Your AI workstation in the cloud
A Linux desktop with a dedicated GPU — in your browser or a native app on Mac, Windows or Linux. Your files and environment stay put. Boost to a bigger GPU in one click.
From $0.35/hr · Auto-pause when idle · No commitment
The problem
GPU rental wasn't built for daily work
Rented pods are disposable. Your work isn't. The same evening, two ways.
sent 3.2 GB · done
$ python generate.py --batch 24
done · 24 images · 214 s
Terminating instance and cleaning up...
Pods tear down when you stop. Your models, nodes and files go with them. The next day, you start over from scratch.
loras/
outputs/
Hinode auto-pauses instead. Close the lid — files, apps and processes wait intact. Open it and continue where you left off.
Boost
When the model doesn't fit
The same job on the same 24 GB card — with and without a way out.
loading flux-dev · 23.8 GB
CUDA out of memory — job killed
Fixed VRAM is a wall. The model needs more than the card has, and the job dies. Creating a new pod means starting over from scratch.
loading flux-dev · 23.8 GB
✓ loaded · 23.8 / 96 GB VRAM
Boost raises the ceiling. Pick the card the job needs, pay for the minutes you spend on it, and drop back after.
Features
One workstation, always ready
Everything an AI creator needs in a single environment that follows you across devices.
Persistent state + auto-pause
Your workstation hibernates after 15 minutes idle and wakes with everything intact. Paused time costs storage only, not compute.
Boost
Pick 24, 48 or 96 GB of GPU memory and the desktop moves onto that card in about a minute. Billed by the minute; press it again to come back.
You pay for what you use
No plan and no commitment. The desktop costs $0.35 an hour while it runs, a GPU from $1.65 an hour while you boost, storage $0.07 per GB a month, traffic $0.10 per GB. Billed at the end of the month.
H.265 60fps streaming
Every resolution up to 4K60 over HEVC is included. You pay for the traffic the stream sends: about $0.08 an hour on a still desktop, $0.36 on ordinary work.
Set it up once
Hinode Apps installs PyTorch, ComfyUI or VS Code in one click, and updates them the same way. They land in your home directory, so they are still there after a pause.
Browser or native app
Run it in your browser, or get the native app for Mac, Windows or Linux — the same live session either way.
No lock-in
Your tools, your data
A full Linux desktop where you install the tools you already use, and they stay. Import your data from any source and export it to any destination. No vendor lock-in, no closed ecosystems.
Pricing
A cheap machine, and a big GPU when you need one
No plan, no commitment. The desktop costs cents an hour; you pay for a GPU only in the minutes you use it.
Your workstation
A persistent Linux desktop with your files, your tools and your environment. Billed by the minute while it is on. While it sleeps you pay for the disk and nothing else.
- Ready in about forty seconds
- Auto-pause stops the meter when you step away
- Storage billed separately, $24.32/mo for 128 GB
Boost
Press Boost, pick a GPU, pay by the minute
The desktop moves onto the card you chose and back again when you are done. Nothing to cancel.
24 GB of VRAM
NVIDIA L4
- 32, 64, 128 or 256 GB of system RAM
- Image generation, light fine-tuning, most ComfyUI work
- Billed by the minute, 15 minutes minimum
48 GB of VRAM
NVIDIA L40S
- 32, 64, 128 or 256 GB of system RAM
- Video models, larger LoRAs, 30B-class inference
- Billed by the minute, 15 minutes minimum
96 GB of VRAM
RTX PRO 6000 Blackwell
- 64, 128 or 256 GB of system RAM
- 70B-class models, long video, anything that will not fit elsewhere
- Billed by the minute, 15 minutes minimum
What a month actually costs
Two numbers decide it: the hours you work, and how many of those need a GPU. The rest of the time you are paying for a desktop.
≈ 87 active hours a month
≈ 26 hours on a 48 GB GPU
Plus storage, and data transfer at $0.10/GB. Ordinary work is cents an hour.
How we compare
The same GPU. A full desktop.
Boost to 48 GB and you are on a dedicated L40S — the same card RunPod rents — with a full desktop that stays put instead of a terminal you rebuild.
| RunPod | Vast | Lambda | HF Pro | Hinode Pro | |
|---|---|---|---|---|---|
| GPU | L40S · 48 GB | RTX A6000 · 48 GB | A100 · 40 GB | ZeroGPU · shared | L40S · 48 GB |
| GPU access | Dedicated | Dedicated | Dedicated | Quota + queue | Dedicated |
| Interface | SSH / Jupyter | SSH / Jupyter | SSH / Jupyter | Spaces (web) | SSH, Native desktop |
| Session | Lost on stop | Tied to one host | Lost on terminate | Ephemeral | Persistent + auto-pause |
| Devices | Browser | Browser | Browser | Browser | Browser + native apps |
How it works
From signup to GPU in minutes
- 01
Sign up
Create an account in seconds. No time-limited trial.
- 02
Launch your workstation
It boots to an Ubuntu desktop in about a minute, GPU ready and nothing in your way.
- 03
Connect anywhere
Open the desktop in your browser, or in the native app for Mac, Windows or Linux.
- 04
Boost when you need it
Jump to a bigger GPU for a heavy run, then drop back automatically.
FAQ
Questions, answered
A cloud Linux desktop with a dedicated GPU, in your browser or a native app on Mac, Windows or Linux. Your environment persists between sessions, so you never set it up twice.
A full Ubuntu desktop, and whatever you add. Hinode Apps, on the dock, installs PyTorch, ComfyUI, Ollama, VS Code or Chrome in one click and updates them the same way. The terminal works too: apt install and pip install do what you expect. It all lands in your home directory, which survives a pause.
By the minute, charged to your card at the end of the month. The desktop is $0.35 an hour while it runs and a GPU from $1.65 an hour while you boost. Auto-pause stops the compute meter when you step away. Storage is billed separately, at $0.07 per GB a month for as long as the volume exists, and outbound traffic at $0.10 per GB.
Pick how much GPU memory the job needs — 24, 48 or 96 GB — and your desktop moves onto that card, files and all. Billed by the minute, fifteen minutes minimum. Press it again and you are back on the base machine.
Only for storage. Your workstation pauses after 15 minutes of inactivity and the compute meter stops, while your files keep their disk at $0.07 per GB a month. Waking takes about a minute and brings back everything you installed — a pause shuts the machine down, so anything that was running does not survive it.
Any supported browser, plus native apps for Mac, Windows and Linux. The same live session follows you across all of them.
Chrome, Edge, and other recent Chromium browsers. The client needs WebTransport and WebCodecs, which Safari and Firefox don't fully support yet.
Yes. Every workstation answers on SSH at a name that never changes, so `ssh mybox` works from any terminal — same GPU, same files, no streaming involved. Forward a port to run Jupyter or ComfyUI in your own browser. Handy when you're far from the region your workstation runs in, since a terminal doesn't care about latency the way a live desktop does.
Active workstations run on encrypted volumes; paused ones are snapshotted to object storage. Export your data anytime.