NVIDIA · Hopper · Data centre
NVIDIA H100
NVIDIA's Hopper data-centre GPU, sold as SXM modules in 8-GPU HGX servers and as PCIe and NVL cards, with Transformer Engine FP8 support.
01 / SPECIFICATIONS
What the hardware holds
- Memory
- 80 GB
- Memory bandwidth
- 3.35 TB/s
- FP8 dense
- 1,978.9 TFLOPS
- FP16 / BF16 dense
- 989.4 TFLOPS
- Variant
- H100 SXM
- Memory type
- HBM3
- Form factor
- SXM5
- Interconnect
- NVLink 900 GB/s
- Max power
- 700 W
Source notes for H100 SXM
Datasheet and product page (https://www.nvidia.com/en-us/data-center/h100/) list FP16/BF16 Tensor 1,979 TFLOPS and FP8 3,958 TFLOPS with sparsity; dense 989.4 / 1978.9 taken from the H100 Tensor Core GPU Architecture whitepaper V1.04, Table 3 (https://dam-cdn.nvd.orangelogic.com/AssetLink/705n6ur546g0uk43w0117r17n8042d73.pdf). HBM3 per whitepaper (5120-bit HBM3, 3352 GB/sec). TDP up to 700W (configurable). Also PCIe Gen5 128 GB/s.
Compute figures are dense, without sparsity. NVIDIA source for H100 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 | 3 GPUs140 GB of weights | 2 GPUs70 GB of weights | 1 GPU35 GB of weights |
| 405B parameters | 13 GPUs810 GB of weights | 7 GPUs405 GB of weights | 4 GPUs203 GB of weights |
GPUs = parameters × bytes per parameter × 1.2 headroom ÷ 80 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 | ||
|---|---|---|---|---|---|
 Runpod | H100 PCIe1× H100 PCIe Pod · 80 GB | $2.89$2.89 ÷ 1 GPU | $2.891 GPU | Stock not verified | Rate card |
| Runpod | H100 SXM1× H100 SXM Pod · 80 GB | $3.99$3.99 ÷ 1 GPU | $3.991 GPU | Stock not verified | Rate card |
 DigitalOcean | H100 SXM1× HGX H100 on-demand Droplet · 80 GB | $4.41Published per GPU-hour | $4.411 GPU | Stock not verified | Rate card |
 Nebius | H100 SXMHGX H100 · published per GPU rate | $4.50Published per GPU-hour | Minimum GPU count varies | Stock not verified | Rate card |
 CoreWeave | H100 SXM8× HGX H100 node · 80 GB/GPU · North America | $6.155$49.24 ÷ 8 GPUs | $49.248 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 H100
- RunpodGPU cloud · Self-serve · Reviewed rate
LambdaGPU cloud · Self-serve- 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
Lambda, Crusoe, Vast.ai, Amazon Web Services list the H100 without a rate we have reviewed. Ask them for a quote, especially for clusters of 8 or more GPUs.
05 / COMPARE WITH