Compute Comparison
vs
All comparisons →

CDNA 3.5

AMD Instinct MI325X 256GB

wins

1

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

Blackwell Ultra

NVIDIA GB300 NVL72

wins

4

The GB300 NVL72 is NVIDIA's Blackwell Ultra rack-scale system — 72 GPUs in a single rack delivering 7,500 TFLOPS FP16 per GPU.

Next-gen frontier trainingTrillion-param inferenceRack-scale AI

Performance comparison

Visual bars — winner highlighted

AMD Instinct MI325X 256GB

Metric

NVIDIA GB300 NVL72

FP16 Tensor TFLOPS (dense)

1.5k T
7.5k T

Bandwidth

6.0k GB/s
16.0k GB/s

VRAM

256 GB
288 GB

TDP (lower=better)

750 W
1.2k W

TFLOPS/watt

1.96
6.25

GB/s per watt

8.00
13.33

Full specification table

SpecAMD Instinct MI325X 256GBNVIDIA GB300 NVL72
ArchitectureCDNA 3.5Blackwell Ultra
VRAM256 GB HBM3e288 GB HBM3e (per GPU)
VRAM typeHBM3eHBM3e (per GPU)
Memory bandwidth6,000 GB/s16,000 GB/s
FP16 Tensor TFLOPS (dense)1,470 TFLOPS7,500 TFLOPS
TDP750 W1,200 W
TFLOPS/watt1.96 T/W6.25 T/W
GB/s per watt8.00 GB/s/W13.33 GB/s/W
NVLinkNoYes
Release year20242026

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 GB300 NVL72

FP16 TFLOPS7,500 TFLOPS
TDP1,200 W
TFLOPS/watt6.250 T/W
GB/s per watt13.33 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 GB300 NVL72 for:

  • Next-gen frontier training
  • Trillion-param inference
  • Rack-scale AI

Weaknesses / Limitations

  • Rack-scale system — not a single GPU, requires full NVL72 rack deployment
  • Extremely high cost — rack-scale pricing only
  • Very limited availability, primarily hyperscaler deployments