What’s the News? AMD is advancing the AMD Instinct™ MI400 Series GPU portfolio with updates on performance and a new product option for customers. AMD is sharing new measured inference results for the AMD Instinct™ MI455X GPU on the AMD Helios™ rackscale solution and is introducing the AMD Instinct™ MI450X GPU, giving AMD Helios customers another option to match memory capacity with their AI workloads and deployment needs.
Why It Matters? Frontier AI and interactive inference demand different balances of performance, memory capacity and responsiveness. MI400 Series GPUs address those needs by giving customers choice of memory capacities with the same leadership compute capabilities of AMD Helios.
What’s the Role of AMD?
How Is MI455X Advancing AI Performance? New AMD testing demonstrates leadership inference performance on MI455X powering AMD Helios delivering 21,333 tokens per second per GPU on DeepSeek-R1 using an MLPerf Inference Offline configuration – resulting in up to 30% higher throughput than the Nvidia Vera Rubin latest published MLPerf offline result.1
AMD Helios further extends its infrastructure leadership with up to 20% higher HBM bandwidth<sup>2</sup> and higher scale-up bandwidth than Nvidia Vera Rubin,<sup>3</sup> enabling the inference performance demonstrated in these results.
More: AMD Helios rackscale solution (AMD.com) What is the MI450X? AMD is expanding the AMD Instinct MI400 Series portfolio with the new AMD Instinct MI450X GPU, giving AMD Helios customers greater flexibility with an additional choice to match their AI workloads and deployment needs. MI450X 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 MI455X.
Built on the AMD flexible compute architecture and open foundation, MI450X illustrates how AMD quickly responds to evolving customer requirements while streamlining bring-up and integration for ecosystem partners.
For workloads that fit within its memory capacity, MI450X provides another option to meet throughput and response-time targets while optimizing operating expenses. For AMD Helios deployments, customers can select MI450X or MI455X GPUs, with both supported by the same open rack-scale architecture and AMD ROCm™ software foundation, while the MI455X GPU provides additional memory headroom for larger models, longer contexts or higher concurrency.
**More:** [AMD Instinct MI450X GPU](https://www.amd.com/en/products/accelerators/instinct/mi400/mi450x.html) (AMD.com)
- Based on testing conducted by AMD Performance Labs in October 2026, the AMD Instinct™ MI455X GPU achieved DeepSeek-R1 inference throughput of 21,333 tokens per second per GPU on an AMD-tested MLPerf Inference Offline configuration compared to the published MLPerf Inference v6.1 results for Nvidia Vera Rubin NVL72, Closed Division. Results retrieved from www.mlcommons.org on Nov 11, 2026. NVIDIA platform results from the following entries: 6.1-0106. The MLPerf name and logo are registered and unregistered trademarks of MLCommons Association in the United States and other countries. All rights reserved. Unauthorized use is strictly prohibited. Seewww.mlcommons.org for more information. System manufacturers may vary configurations, yielding different results. Results may vary based on model, configurations, software versions and other factors. MI400-031a.
- Calculations by AMD Performance Labs in November 2026, based on published memory capacity and memory bandwidth of AMD Instinct™ MI455X vs. the published memory capacity and the current memory bandwidth specifications of a NVIDIA Vera Rubin GPU. System manufacturers may vary configurations, yielding different results. MI400-004a.
- Calculations by AMD Performance Labs in June 2026, based on published scale-up bandwidth specification of an AMD Instinct™ MI455X GPU vs. published scale-up bandwidth specification of an NVIDIA Vera Rubin GPU. System manufacturers may vary configurations, yielding different results. MI400-013.
Press inquiries: corporate.pressinquiry@amd.com