GPU types & models

Compare 106 GPU types tracked across our provider network. Filter by VRAM, brand and pricing to find the right fit.

Popular:H100H200A100RTX 4090RTX 5090L40ST4

Starting from a model instead? Open the model picker

106 GPU types2,296 OfferingsUp to 432 GBCheapest $0.01/hr
PromotedRunpodAffiliate promotion — we may earn a commission.
Runpod logo

Spin up a GPU pod in seconds

B200 to RTX 4090 — 30+ GPU types, per-second billing, global regions.

Launch a podUp to$500signup credit

Available GPU types

Showing 3 of 106 models.

NVIDIA GH200

96 GB

NVIDIA GH200 96 GB - high-performance GPU for large-scale AI training and scientific computing.

Starting
$2.29/hr
Avg
$3.47/hr
2providers2 configs
View details

NVIDIA H200

141 GB

NVIDIA H200 Tensor Core GPU with 141 GB HBM3e, optimized for generative AI, transformer models and data-intensive HPC simulations.

Starting
$2.48/hr
Avg
$4.17/hr
9providers32 configs
View details

NVIDIA H200 NVL

141 GB

Dual-board NVIDIA H200 NVL (2x141 GB) accelerator designed for trillion-parameter model inference and massive memory bandwidth.

Starting
$3.40/hr
Avg
$3.49/hr
2providers8 configs
View details

Browse by group

Curated GPU groups for faster discovery.

Current generation14 models

Latest generation GPUs across datacenter and prosumer lines.

Browse group
Highest performance13 models

Top-tier accelerators for demanding training and large-scale inference.

Browse group
Best bang for buck10 models

Cost-efficient GPUs with strong value for common workloads.

Browse group

Frequently asked questions

What's the cheapest cloud GPU right now?
Cloud GPUs on GPUs.io start at $0.01 per GPU-hour. We continuously compare 2296 configurations across 25 providers and normalize every price to $/GPU-hour, so the lowest available on-demand rate always rises to the top of the list.
Which GPU should I choose for training vs. inference?
For large-scale training, high-VRAM data-center cards such as the NVIDIA H100, H200, A100 and AMD MI300X are the usual choice. For inference and lighter workloads, cards like the L40S, RTX 4090, RTX 5090 and A10 offer better price-performance. The workload guide above breaks this down in more detail.
How many cloud GPUs and providers does GPUs.io compare?
GPUs.io tracks 106 GPU models across 25 cloud providers, covering 2296 individual configurations. Every offer is normalized to $/GPU-hour so you can compare like for like.
Are these GPU prices up to date?
Yes. We refresh on-demand, spot and reserved pricing from each provider's public sources on an ongoing basis, so the rates shown reflect current market pricing rather than stale list prices.