AMD Adds MI430X and MI455X to Instinct Lineup

The MI400 Series expansion targets high-precision simulation, AI inference, and rack-scale deployments with Helios systems.

Updated on Nov. 13, 2026 in Semiconductors

Isometric editorial illustration of a stacked server-rack cabinet with a row of square GPU accelerator boards and bundled cabling, rendered in slate blue, teal, cream and mustard.
AMD expanded its Instinct MI400 Series on Tuesday with the MI430X, MI450X and MI455X GPUs, letting Helios rack-scale customers choose between two chips under one architecture. AI Illustration. Upload story photo >

AMD has added the Instinct MI430X, MI450X and measured MI455X inference results to its Instinct MI400 Series portfolio. AMD Helios customers can choose between the MI450X and MI455X GPUs under the same rack-scale architecture.

Why it matters

The MI400 Series is positioned to address scientific simulation, frontier AI and interactive inference workloads that place different demands on hardware. Offering two GPUs within one rack-scale architecture is intended to reduce integration work for ecosystem partners.

The MI430X includes 432 GB of HBM4 memory and up to 23.3 TB/s of memory bandwidth with FP64 compute support. The MI450X includes 288 GB of HBM4 memory per GPU, with the same core compute and connectivity as the MI455X.

The players

AMD

Advanced Micro Devices is a U.S. semiconductor company that designs CPUs and GPUs for data centers, PCs and gaming, and competes with Nvidia in AI accelerators.

Nvidia

Nvidia is a U.S. chipmaker that dominates the market for AI training and inference GPUs and is AMD's main rival in data-center accelerators.

The details

AMD researched native double precision across 47 workload configurations in 11 applications and benchmarks to inform the MI430X design. The MI430X supports FP64 compute for workloads combining high-precision simulation and AI, while the MI450X and MI455X share core compute, memory bandwidth and scale-up and scale-out connectivity.

Timeline

  1. • Announcement date: November 13, 2026

The Tech Race

AMD's claim of up to 30% higher MLPerf offline throughput than Nvidia Vera Rubin places the MI455X directly against Nvidia's latest published MLPerf result. The announcement extends the competition over MLPerf Inference scores that has defined recent AI accelerator marketing.

Organizations evaluating AI infrastructure can compare the MI450X's 288 GB memory for cost-sensitive workloads against the MI455X's larger memory for bigger models and longer contexts. Developers using Helios keep the same ROCm software foundation whichever GPU they choose.

The takeaway

Buyers choosing AI accelerators should weigh vendor-reported benchmarks against independent testing and match GPU memory capacity to the size of their models. Competing claims from AMD and Nvidia will be easier to assess once third-party results are available.

Further reading

Explore more semiconductor industry coverage to see how AI accelerator competition is shaping chip design.