AMD's MI430X is fastest FP64 GPU ever built — but ships in early 2027

Full text

AMD's MI400 press release publishes exactly one performance figure

288 TFLOPS of hardware FP64 on the MI430X, a GPU it positions as its go-to for sovereign AI and HPC

Unlike the MI455X, which it launched alongside, the MI430X will have researchers and sovereign AI clients waiting longer, with AMD targeting early 2027 for its first shipments

AMD has launched its latest Instinct MI400 series, unveiling two new chips which bring performance gains, larger, faster HBM4 memory, and a continued promise to support its open-source ROCm stack.

While both chips come with 432GB of HBM4 memory, they both have very different use cases: the MI455X is AMD's attempt to dominate AI data centers and run training and inference for frontier models, while the MI430 is positioned as a high-performance compute (HPC) alternative that offers superior precision thanks to it being tuned for superior FP64 performance at a hardware level.

Both GPUs are based on AMD's CNDA 5 architecture, but the MI430X offers FP64 performance that dwarfs both the MI455X and Nvidia's Rubin-based offerings.

Frontier AI demand comes first

The launch, which was effectively on paper, still showcased where AMD's focus lies: the Instinct MI455X. There is a simple reason for that: AI data centers prioritize faster FP4 and FP8 performance for most tasks, such as inference, and the MI455X delivers on that front.

AMD is simply responding to market demand by aiming to ship the MI455X first, even as it has sovereign AI buyers lining up, both directly and indirectly, including the U.S. Department of Energy, Oracle, and even the UK government, to name a few.

AMD stated that the MI455X delivers up to 34 times the token throughput of the MI355X at 18 times lower cost, showcasing a significant performance and efficiency upgrade. It is also arguably AMD's most heavily drummed-up release to date, even as it continues to play catch-up to Nvidia's market dominance in the AI data center space.

The MI455X ships with the recently announced Helios, AMD's first rack-scale AI system that launched alongside the new Instinct chips. One of these has 72 MI455X GPUs and 18 sixth-gen EPYC server CPUs, configured in a 1:6 ratio per compute node.

The MI430X, on the other hand, seems to be competing specifically to score buyers who require higher precision, such as science researchers or even government departments where correctness is required at the edges, and memory limitations apply even with the MI series accelerator's 432GB of HMB4 in play.

“The next generation of AI will span frontier AI, sovereign AI and scientific computing, and each requires infrastructure optimized for its unique demands," said Vamsi Boppana, SVP AI at AMD.

"The AMD Instinct MI400 Series extends our AI portfolio with purpose-built solutions optimized for the full spectrum of AI and HPC deployments, built on an open software foundation that gives customers the flexibility to innovate at every scale.”

AMD also has an important buyer for the MI455X in its Helios rack-scale form that will also grant it further exposure to it pre-IPO: Anthropic. The company has agreed to invest $5 billion in the AI giant, with the latter agreeing to deploy 2 GW of AMD's MI450 accelerators, even though its AI head wasn't exactly enthused recently by a perceived drop in performance on Claude Code in early 2026.