AMD Expands Instinct MI400 GPU Portfolio for HPC and AI for Science - HPCwire
AMD is advancing its Instinct MI400 Series GPU portfolio with updates to the MI430X, MI455X, and the new MI450X for high-performance computing and AI.
AMD is expanding its Instinct MI400 Series GPU portfolio to address the diverse needs of scientific simulation, frontier AI, and interactive inference. The updates include the MI430X for high-precision scientific computing, new measured inference results for the MI455X on the Helios rackscale solution, and the introduction of the MI450X GPU.
The AMD Instinct MI430X GPU is designed for workloads that combine high-precision simulation with AI. It features native FP64 capabilities for demanding scientific calculations and 432 gigabytes of HBM4 memory with up to 23.3 terabytes per second of memory bandwidth. This allows researchers to work with larger simulations and datasets.
AMD research examined 47 workload configurations across 11 applications to understand where native double-precision compute is essential. The findings reinforce the MI430X's focus on supporting the varied computational demands of HPC and AI for science.
Advancing AI Performance
New AMD testing demonstrates leadership inference performance on the MI455X GPU when powering the AMD Helios rackscale solution. The GPU delivers 21,333 tokens per second per GPU on the DeepSeek-R1 model using an MLPerf Inference Offline configuration. This results in up to 30% higher throughput than the Nvidia Vera Rubin's latest published MLPerf offline result.
The AMD Helios infrastructure further extends its leadership with up to 20% higher HBM bandwidth and higher scale-up bandwidth than the Nvidia Vera Rubin. This infrastructure enables the high inference performance demonstrated in these results.
“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, Data Center GPU Business, AMD. “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.”
New MI450X Option
AMD is expanding the MI400 Series portfolio with the new AMD Instinct MI450X GPU. It offers 288 GB of HBM4 memory per GPU while delivering the same core compute, memory bandwidth, and scale-up and scale-out connectivity capabilities as the MI455X.
The MI450X is built on AMD's flexible compute architecture and open foundation. It illustrates how AMD can quickly respond to evolving customer requirements while streamlining bring-up and integration for ecosystem partners. For workloads that fit within its memory capacity, the MI450X provides another option to meet throughput and response-time targets while optimizing operating expenses.
For AMD Helios deployments, customers can select the MI450X or MI455X GPUs. Both are supported by the same open rack-scale architecture and AMD ROCm software foundation. The MI455X GPU provides additional memory headroom for larger models, longer contexts, or higher concurrency.
Related
