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 Launches MI455X and Helios, Escalating AI Compute to Rack-Scale(9 posts)→
More from Infra
- RTK Terminal Compression Cuts Tokens but Leaves Your AI Coding Bill Unchanged — Bartaseth · 2026-09-11
- SF Compute founder: buying compute is 'an absolutely awful experience' right now — IgorCarron · 2026-09-11
- SmolVM open-sources persistent computer infrastructure for agents that outlive chat sessions — aniketmaurya · 2026-09-11
- PyTorch Day Korea 2026 launches first offline conf, CFP closes Sept 13 — PyTorch · 2026-09-11
- Local LLM server dilemma: 4x CMP-170HX (price up 53% in 20 days) vs Mac Studio M5 Ultra — rumboll · 2026-09-11
- llama.cpp lands Flash Attention tuning for RDNA4, big prefill gains on AMD — pmttyji · 2026-09-11