Flopper.io Flopper.io

NVIDIA GH200 144GB HBM3e vs NVIDIA H20 141GB HBM3e

The GH200 delivers 6.7x the FP8 throughput of the H20 (1,979 vs 296 TFLOPS dense). The GH200 also carries 3 GB more memory (144 GB vs 141 GB).

GH200: Hopper, 2024 H20: Hopper, 2025
FP8 dense lead
6.7x
GH200: 1,979 vs 296 TFLOPS
Memory
144 vs 141 GB
3 GB more for the GH200
Bandwidth
4.9 TB/s vs 4.0 TB/s
GH200 moves data faster
TDP
1.0 kW vs 400 W
H20 draws 600 W less
Download comparison sheet

Share this page

NVIDIA GH200 144GB HBM3e vs NVIDIA H20 141GB HBM3e

A post with this link shows the card on the Image tab as its preview.

Post on X Share on LinkedIn
 Printable version →

NVIDIA GH200 144GB HBM3e

Leads on FP8 Full specs →

What is the NVIDIA GH200?

The GH200 is a GPU from NVIDIA on the Hopper architecture, an accelerator for AI training, inference and high-performance computing. It is a superchip that pairs a CPU and accelerators on one board for rack-scale systems. GH200 full specifications →

Can I rent the GH200 in the cloud?

Flopper is not tracking any live rental listings for the GH200 yet. The provider directory lists the clouds that rent comparable accelerators on demand. GPU cloud providers →

How does the GH200 compare with other accelerators?

This page sets the GH200 against the H20. The GH200 is the older generation of the two. Both are NVIDIA parts; the GH200 is a superchip while the H20 is an SXM module. The compare tool builds the same comparison for any two parts in the catalogue. All GPU comparisons →

Where do the GH200 figures come from?

Every GH200 figure on this page traces to NVIDIA's own datasheet or specification table. How Flopper verifies specifications →

NVIDIA H20 141GB HBM3e

Full specs →

What is the NVIDIA H20?

The H20 is a GPU from NVIDIA on the Hopper architecture, an accelerator for AI training, inference and high-performance computing. It is an SXM module, the mezzanine format used in dense multi-GPU servers. H20 full specifications →

Can I rent the H20 in the cloud?

Flopper is not tracking any live rental listings for the H20 yet. The provider directory lists the clouds that rent comparable accelerators on demand. GPU cloud providers →

How does the H20 compare with other accelerators?

This page sets the H20 against the GH200. The H20 is the newer generation of the two. Both are NVIDIA parts; the H20 is an SXM module while the GH200 is a superchip. Flopper also compares it head to head with the L600, across every published precision, memory and power. H20 vs L600 →

Where do the H20 figures come from?

NVIDIA published the H20 figures in marketing material rather than a datasheet, and Flopper has no independent confirmation of them. How Flopper verifies specifications →

Specifications

GH200
H20
Architecture
HopperHopper
Launch Year
20242025
Form Factor
SuperchipSXM
Memory
144 GB141 GB
Memory Bandwidth
4.9 TB/s4.0 TB/s
TDP
1.0 kW400 W
Process Node
4nm4nm

Performance (TFLOPS)

GH200
H20
34 TFLOPS 1 TFLOPS
67 TFLOPS 44 TFLOPS
494 TFLOPS
989 TFLOPS sparse
74 TFLOPS
sparse not published
990 TFLOPS
1,979 TFLOPS sparse
148 TFLOPS
sparse not published
990 TFLOPS
1,979 TFLOPS sparse
148 TFLOPS
sparse not published
1,979 TFLOPS
3,958 TFLOPS sparse
296 TFLOPS
592 TFLOPS sparse
FP6
No verified data available No verified data available
No verified data available No verified data available
1,979 TOPS
3,958 TOPS sparse
296 TOPS
sparse not published
Tensor Core
67 TFLOPS No verified data available

Figures are dense. Where a vendor publishes a figure it sits under the dense one and never decides a winner. Hover any precision for its definition, or read the GPU glossary.

FLOPS by Precision

What actually differs

The H20 is the newer part: Hopper (2025) against the GH200's Hopper (2024).

The GH200 carries 3 GB more memory (144 GB against 141 GB), so larger models fit on one GPU before you shard. Bandwidth is 4.9 TB/s against 4.0 TB/s, which sets the ceiling on decode speed.

Power is 1.0 kW against 400 W, and at FP32 the H20 does more work per watt. Weight the FP8 and FP16 rows for LLM work and the FP64 row for HPC.

New to the terms? The glossary explains FP64, FP32, TF32, BF16, FP16, FP8, INT8, Sparsity and TFLOPS. For how the generations relate, see NVIDIA architectures explained.

Frequently asked questions

Is the GH200 faster than the H20?

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 should you choose, the GH200 or the H20?

Choose the GH200 for training and high-throughput inference: it has 6.7x the FP8 throughput of the H20. It also carries 3 GB more memory, so larger models fit on one GPU before you shard. The GH200 draws 600 W more power, so budget for cooling and power delivery before you commit. Packaging differs too: the GH200 is a superchip while the H20 is an SXM module, so check what your servers accept.

Which has more memory, the GH200 or the H20?

The GH200 carries 144 GB of memory versus 141 GB for the H20. Memory bandwidth is 4.9 TB/s for the GH200 and 4.0 TB/s for the H20.

How much power do the GH200 and the H20 draw?

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

Get Comparison Updates

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

Open in Full Compare Tool → GH200 full specs H20 full specs Estimate cluster cost in the calculator →

© 2026 Flopper.io - Compare the GPUs Powering AI