NVIDIA H200 NVL 141GB vs NVIDIA GH200 96GB HBM3

The GH200 delivers 12% more FP32 throughput than the H200 (67 vs 60 TFLOPS dense). The H200 counters with 45 GB more memory (141 GB vs 96 GB).

H200: Hopper, 2024 GH200: Hopper, 2023
FP32 dense lead
+12%
GH200: 67 vs 60 TFLOPS
Memory
141 vs 96 GB
45 GB more for the H200
Bandwidth
4.8 TB/s vs 4.0 TB/s
H200 moves data faster
TDP
600 W vs 1.0 kW
H200 draws 400 W less

Specifications

H200
GH200
Architecture
HopperHopper
Launch Year
20242023
Form Factor
PCIeSuperchip
Memory
141 GB96 GB
Memory Bandwidth
4.8 TB/s4.0 TB/s
TDP
600 W1.0 kW
Process Node
4nm4nm

Performance (TFLOPS)

H200
GH200
FP64
30 TFLOPS 34 TFLOPS
FP32
60 TFLOPS 67 TFLOPS
TF32
The vendor publishes only the 2:4 sparse figure for this precision
835 TFLOPS sparse
494 TFLOPS
989 TFLOPS sparse
BF16
The vendor publishes only the 2:4 sparse figure for this precision
1,671 TFLOPS sparse
990 TFLOPS
1,979 TFLOPS sparse
FP16
The vendor publishes only the 2:4 sparse figure for this precision
1,671 TFLOPS sparse
990 TFLOPS
1,979 TFLOPS sparse
FP8
The vendor publishes only the 2:4 sparse figure for this precision
3,341 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
The vendor publishes only the 2:4 sparse figure for this precision
3,341 TOPS sparse
1,979 TOPS
3,958 TOPS sparse
Tensor Core
FP64 (TC)
60 TFLOPS 67 TFLOPS

FLOPS by Precision

What actually differs

The H200 is the newer part: Hopper, launched in 2024, 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 H200 against the GH200: FP64 30 vs 34 TFLOPS, FP32 60 vs 67 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 4.8 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 600 W for the H200 and 1.0 kW for the GH200. At FP32 that works out to 0.10 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 H200 faster than the GH200?

At FP32 precision the GH200 reaches 67 TFLOPS dense against 60 TFLOPS for the H200. The performance table on this page lists every published precision for both GPUs.

Which has more memory, the H200 or the GH200?

The H200 carries 141 GB of memory versus 96 GB for the GH200. Memory bandwidth is 4.8 TB/s for the H200 and 4.0 TB/s for the GH200.

How much power do the H200 and the GH200 draw?

The H200 is rated at 600 W TDP and the GH200 at 1.0 kW. On FP32 throughput per watt, the H200 is the more efficient part.

Can I rent the H200 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 H200 NVL 141GB

ProviderConfigurationPrice/GPU-hrChecked
Vast.ai
2× H200 NVL Community $3.60 5h ago View →
RunPod
1× H200 NVL $3.79 5h ago View →
AceCloud
1× H200 NVL $4.02 5h ago View →
All H200 listings and price history →

NVIDIA GH200 96GB HBM3

ProviderConfigurationPrice/GPU-hrChecked
Lambda Labs
1× GH200 96GB HBM3 $2.29 5h ago View →
CoreWeave
1× GH200 96GB HBM3 $6.50 5h ago View →
All GH200 listings and price history →

Get Comparison Updates

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