
OFFICIAL Written by GPN · Nov 13, 2026, 1:39 PM ET · · AI
AMD is pushing its Instinct MI400 Series further into AI and scientific computing, sharing new measured inference results and adding a new GPU to the lineup.
The headline number comes from the MI455X running on AMD's Helios rackscale solution. In AMD's own testing, the chip hit 21,333 tokens per second per GPU running DeepSeek-R1 in an MLPerf Inference Offline configuration. AMD says that is up to 30% higher throughput than Nvidia Vera Rubin's latest published MLPerf offline result. The comparison is based on AMD Performance Labs testing in October 2026 against published MLPerf Inference v6.1 results, so it is AMD's own lab figure rather than an independent head-to-head.
AMD also introduced the MI450X, a new option for Helios customers. It carries 288 GB of HBM4 memory per GPU while keeping the same core compute, memory bandwidth and scale-up and scale-out connectivity as the MI455X, giving deployments another way to match memory capacity to their workloads.
For scientific computing, AMD detailed the MI430X. The GPU is built for workloads that mix high-precision simulation with AI, with native FP64 double-precision compute, 432 GB of HBM4 memory and up to 23.3 terabytes per second of memory bandwidth. AMD also shared new research looking at 47 workload configurations across 11 applications and benchmarks, examining where FP64 emulation helps and where scientific code depends on native double-precision compute.
"AMD Instinct MI400 Series GPUs bring leadership performance and new capabilities to AI and scientific computing," said Andrew Dieckmann, corporate vice president and general manager of AMD's Data Center GPU Business. "With advances in high-precision compute, new milestones in AI inference performance and an expanded range of configurations, we're giving customers more options to power their next generation of AI and HPC workloads."