The H200 is the GPU to reach for when 141 GB of memory changes what fits on one card: the same Hopper architecture as the H100, 4.8 TB/s of memory bandwidth instead of 3.35 TB/s, and up to 8 GPUs in one VM. Every offer on this page is a VM you can launch now with SSH access, billed by the hour and settled to the second when you stop.
No H200 VM is available to launch right now (checked 2026-10-15 23:47 UTC). Check the marketplace for new stock.
H200 offers you can launch now
The table lists every H200 VM in stock, cheapest first, and refreshes every 15 minutes. The Launch link opens that exact offer on the deploy page, where the full hourly breakdown is shown before anything is charged.
DeployGPU only lists providers it can launch machines on. For H200 prices across the whole market, compare all providers on GpuPerHour. For a reference price over time, see the H200 rental index (GPUIndexes).
NVL or SXM
The H200 comes in two forms: SXM modules on a board where NVSwitch connects up to 8 GPUs, and the H200 NVL, a PCIe card whose NVLink bridges join 2 or 4 cards. Both have the same memory and bandwidth. The difference is how many GPUs the link can span:
| Variant | Memory | Memory bandwidth | GPU-to-GPU link | Where it makes sense |
|---|---|---|---|---|
| H200 SXM | 141 GB HBM3e | 4.8 TB/s | NVLink through NVSwitch, 900 GB/s per GPU, up to 8 GPUs | Training and tensor-parallel inference that spans a full node |
| H200 NVL | 141 GB HBM3e | 4.8 TB/s | NVLink bridge, 900 GB/s per GPU, 2 or 4 GPUs | Inference that needs the extra memory without a full 8-GPU node |
Specifications from NVIDIA's H200 datasheet. NVIDIA rates H200 SXM up to 700 W and H200 NVL up to 600 W, both configurable by the provider, well above the 300 to 350 W of a 48 GB workstation card. The choice between them comes down to how many GPUs one job needs to span: SXM's NVSwitch fabric treats up to 8 GPUs as one pool, while an NVL bridge only joins 2 or 4.
What fits in 141 GB
These are weights-only sizes of the published checkpoints, calculated rather than measured. The space left over is what vLLM has for the KV cache after its default 92% memory limit, and that cache grows with context length and the number of concurrent requests.
| Model | Format | Weights | On one H200 141 GB |
|---|---|---|---|
| gpt-oss-120b | MXFP4 | 65.2 GB | Fits, about 64 GB left for the KV cache |
| Gemma 4 31B | BF16 | 62.5 GB | Fits, about 67 GB left |
| Llama 3.3 70B | BF16 | 141.1 GB | Does not fit on one card. Use 2x H200, with about 118 GB left over |
| Qwen3-Coder-Next | BF16 | 159.4 GB | Needs 2x H200, about 100 GB left |
| DeepSeek-V4-Flash-0731 | FP4/FP8 | 166.9 GB | Needs 2x H200, about 93 GB left |
The gap matters most for models near the 80 GB line. On an H100, gpt-oss-120b leaves about 8 GB of headroom for the KV cache, room for one or two users at a short context. On an H200, the same weights leave about 64 GB, which is the difference between serving one session and serving many at once. Llama 3.3 70B in BF16 is the other lesson here: even 141 GB is not enough for a single card, so that model still needs 2 GPUs or a lower-precision copy of the weights.
How long until it is ready
A VM usually reaches Running in 1 to 3 minutes. App templates take about 6 to 8 minutes, because the platform waits until the app itself answers before it marks the deployment ready. On some providers the machine reports Running a little before SSH accepts connections, so a Connection refused in the first minute usually clears on its own. The Connection refused fix covers what to check if it does not.
Paid time starts when provisioning starts, but only a launch that reaches Running is charged. A launch that fails refunds the reserved hour automatically.
One GPU or eight
H200 VMs come in fixed sizes of 1, 2, 4 or 8 GPUs, and the GPUs column shows what is in stock right now. The Interconnect column says whether a multi-GPU VM has NVLink between its cards. One VM is the largest unit DeployGPU launches: there is no self-serve multi-node cluster. For more than 8 H200s, or a fixed term at a fixed price, request a cluster quote and a person replies by email. The multi-GPU page goes deeper on how GPUs inside one node talk to each other.
Software on the machine
In the dashboard, every H200 VM boots an Ubuntu image with NVIDIA drivers and Docker installed by the provider, and you connect over SSH as the user shown on the deployment. Three app templates set up a ready app on a private HTTPS link instead: JupyterLab with PyTorch, ComfyUI and Open WebUI with Ollama. For an OpenAI-compatible server on your own H200, follow the vLLM guide.
What you pay
The price on each row is the hourly rate you are billed, including any platform fee. Launching reserves the first hour from your credit balance, and when you stop, the unused seconds of that hour go back to the balance. Credits are bought by card with a 5% + $0.30 processing fee on each top-up. The pricing page has the full breakdown.
Stopping an H200 deletes its disk, so copy results off first. The guide to stopping a GPU without losing work shows how.
Frequently asked questions
How much does an H200 cost per hour on DeployGPU?
The table on this page lists the current hourly price of every H200 VM you can launch, cheapest first, refreshed every 15 minutes. The price shown is the rate you are billed, including any platform fee.
What is the difference between H200 NVL and H200 SXM?
Both variants have the same 141 GB of HBM3e memory and 4.8 TB/s of bandwidth. SXM connects up to 8 GPUs through NVSwitch at 900 GB/s per GPU. NVL uses a bridge that joins only 2 or 4 GPUs at the same per-GPU rate.
Is the H200 faster than the H100?
The compute cores are the same Hopper generation. The difference is memory: 141 GB against 80 GB, and 4.8 TB/s of bandwidth against 3.35 TB/s, which matters most for long context and for models that do not fit on an H100.
Can I rent a single H200?
Yes, whenever a 1x offer is in stock. The GPUs column shows how many H200s come with each VM, and you pay only for the time the VM runs.
How long until I can SSH into an H200?
Most VMs reach Running in 1 to 3 minutes. On some providers SSH starts accepting connections a little after the machine reports Running, so retry for a minute before assuming something is wrong.
What happens if no H200 is available?
The table says so and links to the marketplace. Stock changes by the minute. An H100 costs less if 80 GB is enough, and a B200 has more memory when one is in stock.
Do I need a contract to rent an H200?
No. Every H200 here is on demand with no minimum term. You prepay credits, launch, and stop whenever you want.