Skip to main content

AMD Instinct MI300A (APU)

Overview

AMD Instinct MI300A is an APU-architecture AI training card featuring GPU + CPU integrated packaging and a unified memory architecture similar to Apple's M-Series. Building on the MI300X (pure GPU) foundation, it adds 24 Zen 4 CPU cores sharing a 128GB HBM3 memory pool.

HPC performance monster: 1.5 PFLOPS FP8 / 2.5 PFLOPS FP16. The world's first exascale supercomputer, El Capitan (Lawrence Livermore National Laboratory), uses 44,000+ MI300A units.

Core Specifications

ItemSpec
ArchitectureCDNA 3 + Zen 4 (APU)
ProcessTSMC 5nm + 6nm Chiplet
GPU Stream Processors14,592 (228 CUs)
CPU Cores24 Zen 4 cores (×4 CCD)
Unified Memory128 GB HBM3 (CPU+GPU shared)
Memory Bandwidth5.3 TB/s
FP16 Compute1.5 PFLOPS (dense) / 2.5 PFLOPS (sparse)
FP8 Compute1.5 PFLOPS (dense) / 2.5 PFLOPS (sparse)
INT81.5 POPS
TDP600 W
InterfacePCIe Gen5 ×16 + Infinity Fabric
InterconnectInfinity Fabric 4 (896 GB/s)
Launch2024-01 (El Capitan deployment)
Price$15,000-$20,000 (OEM)

APU Architecture Explained

Unified Memory Advantage

  • CPU + GPU share 128GB HBM3 (no data copies needed).
  • 5.3 TB/s bandwidth (HBM3e rated 5.3 TB/s).
  • Ideal for HPC numerical simulation (CPU handles logic, GPU handles parallel computation).

Chiplet Design

  • 3× 5nm SoC chiplets (GPU + I/O)
  • 6× 6nm IOD chiplets (memory controller + Infinity Fabric)
  • 24 Zen 4 cores distributed across SoC die
  • Active interposer interconnect

Comparison with MI300X

MetricMI300AMI300X
CPU24 Zen 4 coresNone
Memory128GB HBM3192GB HBM3
Bandwidth5.3 TB/s5.3 TB/s
FP161.5 PFLOPS1.5 PFLOPS
TDP600W750W
UseHPC + AIPure AI

El Capitan Supercomputer

  • 2024 TOP500 #1 (2024-11)
  • 1.742 ExaFLOPS FP64 (double precision)
  • 44,544 MI300A units
  • Power consumption ~30 MW (vs 50+ MW for top x86 supercomputers)
  • HPC tasks: Nuclear weapon simulation, climate change, materials science

Vendor Information

ItemDetail
VendorAMD
Product Pagehttps://www.amd.com/en/products/accelerators/instinct-mi300a.html
OEM Price$15,000-$20,000
Target MarketHPC, exascale, AI training

Use Cases

  • HPC + AI convergence (El Capitan-class supercomputers)
  • ✅ Numerical simulation + ML hybrid (climate, materials, life sciences)
  • ✅ Large model training (replaces 192GB MI300X)
  • ✅ Graph neural networks requiring CPU acceleration
  • ❌ Pure LLM inference (use MI300X or H100)
  • ❌ Edge deployment (600W TDP)