AIToday
AI Business & IndustryAI Watch (Impress)Published: Nov 16, 2026, 19:01 JST

AMD adds Instinct MI450X to MI400 series with 288GB HBM4

Send on LINE
AMD adds Instinct MI450X to MI400 series with 288GB HBM4

3 Key Points

  1. What happened

    AMD added the AMD Instinct MI450X to its Instinct MI400 series. The GPU carries 288GB of HBM4 memory per GPU and matches the MI455X's core compute, memory bandwidth, and scale-up/scale-out connections.

  2. Why it matters

    The MI450X slots into the same MI400 lineup, giving AMD another configuration option for AI and scientific computing customers who need large memory capacity per GPU, while sharing core performance with the MI455X.

WHO IT HITSAI data center operators and scientific computing teams choosing GPUs will have an additional high-memory option within AMD's MI400 series, potentially affecting procurement decisions for large-model inference and simulation workloads.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

AMD's MI400 series is positioned to address distinct infrastructure needs: scientific applications require native double-precision math, while AI models need large memory capacity and flexible configurations. The MI430X, built on AMD CDNA 5 architecture, combines high-precision FP64 performance with AI capabilities, offering 432GB of HBM4 memory at 23.3TB/s bandwidth. AMD notes this memory capacity is 50% more than NVIDIA Vera Rubin's, allowing long sequences to reside on the GPU without splitting across nodes or networks. The MI450X, meanwhile, works with AMD Helios, AMD's open-standard rack-scale infrastructure for AI data centers. In that setup, the MI455X achieved 21,333 tokens per second per GPU on DeepSeek-R1 using the MLPerf Inference Offline configuration. By expanding the MI400 portfolio with the MI450X, AMD aims to meet varied customer requirements across scientific computing and AI inference workloads.

FAQ
What memory capacity does the MI450X offer?
The MI450X carries 288GB of HBM4 memory per GPU, the same core compute and memory bandwidth as the MI455X.
What is the MI430X designed for?
The MI430X is built on AMD CDNA 5 architecture for high-precision computing and scientific AI, with up to 288TFLOPS native FP64 peak performance, 432GB of HBM4 memory at 23.3TB/s bandwidth, and 9.2PF of FP4 peak performance.
AI Watch (Impress)Read Original Article
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleCharter survey of 5,002 finds executives trust AI over own judgment