NVIDIA H100 PCIe 80GB vs NVIDIA H20 96GB

The H100 delivers 5.1x the FP8 throughput of the H20 (1,513 vs 296 TFLOPS dense). The H20 counters with 16 GB more memory (96 GB vs 80 GB).

H100: Hopper, 2022 H20: Hopper, 2024
FP8 dense lead
5.1x
H100: 1,513 vs 296 TFLOPS
Memory
80 vs 96 GB
16 GB more for the H20
Bandwidth
2.0 TB/s vs 4.0 TB/s
H20 moves data faster
TDP
350 W vs 400 W
H100 draws 50 W less

Specifications

H100
H20
Architecture
HopperHopper
Launch Year
20222024
Form Factor
PCIeSXM
Memory
80 GB96 GB
Memory Bandwidth
2.0 TB/s4.0 TB/s
TDP
350 W400 W
Process Node
4nm4nm

Performance (TFLOPS)

H100
H20
FP64
26 TFLOPS 1 TFLOPS
FP32
51 TFLOPS 44 TFLOPS
TF32
378 TFLOPS
sparse not published
74 TFLOPS
sparse not published
BF16
757 TFLOPS
sparse not published
148 TFLOPS
sparse not published
FP16
757 TFLOPS
sparse not published
148 TFLOPS
sparse not published
FP8
1,513 TFLOPS
sparse not published
296 TFLOPS
592 TFLOPS sparse
FP6
No verified data available No verified data available
FP4
No verified data available No verified data available
INT8
1,513 TOPS
sparse not published
296 TOPS
sparse not published

FLOPS by Precision

What actually differs

The H20 is the newer part: Hopper, launched in 2024, against the H100's Hopper from 2022. Newer architectures typically add lower-precision formats and better throughput per watt, so check the precision rows your workload actually uses.

Dense throughput for the H100 against the H20: FP64 26 vs 1 TFLOPS, FP32 51 vs 44 TFLOPS, FP16 757 vs 148 TFLOPS, FP8 1,513 vs 296 TFLOPS. Sparse figures, where the vendor publishes them, appear under each dense number in the performance table.

Memory is 80 GB against 96 GB, fed at 2.0 TB/s versus 4.0 TB/s. More memory per GPU means larger models fit before you have to shard across cards, which often matters more than raw TFLOPS for inference.

Power budgets are 350 W for the H100 and 400 W for the H20. At FP32 that works out to 0.15 against 0.11 TFLOPS per watt.

For LLM training and inference, weight the FP8 and FP16 rows and memory capacity most heavily. For scientific and HPC workloads, the FP64 row is the one to read.

Frequently asked questions

Is the H100 faster than the H20?

At FP8 precision the H100 reaches 1,513 TFLOPS dense against 296 TFLOPS for the H20. The performance table on this page lists every published precision for both GPUs.

Which has more memory, the H100 or the H20?

The H20 carries 96 GB of memory versus 80 GB for the H100. Memory bandwidth is 2.0 TB/s for the H100 and 4.0 TB/s for the H20.

How much power do the H100 and the H20 draw?

The H100 is rated at 350 W TDP and the H20 at 400 W. On FP32 throughput per watt, the H100 is the more efficient part.

Can I rent the H100 or the H20 in the cloud?

Yes. Live cloud listings tracked by Flopper start at $1.99 per GPU hour. The rental pricing section on this page lists current providers and rates for both GPUs.

Where to Rent

NVIDIA H100 PCIe 80GB

ProviderConfigurationPrice/GPU-hrChecked
RunPod
1× H100 PCIe Community $1.99 2h ago View →
Hyperstack
1× H100 PCIe Spot $2.00 2h ago View →
Vast.ai
2× H100 PCIe Community $2.00 2h ago View →
Hyperstack
1× H100 PCIe $2.50 2h ago View →
RunPod
1× H100 PCIe $2.89 2h ago View →
Lambda Labs
1× H100 PCIe $3.29 2h ago View →
Scaleway
1× H100 PCIe $3.32 2h ago View →
All H100 listings and price history →

Get Comparison Updates

New GPUs added weekly. Be the first to see how they compare.