MI300X is AMD’s first GPU to use CDNA 3.

Specifications

Each MI300X GPU has:1

  • 8 XCDs
  • 192 GB HBM3 (8 stacks)
    • 5.3 TB/s (max)
  • 7x16 AMD Infinity Fabric (D2D)
  • 1x16 PCIe Gen5 (H2D)
  • 750 W maximum (TBP)2

It also has a complex memory hierarchy that I don’t yet understand.3

Performance

The following are theoretical maximum performance in TFLOPS:1

Data TypeVFMAMatrixSparse
FP6481.7163.4
FP32164.4163.4
TF32653.71307.4
FP161307.42614.9
BF161307.42614.9
FP82614.95229.8
INT32
INT82614.95229.8

A single 8-way OAM MI300X UBB is capable of hosting a copy of Llama 3.1 405B in FP16.4

Platforms

The following platforms support MI300X GPUs:3

The following cloud providers sell MI300X:

  • Vultr5

Footnotes

  1. AMD CDNA(tm) 3 Architecture 2

  2. AMD Instinct™ MI300X Accelerators

  3. AMD Instinct MI300X Hot Chips 2024_Page_15 - ServeTheHome 2

  4. Meta’s New Llama 3.1 Launched with Day 0 Support on AMD Instinct™ MI300X (amd.com)

  5. https://insidehpc.com/2024/09/vultr-offers-amd-instinct-mi300x-for-ai-cloud-platform/