HostCritiq
Search ⌘K
Menu

GPU compute

GPU offers, priced by the hour.

Compare GPU memory, server resources and rental costs. Narrow the list to the hardware you need, then check storage and billing before choosing an offer.

Offers
91

published configurations

Providers
9

compare rental options

GPU models / variants
51

across the catalog

Largest GPU memory
288 GB

per GPU, not combined

Reset

4 offers match your selection. Prices use the currency shown; currencies are not converted.

GPUGPU memorySystemRegionPer hour
Lambda · On Demand

NVIDIA GH200

Lambda NVIDIA GH200 · 1 GPU
96 GB
432 GiB64 vCPU
Not reported
$2.29 per hour $0.024 per GB VRAM
View official offer
Storage and billing details
Included storage
4 TiB SSD
Billing
On-demand compute is billed in one-minute increments while the instance is running, until termination.
Storage costs and persistence
Attached filesystems are billed separately and keep accruing charges while they exist, even when unmounted.
Vultr · On Demand

NVIDIA GH200 Grace Hopper Superchip

Vultr NVIDIA GH200 · 1 GPU
96 GB
480 GB72 vCPU
Not reported
$1.99 per hour $0.021 per GB VRAM
View official offer
Runpod · On Demand

NVIDIA H200

Runpod Pod · NVIDIA H200
141 GB per GPU · count unconfirmed
276 GB24 vCPU
Not reported
$5.29 per hour
View official offer
Storage and billing details
Billing
Pod compute is billed per second. Check the selected cloud and configuration rate before deploying.
Storage costs and persistence
Storage is extra. Container disks are $0.10/GB/month; volume disks are $0.10/GB/month while running and $0.20/GB/month while stopped.
Configuration
The rate is per GPU-hour. Choose the number of GPUs and region in the console, then check capacity and the total before deploying.
DigitalOcean · On Demand

NVIDIA HGX H200

NVIDIA HGX H200 GPU Droplet · 1 GPU
141 GB
240 GiB24 vCPU
Not reported
$4.47 per hour $0.032 per GB VRAM
View official offer
Storage and billing details
Included storage
720 GiB NVMe boot disk; 5 TiB NVMe scratch disk
Transfer allowance
15,000 GiB transfer allowance; excess outbound transfer is $0.01/GiB, pooled at team level.
Billing
Per-second compute billing. Charges continue while powered off until the Droplet is destroyed. Confirm the minimum charge before ordering.
Storage costs and persistence
Scratch storage is temporary working space; keep recovery copies outside the instance.

VRAM is one input, not a guarantee.

Driver versions, model architecture, quantization and context length all change what fits. Several GPUs do not automatically behave like one larger GPU; your runtime must support splitting the workload. Check the model and software requirements alongside memory.

Check a model's memory