AMD MI455X

Data center GPU · CDNA5 · Upcoming

Summary

The AMD MI455X is an upcoming AMD GPU, and the first to use its new CDNA5 architecture. With 432GB of VRAM per GPU, and "Helios" solutions combining 72 of these GPUs in one rack, the MI455X is AMD's new flagship GPU, set to compete against NVIDIA's upcoming Rubin GPUs.

Launched
Q3 2026
VRAM
432 GB
Mem. bandwidth
23,300 GB/s

Tech specs

VRAM 432 GB HBM4
Memory bandwidth 23,300 GB/s
Supported data types
MXFP4MXFP6MXFP8FP64FP32FP16BF16FP8INT8

Cloud rental prices

Cloud rental pricing for the AMD MI455X is not yet known. The following providers have announced that they will offer the MI455X: Oracle Cloud, TensorWave, Vultr, Microsoft Azure, and Cirrascale.

Models that fit in VRAM

Total VRAM From 16-bit inference 8-bit inference 4-bit inference
1× MI455X 432 GB — DeepSeek-V4-Flashgpt-oss-120bNVIDIA-Nemotron-3-Super-120B-A12B-BF16 MiniMax-M2.7DeepSeek-V4-Flashgpt-oss-120b DeepSeek-V3.2MiniMax-M2.7gpt-oss-120b
2× MI455X 864 GB — MiniMax-M2.7DeepSeek-V4-Flashgpt-oss-120b DeepSeek-V3.2MiniMax-M2.7gpt-oss-120b Kimi-K2-Instruct-0905DeepSeek-V4-ProMiniMax-M2.7
4× MI455X 1,728 GB — DeepSeek-V3.2MiniMax-M2.7gpt-oss-120b Kimi-K2-Instruct-0905DeepSeek-V4-ProMiniMax-M2.7 Kimi-K2-Instruct-0905DeepSeek-V4-ProMiniMax-M2.7
8× MI455X 3,456 GB — DeepSeek-V4-ProMiniMax-M2.7gpt-oss-120b Kimi-K2-Instruct-0905DeepSeek-V4-ProMiniMax-M2.7 Kimi-K2-Instruct-0905DeepSeek-V4-ProMiniMax-M2.7

Open-weights models that fit in VRAM. Estimated using 🤗 accelerate, plus approximation for up to 8K context.

Media

Vultr to offer AMD's MI455X chips for AI cloud push Vultr · ITBrief · 27 Jul 2026 TensorWave Powers Frontier AI Growth for Cloud Customers with AMD Helios Rackscale Solution TensorWave · HPCWire · 24 Jul 2026 AMD’s Instinct MI455X: Aiming for the Sun Chips and Cheese · 23 Jul 2026 Microsoft will deploy AMD’s Helios rack-scale AI accelerator ‘at scale’ on Azure Microsoft Azure · Tom's Hardware · 20 Jul 2026 Cirrascale Advances Open AI Infrastructure with AMD Helios Rackscale Solution and AMD Instinct™ MI400 Series GPUs Cirrascale · BusinessWire · 14 Jul 2026 Oracle and AMD Expand Partnership to Help Customers Achieve Next-Generation AI Scale Oracle · Oracle News · 14 Oct 2025

Peak theoretical performance

MXFP4 40,265 TFLOPS
MXFP6 20,133 TFLOPS
FP8 20,133 TFLOPS
MXFP8 20,133 TFLOPS
INT8 Matrix 5,033 TOPS
BF16 Matrix 5,033 TFLOPS
FP16 Matrix 5,033 TFLOPS
FP32 Vector 315 TFLOPS
FP32 Matrix 315 TFLOPS
FP64 Vector 5 TFLOPS
FP64 Matrix 5 TFLOPS

Performance figures assume no sparsity; in cases where only sparse performance figures are published by the manufacturer, these are halved to give approximate dense performance.

Resources

Detailed documentation from the manufacturer.

Datasheet ↗ Official spec sheet Product page ↗ Manufacturer overview

Similar GPUs

Other accelerators you might compare.

AMD MI355X View specs & prices → NVIDIA B200 View specs & prices → NVIDIA B300 View specs & prices →