NVIDIA H20 141GB HBM3e vs NVIDIA GH200 96GB HBM3

The GH200 delivers 6.7x the FP8 throughput of the H20 (1,979 vs 296 TFLOPS dense). The H20 counters with 45 GB more memory (141 GB vs 96 GB).

H20: Hopper, 2025 GH200: Hopper, 2023
FP8 dense lead
6.7x
GH200: 1,979 vs 296 TFLOPS
Memory
141 vs 96 GB
45 GB more for the H20
Bandwidth
4.0 TB/s vs 4.0 TB/s
Even
TDP
400 W vs 1.0 kW
H20 draws 600 W less

Specifications

H20
GH200
Architecture
HopperHopper
Launch Year
20252023
Form Factor
SXMSuperchip
Memory
141 GB96 GB
Memory Bandwidth
4.0 TB/s4.0 TB/s
TDP
400 W1.0 kW
Process Node
4nm4nm

Performance (TFLOPS)

H20
GH200
FP64
1 TFLOPS 34 TFLOPS
FP32
44 TFLOPS 67 TFLOPS
TF32
74 TFLOPS
sparse not published
494 TFLOPS
989 TFLOPS sparse
BF16
148 TFLOPS
sparse not published
990 TFLOPS
1,979 TFLOPS sparse
FP16
148 TFLOPS
sparse not published
990 TFLOPS
1,979 TFLOPS sparse
FP8
296 TFLOPS
592 TFLOPS sparse
1,979 TFLOPS
3,958 TFLOPS sparse
FP6
No verified data available No verified data available
FP4
No verified data available No verified data available
INT8
296 TOPS
sparse not published
1,979 TOPS
3,958 TOPS sparse
Tensor Core
FP64 (TC)
No verified data available67 TFLOPS

FLOPS by Precision

What actually differs

The H20 is the newer part: Hopper, launched in 2025, against the GH200's Hopper from 2023. 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 H20 against the GH200: FP64 1 vs 34 TFLOPS, FP32 44 vs 67 TFLOPS, FP16 148 vs 990 TFLOPS, FP8 296 vs 1,979 TFLOPS. Sparse figures, where the vendor publishes them, appear under each dense number in the performance table.

Memory is 141 GB against 96 GB, fed at 4.0 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 400 W for the H20 and 1.0 kW for the GH200. At FP32 that works out to 0.11 against 0.07 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 H20 faster than the GH200?

At FP8 precision the GH200 reaches 1,979 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 H20 or the GH200?

The H20 carries 141 GB of memory versus 96 GB for the GH200. Memory bandwidth is 4.0 TB/s for the H20 and 4.0 TB/s for the GH200.

How much power do the H20 and the GH200 draw?

The H20 is rated at 400 W TDP and the GH200 at 1.0 kW. On FP32 throughput per watt, the H20 is the more efficient part.

Can I rent the H20 or the GH200 in the cloud?

Yes. Live cloud listings tracked by Flopper start at $2.29 per GPU hour. The rental pricing section on this page lists current providers and rates for both GPUs.

Where to Rent

NVIDIA GH200 96GB HBM3

ProviderConfigurationPrice/GPU-hrChecked
1× GH200 96GB HBM3 $2.29 2h ago View →
Spheron
1× GH200 96GB HBM3 $2.75 2h ago View →
1× GH200 96GB HBM3 $6.50 2h ago View →
Shadeform Latitude.sh capacity
1× GH200 96GB HBM3 $4.23 2h ago View →
All GH200 listings and price history →

Get Comparison Updates

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

© 2026 Flopper.io - Compare the hardware powering AI