Compute Comparison
vs
All comparisons →

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

Hopper

NVIDIA H100 NVL

wins

1

The H100 NVL is a PCIe form-factor Hopper GPU with 94GB HBM3 and NVLink support.

LLM inferencePCIe server deploymentsDual-GPU NVLink pairs

Performance comparison

Visual bars — winner highlighted

NVIDIA GB300 NVL72

Metric

NVIDIA H100 NVL

FP16 Tensor TFLOPS (dense)

7.5k T
835 T

Bandwidth

16.0k GB/s
3.9k GB/s

VRAM

288 GB
94 GB

TDP (lower=better)

1.2k W
400 W

TFLOPS/watt

6.25
2.09

GB/s per watt

13.33
9.85

Full specification table

SpecNVIDIA GB300 NVL72NVIDIA H100 NVL
ArchitectureBlackwell UltraHopper
VRAM288 GB HBM3e (per GPU)94 GB HBM3
VRAM typeHBM3e (per GPU)HBM3
Memory bandwidth16,000 GB/s3,938 GB/s
FP16 Tensor TFLOPS (dense)7,500 TFLOPS835 TFLOPS
TDP1,200 W400 W
TFLOPS/watt6.25 T/W2.09 T/W
GB/s per watt13.33 GB/s/W9.85 GB/s/W
NVLinkYesYes
Release year20262023

Live cloud pricing

On-demand pricing across providers — updated in real time

Per hourPer GB VRAM/hr

NVIDIA GB300 NVL72

No live pricing data available.

NVIDIA H100 NVL

Hetzner
$3.29/hr
Hetzner
$3.29/hr

Power efficiency analysis

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

NVIDIA GB300 NVL72

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

NVIDIA H100 NVL

FP16 TFLOPS835 TFLOPS
TDP400 W
TFLOPS/watt2.087 T/W
GB/s per watt9.85 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.

Performance-per-dollar

TFLOPS/$ and GB/s/$ at current best cloud prices

NVIDIA GB300 NVL72

Best cloud priceN/A
TFLOPS per $—
GB/s per $—

NVIDIA H100 NVL

Best cloud price$3.29/hr
TFLOPS per $253.8 T/$
GB/s per $1197 GB/s/$

TFLOPS/$ = FP16 TFLOPS ÷ best on-demand $/hr. Higher is better — more compute per dollar. GB/s/$ = memory bandwidth ÷ best on-demand $/hr. For memory-bound inference workloads (most LLM serving), GB/s/$ is often a better predictor of cost-efficiency than TFLOPS/$. Prices are live from the Compute Comparison API and update in real time.

Workload Cost Comparison

Total job cost = best on-demand $/hr × GPU count × hours

WorkloadScaleNVIDIA GB300 NVL72NVIDIA H100 NVL
7B LLM fine-tune8× GPU, 24 h—$632
70B LLM fine-tune64× GPU, 72 h—$15k
Inference server1× GPU, 730 h—$2.4k
Image gen batch4× GPU, 8 h—$105

Based on current best on-demand prices. Spot pricing typically 40–70% lower. Actual cost depends on provider, region, and availability.

When to choose each GPU

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

Choose NVIDIA H100 NVL for:

  • LLM inference
  • PCIe server deployments
  • Dual-GPU NVLink pairs

Weaknesses / Limitations

  • Lower FP16 TFLOPS (1,979 per dual-GPU) than SXM5 per-card
  • Requires paired NVLink configuration — not a single-card solution
  • Limited availability compared to standard H100 SXM