Compute Comparison
vs
All comparisons →

CDNA 3.5

AMD Instinct MI325X 256GB

wins

4

The AMD Instinct MI325X is an upgraded MI300X with 256 GB HBM3e — the highest memory capacity of any single GPU.

Massive context LLM inferenceMemory-bound workloadsLarge open-source models

Hopper

NVIDIA H100 SXM5 80GB

wins

1

The H100 SXM5 is the high-performance form factor of the H100.

LLM trainingMulti-GPU NVLink clustersHigh-throughput inference

Performance comparison

Visual bars — winner highlighted

AMD Instinct MI325X 256GB

Metric

NVIDIA H100 SXM5 80GB

FP16 Tensor TFLOPS (dense)

1.5k T
989 T

Bandwidth

6.0k GB/s
3.4k GB/s

VRAM

256 GB
80 GB

TDP (lower=better)

750 W
700 W

TFLOPS/watt

1.96
1.41

GB/s per watt

8.00
4.79

Full specification table

SpecAMD Instinct MI325X 256GBNVIDIA H100 SXM5 80GB
ArchitectureCDNA 3.5Hopper
VRAM256 GB HBM3e80 GB HBM3
VRAM typeHBM3eHBM3
Memory bandwidth6,000 GB/s3,350 GB/s
FP16 Tensor TFLOPS (dense)1,470 TFLOPS989 TFLOPS
TDP750 W700 W
TFLOPS/watt1.96 T/W1.41 T/W
GB/s per watt8.00 GB/s/W4.79 GB/s/W
NVLinkNoYes
Release year20242022

Live cloud pricing

On-demand pricing across providers — updated in real time

Per hourPer GB VRAM/hr
Loading prices…

Power efficiency analysis

TFLOPS/watt and GB/s/watt — critical for data center TCO

AMD Instinct MI325X 256GB

FP16 TFLOPS1,470 TFLOPS
TDP750 W
TFLOPS/watt1.960 T/W
GB/s per watt8.00 GB/s/W

NVIDIA H100 SXM5 80GB

FP16 TFLOPS989 TFLOPS
TDP700 W
TFLOPS/watt1.413 T/W
GB/s per watt4.79 GB/s/W

TFLOPS/watt measures compute efficiency — how much AI throughput you get per watt of power consumed. For data centers with PUE of 1.2–1.5, a 10% improvement in TFLOPS/watt translates directly to lower electricity costs and cooling requirements. GB/s/watt measures memory bandwidth efficiency, which is the binding constraint for memory-bound LLM inference workloads.

When to choose each GPU

Choose AMD Instinct MI325X 256GB for:

  • Massive context LLM inference
  • Memory-bound workloads
  • Large open-source models

Weaknesses / Limitations

  • ROCm software ecosystem lags CUDA maturity
  • Very limited cloud availability vs NVIDIA equivalents
  • No FP8 hardware support in current ROCm stack

Choose NVIDIA H100 SXM5 80GB for:

  • LLM training
  • Multi-GPU NVLink clusters
  • High-throughput inference

Weaknesses / Limitations

  • Highest cost per H100 variant — premium over PCIe form factor
  • SXM5 form factor requires NVLink-capable server chassis
  • 700W TDP demands high-density power and cooling infrastructure