The B200 is DeployGPU's largest-memory GPU: 180 GB of HBM3e per GPU on NVIDIA's Blackwell architecture, FP4 (NVFP4) Tensor Core support that Hopper cards do not have, and NVLink at 1.8 TB/s between GPUs. It is also the scarcest card on the platform, so this page is written to be useful even when the table below is empty.
- Offers now
- 1
- From
- $8.03/GPU-hr
- GPUs per VM
- 8x
- Providers
- 1
B200 offers you can launch now
The table shows any B200 VM in stock right now, cheapest first, refreshed every 15 minutes.
| GPU | GPUs | Provider | Region | vCPU / RAM / Disk | Interconnect | Per GPU-hr | Per hour | Launch |
|---|---|---|---|---|---|---|---|---|
| B200 SXM 192 GB | 8x | Verda | Helsinki, FI | 240 / 1360 GB / - | NVLink (SXM) | $8.03 | $64.23 | Launch |
For B200 prices across the whole market, compare all providers on GpuPerHour. For a reference price over time, see the B200 rental index (GPUIndexes).
180 GB, not 192 GB
Some listings advertise the B200 at 192 GB. NVIDIA's own DGX B200 datasheet lists 1,440 GB of total GPU memory across the system's 8 GPUs, which works out to 180 GB per GPU, and that is the figure this page uses.
| Spec | B200 |
|---|---|
| Memory | 180 GB HBM3e per GPU |
| Memory bandwidth | About 8 TB/s per GPU |
| GPU-to-GPU link | Fifth-generation NVLink, 1.8 TB/s per GPU, NVSwitch mesh across up to 8 GPUs |
| Low precision | FP4 (NVFP4) and FP8/FP6 Tensor Core support |
| Architecture | NVIDIA Blackwell |
Specifications from NVIDIA's DGX B200 datasheet. Across a full 8-GPU HGX B200 board, NVIDIA lists 72 PFLOPS of dense FP4 Tensor Core throughput and 144 PFLOPS with structured sparsity, a lower-precision format than the FP8 that tops out on Hopper cards like the H100 and H200.
What fits in 180 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.
| Model | Format | Weights | On one B200 180 GB |
|---|---|---|---|
| Llama 3.3 70B | BF16 | 141.1 GB | Fits on one card, about 24 GB left. The same weights need 2 H200s, or 4 H100s once the KV cache has room |
| Qwen3-Coder-Next | BF16 | 159.4 GB | Fits on one card, about 6 GB left, tight for long context |
| gpt-oss-120b | MXFP4 | 65.2 GB | Fits with about 100 GB left for the KV cache |
| DeepSeek-V4-Flash-0731 | FP4/FP8 | 166.9 GB | Needs 2x B200, about 164 GB left |
| GLM-5.3-Flash | FP8 | 328.3 GB | Needs 2x B200, about 3 GB left, no room to spare |
The standout row is Llama 3.3 70B: 180 GB is the first size on DeployGPU where a 70B model in full BF16 fits on a single card with room left over, instead of needing 2 H200s or 4 H100s.
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
B200 VMs come in fixed sizes of 1, 2, 4 or 8 GPUs when stock allows, and the GPUs column shows what is available right now. One VM is the largest unit DeployGPU launches: there is no self-serve multi-node cluster and no InfiniBand. For more than 8 B200s, a fixed term, or reserved capacity while open stock stays thin, request a cluster quote and a person replies by email. The multi-GPU page covers how GPUs inside one node talk to each other.
Software on the machine
In the dashboard, a B200 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 B200, follow the vLLM guide.
Blackwell cards also need newer software than Hopper or Ampere. The images the dashboard offers for these offers carry a driver recent enough for Blackwell, and a launch with an image whose driver is too old is rejected before anything is charged. Your own stack has to keep up too: PyTorch added Blackwell support in version 2.7, with builds for CUDA 12.8 (release notes), so wheels built for older CUDA versions do not run here. Install a current build with:
pip install -U torch --index-url https://download.pytorch.org/whl/cu128
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 a B200 deletes its disk, so copy results off first. The guide to stopping a GPU without losing work shows how.
Frequently asked questions
Why does the B200 table sometimes show no offers?
B200 is the newest card on DeployGPU and stock is limited. Check back, or launch an H200 in the meantime, which has the next-largest memory at 141 GB.
Is the B200 180 GB or 192 GB?
NVIDIA's DGX B200 datasheet lists 1,440 GB of GPU memory across an 8-GPU system, which is 180 GB per GPU. This page uses that figure rather than the 192 GB some listings quote.
What does FP4 support on the B200 change?
FP4, which NVIDIA calls NVFP4, is a lower precision format than the FP8 that Hopper cards support. NVIDIA lists 72 PFLOPS of dense FP4 Tensor Core throughput across a full 8-GPU HGX B200 board.
Can I rent a single B200?
Yes, when a 1x offer is in stock. The GPUs column shows how many B200s come with each VM, and you pay only for the time the VM runs.
Should I wait for a B200 or launch an H200 now?
If the job fits in 141 GB, an H200 usually has more offers in stock. If it needs the extra headroom, such as a 70B model in full BF16 on one card, wait for a B200 or request a cluster quote to ask about reserved capacity.
Do I need a contract to rent a B200?
No. Every B200 here is on demand with no minimum term. Reserved capacity or a fixed term goes through the cluster quote form instead.
How long until I can SSH into a B200?
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.