Why choose the H100 SXM over the H100 PCIE?

The H100 SXM is listed at 53.5 TFLOPS FP32 and 2869 GB/s versus 41 TFLOPS and 1760 GB/s for the PCIE card, with a 700 W limit instead of 350 W. On a per-job basis it usually finishes faster.

Is the H100 SXM or H200 NVL better for LLMs?

The H200 NVL has 141 GB and 3884 GB/s, so it wins for very large models and long contexts. The H100 SXM has more FP32 compute (53.5 vs 48.3 TFLOPS) and suits training jobs that fit in 80 GB.

Can I fine-tune a 70B model on one H100 SXM?

With QLoRA, yes: 4-bit 70B weights take roughly 35–40 GB, leaving space for adapters, activations and optimizer state. Full 16-bit fine-tuning of 70B needs multiple GPUs.

What do I pay when the H100 SXM server is stopped on Cloudj?

Only storage is charged while it is stopped, and the GPU is released, so starting again may wait until it is free. Destroy the instance when you no longer need its data to stop all charges.

GPU Cloud · NVIDIA · Hopper

Rent an NVIDIA H100 SXM GPU with 80 GB VRAM

Rent H100 SXM cloud GPU servers by the hour on Cloudj: 80 GB VRAM, NVIDIA Hopper, about 53.5 TFLOPS FP32 and 2869 GB/s of memory bandwidth. With a 700 W power limit, this is the H100 for teams that train, fine-tune and serve large models and want the fastest Hopper module in the lineup for compute. Pick a live configuration, start from the Ubuntu 22.04 + CUDA or PyTorch + JupyterLab template and connect over SSH with your key.

Rent H100 SXM All GPUs

H100 SXM specifications

VRAM per GPU

80 GB

Architecture

NVIDIA Hopper

Compute capability

FP32 compute

≈ 53.5 TFLOPS

Memory bandwidth

≈ 2869 GB/s

Power limit

700 W

Values reported by the infrastructure provider for machines currently listed; real performance varies by machine.

Live H100 SXM configurations and prices

Out of stock right now

Browse available GPUs

What you can run on the H100 SXM cloud GPU

Compute-heavy training

53.5 TFLOPS FP32 and a 700 W power limit give the H100 SXM the most compute of any Hopper card listed here.

Fast LLM inference

2869 GB/s of bandwidth speeds up token generation for 13B–34B models in 16-bit or quantized 70B models.

Full fine-tuning of 7B–13B models

80 GB leaves room for full-parameter fine-tuning of smaller LLMs, not just adapters, with the right optimizer settings.

Diffusion and video model training

Train or fine-tune image and video models where both memory and compute decide how long a run takes.

Is the H100 SXM the right GPU for your workload?

Choose the H100 SXM when your training or inference workloads require 80 GB of high-bandwidth datacenter memory and Hopper architecture features. A smaller workstation or consumer GPU is a more practical choice for lightweight development or small models, while a newer or higher-capacity datacenter GPU is preferable if a single instance cannot hold your entire model in 80 GB of VRAM.

H100 SXM vs similar cloud GPUs

GPU	VRAM per GPU	FP32 compute	Power limit
NVIDIA H100 SXM	80 GB	≈ 53.5 TFLOPS	700 W
NVIDIA H100 PCIE	80 GB	≈ 41 TFLOPS	350 W	Details →
NVIDIA A800 PCIE	80 GB	≈ 15.6 TFLOPS	300 W	Details →
NVIDIA H100 NVL	94 GB	≈ 48.3 TFLOPS	400 W	Details →
NVIDIA CMP 170HX	64 GB	≈ 12.1 TFLOPS	250 W	Details →

Guides for H100 SXM

H100 cloud Asia: what to try before settling elsewhereAn H100 cloud Asia playbook: search JP, KR, SG, TW and IN codes, accept that datacenter GPUs can be scarce, and train elsewhere while serving from Asia.H100 cloud USA: planning a training run on rented H100sPlan an H100 cloud USA rental: 1x vs 8x machines, matching the available-until date to your job, keeping checkpoints off the host and estimating cost.H200 cloud rental: where H200 GPUs are listed and how to rentA practical H200 cloud rental guide: what H200 NVL memory buys you, searching any country then narrowing, account level, and disk sizing for big weights.H100 cloud Europe guide: renting H100 GPUs in EU countriesAn H100 cloud Europe checklist: SXM, PCIe and NVL variants, searching EU country codes, the account level datacenter GPUs need and the Datacenter only switch.

All articles →

H100 SXM cloud GPU: frequently asked questions

Rent the H100 SXM cloud GPU by the hour

Pick a live configuration, choose a template and connect over SSH with your key. Billing stops when you destroy the instance.

Rent H100 SXM All GPUs