Running in ~30 s
Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.
Products
Developers & company
GPUs · on-demand per hour
Template · Training & fine-tuning · CUDA 12.6
Fine-tune Llama, Qwen or Mistral from one YAML — LoRA, QLoRA and multi-GPU FSDP. Axolotl fine-tunes Llama, Qwen, Mistral and friends from one YAML: LoRA, QLoRA, full-parameter, DeepSpeed and FSDP, with dataset formats handled for you. The template is what our QLoRA walkthrough is written against — checkpoints go to a volume, the trainer instance is disposable. From $0.131/hr on an interruptible RTX 4090.
Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.
You pay the GPU price only: Axolotl — Fine Tuning on a RTX 4090 is $0.262/hr on-demand, $0.131/hr interruptible, billed per second.
Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.
Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.
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 | QLoRA 8B–13B on 24 GB for a couple of dollars. | Deploy |
| Better | A100 SXM4 | 80 GB | $0.583 | $0.291 | 80 GB with NVLink for full fine-tunes and 8× FSDP. | Deploy |
| Best | H100 PCIE | 80 GB | $2.147 | $1.073 | 80 GB + FP8 — full fine-tunes in roughly half the A100 wall-clock. | Deploy |
Need more VRAM? The full catalogue lists all 76 models with live availability; the VRAM guide sizes models to cards.
Pick the template in the console deploy bar, or script it:
$ powergpu launch --gpu rtx-4090 --template axolotl-fine-tuning \
--disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Axolotl — Fine Tuning · RTX 4090 · $0.262/hr · per second
# https://i-7a41c0e2.powergpu.io:8888 (TLS)
$ powergpu stop i-7a41c0e2 # billing ends this second
| Image | axolotlai/axolotl-cloud |
|---|---|
| CUDA | CUDA 12.6 |
| Access | also builds for ARM hosts · SSH shell · JupyterLab on a mapped port |
| Category | Training & fine-tuning |
| Storage | Instance NVMe disk (sized at deploy) + optional network volumes |
| Billing | GPU price only, per second — no template fee, no setup fee |
Environment variables, ports and custom images are covered in the template docs.
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.
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.
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.
About 1.5 hours for 10k instruction pairs on Llama 3.1 8B on one RTX 4090; a 70B QLoRA on an 80 GB card is an overnight job. The guide lists exact configs and bills.
Top up in crypto, benchmark us against your current provider. Per-second billing, fixed prices ≥ 30% below market — cancel by just stopping the instance.