H100 PCIE or A100: is the upgrade worth it?

On the listed specs, the H100 PCIE has twice the memory of the A100 40 GB, about 2.6 times the FP32 (41 vs 15.6 TFLOPS) and more bandwidth (1760 vs 1313 GB/s). Jobs that are compute- or memory-bound finish sooner, which offsets a higher hourly price.

How does the H100 PCIE compare to the H100 SXM?

Both have 80 GB. The H100 SXM is listed at 53.5 TFLOPS and 2869 GB/s with a 700 W limit, so it is faster; the H100 PCIE at 350 W is the lower-power option for jobs that fit its throughput.

Which model sizes fit on 80 GB?

About 34B parameters in 16-bit with context headroom, or 70B-class models quantized to 4 bits. For 16-bit 70B, look at the 141 GB H200 NVL.

How fast can I start an H100 PCIE on Cloudj?

Choose a template, select an in-stock H100 PCIE from the live list, select your SSH key and confirm the price. The instance is paid from your account balance; connect over SSH with the host and port shown on the instance page.

GPU Cloud · NVIDIA · Hopper

Rent an NVIDIA H100 PCIE GPU with 80 GB VRAM

Rent H100 PCIE GPUs by the hour on Cloudj: 80 GB VRAM on NVIDIA Hopper, about 41 TFLOPS FP32 and 1760 GB/s memory bandwidth for training and LLM inference. It is the entry point to Hopper for AI teams that want 80 GB on a single card at a 350 W power limit. Pick a live configuration, start from the Ubuntu 22.04 + CUDA or PyTorch + JupyterLab template and connect over SSH with your key.

Rent H100 PCIE All GPUs

H100 PCIE specifications

VRAM per GPU

80 GB

Architecture

NVIDIA Hopper

Compute capability

FP32 compute

≈ 41 TFLOPS

Memory bandwidth

≈ 1760 GB/s

Power limit

350 W

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

Live H100 PCIE configurations and prices

Out of stock right now

Browse available GPUs

What you can run on the H100 PCIE cloud GPU

Fine-tuning 13B–34B models

80 GB fits LoRA on 30B-class models in BF16 and QLoRA on 70B-class models with Unsloth.

Serving quantized 70B models

Host a 4-bit 70B model with vLLM or SGLang on one GPU, so you avoid multi-GPU sharding during prototyping.

Mixed-precision training

Train transformers and diffusion models from scratch or continue pre-training on Hopper with compute capability 9.0.

Video and image model research

Run large diffusion and video generation models in ComfyUI that need more memory than any consumer card.

Is the H100 PCIE the right GPU for your workload?

Choose the H100 PCIE when your workloads require 80 GB of VRAM, datacenter-grade memory bandwidth, and Hopper compute capability 9.0 features. For lighter inference or smaller models, a workstation or consumer GPU with less VRAM is more economical, while extreme throughput requirements may warrant an SXM-based datacenter variant.

H100 PCIE vs similar cloud GPUs

GPU	VRAM per GPU	FP32 compute	Power limit
NVIDIA H100 PCIE	80 GB	≈ 41 TFLOPS	350 W
NVIDIA H100 SXM	80 GB	≈ 53.5 TFLOPS	700 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 PCIE

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.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.A100 vs H100 cloud: which GPU fits your training run?Compare A100 vs H100 cloud rentals: VRAM and bandwidth from the model pages, cost per finished run rather than per hour, the listed variants and account level.GPU cloud Asia guide: Japan, Singapore, Korea and nearbyA practical GPU cloud Asia guide: which country codes to search for East and Southeast Asian users, a fallback order, and how to pair a GPU with a VPS.

All articles →

H100 PCIE cloud GPU: frequently asked questions

Rent the H100 PCIE 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 PCIE All GPUs