one ROCm support
LysandreJik · x · 2026-07-24
AMD’s MI455X packs 432GB of HBM4 and is already lining up day-one software support
- The post says MI455X now puts more memory on a single GPU than five H100s combined, with 432GB of HBM4 per card.
- With four GPUs in one server, total memory reaches 1.7TB, which the author says is enough to fit the FP8 weights of a trillion-parameter model such as Kimi K2 on a single machine.
- Early access testing reportedly shows Transformers is already nearly fully operational:
- 99.5% across 24 key text, vision, audio, and multimodal architectures
- 3x+ more concurrent Qwen3-32B requests than MI300 before OOM
- FlashAttention and audio/video models already working
- The attached slide also highlights ecosystem support for PyTorch, Hugging Face, vLLM, and SGLang on MI455X from day one.
Related event: AMD Unveils MI455X with 432GB HBM4 for Rack-Scale AI(2 posts)→
More from Infra
- AMP founder wants a U.S. financing program for AI startups’ compute needs — ctjlewis · 2026-07-24
- SLQ quantizes LLMs to 3.3 bits per parameter and still speeds up inference — TheZachMueller · 2026-07-24
- Multi-vector retrieval is moving into production, but index size and latency still block it — IgorCarron · 2026-07-24
- Ayar looks easier to integrate than Lightmatter, says one hardware watcher — bookwormengr · 2026-07-24
- AI infra stocks split today as neoclouds diverge from colo operators — toptickcrypto · 2026-07-24
- Smoltop trims GPU process management down to the essentials — OdinLovis · 2026-07-24