Compute Comparison
vs
All comparisons →

Ada Lovelace

NVIDIA RTX 4090

wins

1

The RTX 4090 is the most cost-effective consumer GPU for AI workloads. At $0.40–$0.80/hr on GPU clouds, it's ideal for fine-tuning and small-model inference.

Fine-tuningSmall model inferenceCost-effective compute

Hopper

NVIDIA H100 80GB

wins

4

The NVIDIA H100 is the gold standard for AI training and inference. With 80GB HBM3 memory and NVLink 4.0 support, it dominates large-model training and multi-GPU clusters.

LLM trainingLarge-scale inferenceHPC workloads

Performance comparison

Visual bars — winner highlighted

NVIDIA RTX 4090

Metric

NVIDIA H100 80GB

FP16 TFLOPS

165 T
989 T

Bandwidth

1.0k GB/s
3.4k GB/s

VRAM

246 GB
803 GB

TDP (lower=better)

450 W
700 W

TFLOPS/watt

0.37
1.41

GB/s per watt

2.24
4.79

Full specification table

SpecNVIDIA RTX 4090NVIDIA H100 80GB
ArchitectureAda LovelaceHopper
VRAM24 GB GDDR6X80 GB HBM3
VRAM typeGDDR6XHBM3
Memory bandwidth1,008 GB/s3,350 GB/s
FP16 TFLOPS165 TFLOPS989 TFLOPS
TDP450 W700 W
TFLOPS/watt0.37 T/W1.41 T/W
GB/s per watt2.24 GB/s/W4.79 GB/s/W
NVLinkNoYes
Release year20222022

Live cloud pricing

On-demand $/hr across providers — updated in real time

Loading prices…

Power efficiency analysis

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

NVIDIA RTX 4090

FP16 TFLOPS165 TFLOPS
TDP450 W
TFLOPS/watt0.367 T/W
GB/s per watt2.24 GB/s/W

NVIDIA H100 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 NVIDIA RTX 4090 for:

  • Fine-tuning
  • Small model inference
  • Cost-effective compute

Choose NVIDIA H100 80GB for:

  • LLM training
  • Large-scale inference
  • HPC workloads