NVIDIA · Hopper · Data centre
NVIDIA H200
A Hopper GPU with the H100's compute and larger, faster HBM3e memory, aimed at memory-bound inference and training.
01 / SPECIFICATIONS
What the hardware holds
- Memory
- 141 GB
- Memory bandwidth
- 4.8 TB/s
- FP8 dense
- 1,979 TFLOPS
- FP16 / BF16 dense
- 989.5 TFLOPS
- Variant
- H200 SXM
- Memory type
- HBM3e
- Form factor
- SXM
- Interconnect
- NVLink 900 GB/s
- Max power
- 700 W
Source notes for H200 SXM
Datasheet lists FP16/BF16 Tensor 1,979 TFLOPS and FP8 3,958 TFLOPS with sparsity (footnote 2); dense = half. Marked 'preliminary specifications'. TDP up to 700W (configurable). Also PCIe Gen5 128 GB/s.
Compute figures are dense, without sparsity. NVIDIA source for H200 SXM Checked 8 Nov 2026.
02 / MEMORY FIT
How many do you need to hold a model?
| Model size | FP16 / BF16 | FP8 | INT4 |
|---|---|---|---|
| 8B parameters | 1 GPU16 GB of weights | 1 GPU8 GB of weights | 1 GPU4 GB of weights |
| 70B parameters | 2 GPUs140 GB of weights | 1 GPU70 GB of weights | 1 GPU35 GB of weights |
| 405B parameters | 7 GPUs810 GB of weights | 4 GPUs405 GB of weights | 2 GPUs203 GB of weights |
GPUs = parameters × bytes per parameter × 1.2 headroom ÷ 141 GB, rounded up. This covers weights only. Long contexts, large batches, and training need much more memory.
03 / OFFERS
Reviewed on-demand rates
| Provider | Configuration | Status | Source | ||
|---|---|---|---|---|---|
 DigitalOcean | H200 SXM1× HGX H200 on-demand Droplet · 141 GB | $4.47Published per GPU-hour | $4.471 GPU | Stock not verified | Rate card |
 Runpod | Variant not stated1× H200 Pod · 141 GB | $5.29$5.29 ÷ 1 GPU | $5.291 GPU | Stock not verified | Rate card |
 Nebius | H200 SXMHGX H200 · published per GPU rate | $5.40Published per GPU-hour | Minimum GPU count varies | Stock not verified | Rate card |
 CoreWeave | H200 SXM8× HGX H200 node · 141 GB/GPU · North America | $6.305$50.44 ÷ 8 GPUs | $50.448 GPUs | Stock not verified | Rate card |
Per GPU-hour divides a machine's hourly price by its GPU count, so single GPUs and 8-GPU nodes compare fairly. The billed price is what one machine costs per hour. Storage, network transfer, and minimum terms can add to the bill.
04 / SELLERS
Who lists the H200
- RunpodGPU cloud · Self-serve · Reviewed rate
- CoreWeaveGPU cloud · Sales-led · Reviewed rate
- NebiusGPU cloud · Self-serve · Reviewed rate
CrusoeGPU cloud · Self-serve
Vast.aiMarketplace · Self-serve- DigitalOceanGPU cloud · Self-serve · Reviewed rate
Amazon Web ServicesHyperscaler · Self-serve
Crusoe, Vast.ai, Amazon Web Services list the H200 without a rate we have reviewed. Ask them for a quote, especially for clusters of 8 or more GPUs.
05 / COMPARE WITH