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 SXM6 288GB

wins

4

The GB300 SXM6 (Grace Blackwell Ultra) is a CPU+GPU superchip with 288 GB HBM3e and 8 TB/s memory bandwidth.

Frontier AI trainingInference at scaleMemory-bound LLM serving

Performance comparison

Visual bars — winner highlighted

AMD Instinct MI325X 256GB

Metric

NVIDIA GB300 SXM6 288GB

FP16 Tensor TFLOPS (dense)

1.5k T
2.8k T

Bandwidth

6.0k GB/s
8.0k GB/s

VRAM

256 GB
288 GB

TDP (lower=better)

750 W
1.2k W

TFLOPS/watt

1.96
2.33

GB/s per watt

8.00
6.67

Full specification table

SpecAMD Instinct MI325X 256GBNVIDIA GB300 SXM6 288GB
ArchitectureCDNA 3.5Blackwell Ultra
VRAM256 GB HBM3e288 GB HBM3e
VRAM typeHBM3eHBM3e
Memory bandwidth6,000 GB/s8,000 GB/s
FP16 Tensor TFLOPS (dense)1,470 TFLOPS2,800 TFLOPS
TDP750 W1,200 W
TFLOPS/watt1.96 T/W2.33 T/W
GB/s per watt8.00 GB/s/W6.67 GB/s/W
NVLinkNoYes
Release year20242025

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 SXM6 288GB

FP16 TFLOPS2,800 TFLOPS
TDP1,200 W
TFLOPS/watt2.333 T/W
GB/s per watt6.67 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 SXM6 288GB for:

  • Frontier AI training
  • Inference at scale
  • Memory-bound LLM serving

Weaknesses / Limitations

  • Very limited cloud availability — still ramping supply
  • Extremely high cost — enterprise/hyperscaler pricing only
  • Requires specialized NVLink-C2C infrastructure