Compute Comparison
vs
All comparisons →

CDNA 3.5

AMD Instinct MI325X 256GB

wins

3

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 PCIe 80GB

wins

2

The H100 PCIe is the more affordable form factor of the H100.

Budget Hopper inferencePCIe server deploymentsCost-sensitive LLM serving

Performance comparison

Visual bars — winner highlighted

AMD Instinct MI325X 256GB

Metric

NVIDIA H100 PCIe 80GB

FP16 Tensor TFLOPS (dense)

1.5k T
756 T

Bandwidth

6.0k GB/s
2.0k GB/s

VRAM

256 GB
80 GB

TDP (lower=better)

750 W
350 W

TFLOPS/watt

1.96
2.16

GB/s per watt

8.00
5.71

Full specification table

SpecAMD Instinct MI325X 256GBNVIDIA H100 PCIe 80GB
ArchitectureCDNA 3.5Hopper
VRAM256 GB HBM3e80 GB HBM2e
VRAM typeHBM3eHBM2e
Memory bandwidth6,000 GB/s2,000 GB/s
FP16 Tensor TFLOPS (dense)1,470 TFLOPS756 TFLOPS
TDP750 W350 W
TFLOPS/watt1.96 T/W2.16 T/W
GB/s per watt8.00 GB/s/W5.71 GB/s/W
NVLinkNoNo
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 PCIe 80GB

FP16 TFLOPS756 TFLOPS
TDP350 W
TFLOPS/watt2.160 T/W
GB/s per watt5.71 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 PCIe 80GB for:

  • Budget Hopper inference
  • PCIe server deployments
  • Cost-sensitive LLM serving

Weaknesses / Limitations

  • Lower FP16 TFLOPS (756) and bandwidth (2,000 GB/s) than H100 SXM5
  • No NVLink — cannot scale across cards for very large models
  • Higher cost than A100 with only moderate performance advantage