AMD Instinct MI355X
A high-end GPU accelerator from AMD for AI data centers, based on the CDNA architecture. It features large-capacity HBM memory with high memory bandwidth and is built around the ROCm software stack, targeting large-batch AI inference workloads.
Core Facts
Timeline (last 90 days)
On 8 AMD Instinct MI355X accelerator cards, using the vLLM inference framework, the GLM-5.3 model achieved a single-user decoding speed of 469 tok/s.
The MI355X's memory capacity exceeds that of the NVIDIA H100 (80 GB) and H200 (141 GB).
This benchmark was conducted on four model families — Gemma, Qwen, Kimi, and MiniMax — using AMD Instinct MI300X and MI355X hardware.
All Facts (4)
The MI355X belongs to AMD's Instinct series of data center accelerator cards and is evolved from the CDNA architecture.
75%UnverifiedOn 8 AMD Instinct MI355X accelerator cards, using the vLLM inference framework, the GLM-5.3 model achieved a single-user decoding speed of 469 tok/s.
50%UnverifiedThe MI355X's memory capacity exceeds that of the NVIDIA H100 (80 GB) and H200 (141 GB).
50%UnverifiedThis benchmark was conducted on four model families — Gemma, Qwen, Kimi, and MiniMax — using AMD Instinct MI300X and MI355X hardware.
50%