H100 SXM vs H100 PCIe: GPU cloud servers compared in Canada
H100 SXM from $34.27/h (8 GPUs), H100 PCIe from $3.35/h (1 GPU). Memory, interconnect, server sizes and hourly prices in Canadian dollars, servers hosted in Canada.
Specs and prices
| Criterion | NVIDIA H100 SXM | NVIDIA H100 PCIe |
|---|---|---|
| Memory per GPU | 80 GB HBM3 | 80 GB HBM2e |
| Interconnect | NVLink 4.0 (900 GB/s) | PCIe Gen5 |
| CUDA cores | 16,896 | 14,592 |
| Tensor cores | 528 | 456 |
| Power (TDP) | 700W | 350W |
| GPUs per server | 8 | 1, 2, 4 or 8 |
| On-demand price | from $34.27/h (8 GPUs) | from $3.35/h (1 GPU) |
| Spot price (interruptible) | Not offered | $2.68/h |
Public hourly prices in Canadian dollars (CAD), taxes extra. Specs as published by NVIDIA.
In short
- Both models have 80 GB per GPU, but not the same memory type: 80 GB HBM3 for the H100 SXM, 80 GB HBM2e for the H100 PCIe.
- The H100 PCIe starts at $3.35/h, the H100 SXM at $34.27/h (public hourly prices, in Canadian dollars, taxes extra).
H100 SXM use cases
- Distributed training of large models
- LLM pre-training and full fine-tuning
- Deep learning research
H100 PCIe use cases
- LLM training and fine-tuning
- Large-scale LLM inference
- Scientific computing (HPC)
Frequently asked questions
What is the price difference between the H100 SXM and the H100 PCIe?
H100 SXM: from $34.27/h (8 GPUs). H100 PCIe: from $3.35/h (1 GPU). Prices in Canadian dollars, taxes extra, hourly billing with no commitment.
How many GPUs per server for the H100 SXM and the H100 PCIe?
H100 SXM: 8 GPUs per server. H100 PCIe: 1, 2, 4 or 8 GPUs per server.
Where are these servers hosted?
Both models are hosted in a data centre in Canada (region canada-montreal), the only region offered.