On-Demand VMs · This page
You own the stack.
For custom kernels, your own Docker daemon, or a GPU node in your Kubernetes cluster.
- Root on the VM via sudo
- Your OS, drivers and runtime
- SSH access · typically about 90 seconds to launch
GPU VMs with
full root access.
Run your AI workloads on 1 to 8 dedicated NVIDIA GPUs, with SSH access and control over your software.
NVIDIA H100 SXM $2.69/hr · Billed by the minute
~/your-vmACCESS EXAMPLE
$ jl ssh <instance-id>
$ sudo whoami
root
about 90 secondsTypical launch · 8-GPU VMs up to ~8 minutes
1 to 8 GPUsScale within a single instance
Pause compute. Keep your disk.Only storage is billed while paused
Measured performance
Full control. Proven GPU performance.
Six internal benchmarks on 8× NVIDIA H200 SXM. VM and container results were within 1.2% across compute, memory, communication and training.
6 / 6benchmarks within 1.2%Inspect the results ↘
Training throughput, nanochat, 8 GPUs
Container1,594,318tok/s
VM1,598,385tok/s
+0.26% observed difference · Final 10-step mean. FP8, Flash Attention 3.
NCCL all-reduce bus bandwidth
Container468.2GB/s
VM467.4GB/s
-0.17% observed difference · 1 GiB message across 8 GPUs. The NVLink and NVSwitch path, preserved.
Internal Jarvislabs benchmark, June 2026, run on separate 8× NVIDIA H200 SXM allocations. A Jarvislabs GPU container compared with a KVM VM using direct GPU passthrough. Training was nanochat with FP8 and Flash Attention 3; probe software versions differed slightly between runs. These figures describe the tested systems and workloads, not a universal guarantee.
All six benchmarks & methodology +
| Workload | Container | VM | Difference |
|---|---|---|---|
| BF16 matrix multiply, mean of 8 GPUs | 694.1 TFLOP/s | 695.9 TFLOP/s | +0.25% |
| FP16 matrix multiply, mean of 8 GPUs | 623.0 TFLOP/s | 630.3 TFLOP/s | +1.16% |
| HBM copy bandwidth, mean of 8 GPUs | 4.286 TB/s | 4.266 TB/s | -0.47% |
| NCCL all-reduce bus bandwidth, 1 GiB message | 468.2 GB/s | 467.4 GB/s | -0.17% |
| nanochat training, 1 GPU | 205,133 tok/s | 206,131 tok/s | +0.49% |
| nanochat training, 8 GPUs | 1,594,318 tok/s | 1,598,385 tok/s | +0.26% |
Differences in either direction at this scale are run-to-run variation: clocks, temperature and software builds move results by fractions of a percent. Read the table as parity, not as the VM winning four rows of six.
Direct GPU passthrough lets CUDA execute on the GPUs. NVLink and NVSwitch connect the eight H200s in the tested configuration.
Choose your compute
Start with one. Scale to eight.
Choose a GPU count to see supported configurations and the hourly price for the whole instance.
Typically ready in about 90 seconds.
| GPU | GPU memory | vCPU | System RAM | Instance / hour |
|---|---|---|---|---|
| NVIDIA H200 SXM | 141 GB | 28 | 300 GB | $3.99/hr |
| NVIDIA H100 SXM | 80 GB | 16 | 200 GB | $2.69/hr |
| NVIDIA RTX Pro 6000 Blackwell | 96 GB | 28 | 160 GB | $1.89/hr |
| NVIDIA A100 80GB | 80 GB | 16 | 112 GB | $1.49/hr |
| NVIDIA A100 40GB | 40 GB | 16 | 112 GB | $0.89/hr |
| NVIDIA A30 | 24 GB | 16 | 112 GB | $0.41/hr |
| NVIDIA L4 | 24 GB | 32 | 124 GB | $0.44/hr |
vCPU and system RAM scale with the number of GPUs in the instance.
Instance storage is $0.00014/GB/hour. A paused instance is charged storage only.
GPU availability and supported counts vary by region. Choose your region in the dashboard.
From launch to work
A familiar machine. On your terms.
GPU drivers are ready. Install your runtime, use your own images, or join an existing Kubernetes cluster.
- 01
Choose your configuration
Select a GPU, count and region in the dashboard. Most VMs launch in about 90 seconds; allow up to ~8 minutes for eight GPUs.
- 02
SSH in. Make it yours.
Take root with sudo. Install Docker, configure your drivers, or bring your own orchestration.
- 03
Pause. Resume on the same disk.
Pausing stops GPU charges. Your data and environment stay on disk, billed at $0.00014/GB/hour.
Choose your level of control
The right starting point for your work.
VMs and Templates use the same GPUs at the same hourly rates. Choose who manages the software underneath your code.
Templates
Get straight to your code.
For notebooks, fine-tuning and image generation in an environment that is already configured.
- PyTorch, ComfyUI and more, pre-built
- Managed kernel and GPU drivers
- JupyterLab, VS Code and SSH
Before you launch
A few useful answers.
The practical details, including launch times, storage charges and configuration limits.
Do I actually get root?+
Yes. SSH in and take root with sudo: install what you like, patch the kernel, load your own modules. It is a virtual machine, not a namespace on somebody else’s host.
Can I run Docker?+
Yes, your own daemon at your own version, pulling from your own registry. The NVIDIA container toolkit is yours to install or skip.
Can I join it to my Kubernetes cluster?+
Yes. The VM behaves as a GPU node: install a kubelet and join an existing cluster, or run a single-node cluster on the box. We do not operate your control plane.
Is a VM slower than a container?+
Not measurably, for the GPU paths we tested. In internal 8× H200 runs, VM and container results stayed within 1.2% of each other across tensor compute, HBM bandwidth, NCCL communication and nanochat training throughput, and five of the six results were within 0.5%. The GPUs reach the VM by direct passthrough, so CUDA runs on the card itself.
How long does a VM take to start?+
About 90 seconds to a root shell. If you want a working PyTorch environment in 1.8 seconds instead, that is what Templates are for.
Do I pay while an instance is paused?+
GPU charges stop the moment you pause. You pay storage only, at $0.00014/GB/hour, and your disk is kept.
Is the GPU shared with anyone else?+
No. Your GPUs are dedicated to your instance for as long as it runs, and on a VM you control the kernel, the drivers and the OS on top of them.
Can I get more than 8 GPUs?+
Eight is the maximum in a single VM. For workloads that need more GPUs, use a multi-node cluster across several VMs. Talk to us about a cluster configuration for your workload.
Which GPUs can I launch at 8 at a time?+
Not every GPU is offered at every count. NVIDIA H200 SXM, NVIDIA H100 SXM, NVIDIA RTX Pro 6000 Blackwell, NVIDIA L4 run at 8 per instance. The rest are capped lower, and the configurations table shows exactly which.
What does it cost to get started?+
A $10 minimum to add credit. Billing is per minute with no minimum rental period. There is no free trial.
Your next experiment starts here
Make it your machine.
Start with $10 in credit. Pay by the minute. No minimum rental period.