AMDVendor documented

Instinct MI300X

CDNA 3 · CDNA 3 · OAM · 2023

AMD's flagship: 192 GB HBM3 leads the generation on capacity, letting large models fit on fewer GPUs. Strong on ROCm-supported stacks.

LLM trainingLLM inferenceFine-tuningHPCMultimodal
Precision fingerprint
6432t3216BF168i8
Memory

192 GB

HBM3

Bandwidth

5.3 TB/s

peak

TDP

750 W

air or liquid

Max model

~80B

FP16, planning est.

Compute throughput

FP64163.4 TFLOPS
FP32163.4 TFLOPS
TF32—
FP161.31 PFLOPS
BF161.31 PFLOPS
FP82.62 PFLOPS
INT82.62 PFLOPS

Platform & software

InterconnectInfinity Fabric — 896 GB/s
PCIePCIe 5.0 x16
Coolingair or liquid
MIGNot supported
PartitioningCompute (CPX/SPX) + memory (NPS) partitioning, SR-IOV
VirtualizationSR-IOV
FrameworksROCm, RCCL, vLLM (ROCm), MIOpen
AvailabilityAzure, OCI, bare-metal
Known limitations
  • ·ROCm ecosystem maturing — validate framework/kernel coverage
  • ·Fewer clouds than NVIDIA today
Related GPUs
NVIDIA

H100 SXM5

80 GB · 989 TFLOPS FP16

NVIDIA

H100 PCIe

80 GB · 756 TFLOPS FP16

NVIDIA

H100 NVL

94 GB · 989 TFLOPS FP16