Running in ~30 s
Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.
Template · Base & frameworks · CUDA 12.8
Run PyTorch on a cloud GPU, in 30 seconds
The training and research default: PyTorch 2.6, cuDNN, JupyterLab and SSH out of the box. PyTorch 2.6 with CUDA 12.8, cuDNN, torchvision and torchaudio, JupyterLab on a mapped port and SSH — the environment most research code assumes. Datasets and checkpoints live on a volume, so the instance stays disposable and per-second billing does the rest. From $0.131/hr on an interruptible RTX 4090.
Template is free
You pay the GPU price only: PyTorch on a RTX 4090 is $0.262/hr on-demand, $0.131/hr interruptible, billed per second.
Volumes for state
Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.
Private by default
Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.
Best GPUs for PyTorch
Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.
| Tier | GPU | VRAM | On-demand | Interruptible | Why this card | Action |
|---|---|---|---|---|---|---|
| Good | RTX 4090 | 24 GB | $0.262 | $0.131 | 24 GB and Ada tensor cores: the best price for single-GPU experiments and LoRA runs. | Deploy |
| Better | A100 SXM4 | 80 GB | $0.583 | $0.291 | 80 GB HBM2e with NVLink for multi-GPU training up to 8× on one machine. | Deploy |
| Best | H100 SXM | 80 GB | $1.587 | $0.793 | FP8 and 3.35 TB/s for full fine-tunes and pre-training at the lowest cost per step. | Deploy |
Need more VRAM? The full catalogue lists all 76 models with live availability; the VRAM guide sizes models to cards.
Deploy PyTorch from the console, CLI or API
Pick the template in the console deploy bar, or script it:
- Console — filter by GPU, choose PyTorch in the template picker, set disk and env, deploy.
- CLI — pip install powergpu, then the command on the right. CLI reference.
- API — POST /v1/instances with "template": "pytorch". REST reference.
- Own image — any OCI reference works too; we inject the NVIDIA runtime. Template docs.
$ powergpu launch --gpu rtx-4090 --template pytorch \
--disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# PyTorch · RTX 4090 · $0.262/hr · per second
# https://i-7a41c0e2.powergpu.io:8888 (TLS)
$ powergpu stop i-7a41c0e2 # billing ends this secondWhat is inside the PyTorch template
| Image | powergpu/pytorch:2.6-cuda12.8 |
|---|---|
| CUDA | CUDA 12.8 |
| Access | also builds for ARM hosts · SSH shell · JupyterLab on a mapped port |
| Category | Base & frameworks |
| Storage | Instance NVMe disk (sized at deploy) + optional network volumes |
| Billing | GPU price only, per second — no template fee, no setup fee |
Other base & frameworks templates
NVIDIA CUDA on a cloud GPU — The bare CUDA base image — drivers, toolkit, SSH and Jupyter. Build your own stack on top.
TensorFlow CUDA on a cloud GPU — TensorFlow with GPU support, Keras and TensorBoard on a mapped port.- PyTorch NGC on a cloud GPU — NVIDIA's NGC-optimised PyTorch build — tuned kernels for the datacenter cards.
- NVIDIA RAPIDS on a cloud GPU — GPU-accelerated data science — cuDF, cuML, cuGraph in a notebook.
All-in-One App Studio on a cloud GPU — A launcher bundling the most-used AI apps behind one desktop — pick and run.- All 37 templates
PyTorch on a cloud GPU: FAQ
Environment variables, ports and custom images are covered in the template docs.
How much does it cost to run PyTorch on a cloud GPU?
Only the GPU price — the template is free. From $0.131/hr on an interruptible RTX 4090, $0.262/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.
How long does PyTorch take to start?
About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.
Can I keep my models and outputs between sessions?
Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.
Does it support torch.compile and FSDP?
Yes — the image ships the matching Triton build for torch.compile, and FSDP/DDP work out of the box on multi-GPU machines (NCCL over NVLink where the hardware has it).