Stop fighting your GPUs.
Launch anything, anywhere.
If you own a multi-GPU rig, you know the drill — terminal, venv,
CUDA_VISIBLE_DEVICES,
a wall of flags, hope GPU 0 is free. GPUCtrl turns the whole ritual into a button.
For ComfyUI, Blender, Ollama, Stable Diffusion, and 10 more apps.
No credit card required
Built for the apps you already use
A desktop launcher for GPU-accelerated apps
If you run ComfyUI, Stable Diffusion, Blender, or local LLMs, you already know the
ritual: spin up a terminal, source the right venv, export
CUDA_VISIBLE_DEVICES,
paste a wall of flags, hope the right GPU is free. GPUCtrl makes that a single
button — and gives you live VRAM, NVLink topology, container isolation, and
workflow automation as a bonus.
Shell history sprawl
- · Memorise flags for every app
- · Manually export GPU env vars
- · Forget which terminal owns which GPU
- · Apps step on each other's VRAM
One surface
- · Saved profiles per app + GPU combo
- · Visual flag picker with inline docs
- · Auto-routed CUDA / HIP / DirectML env
- · Live VRAM, fragmentation alerts, history
Reproducible workflows
- · Launch any setup in one click
- · Run 4 apps on 4 GPUs without conflict
- · Container-isolate experiments
- · Share profiles across your team
The polish your launch scripts will never have
Everything you’d build yourself — if you had a weekend and didn’t mind hunting down the Windows-console handle bug. Profiles, live monitoring, container passthrough, multi-app workflows. GPUCtrl did the boring parts.
Multi-vendor GPU detection
Automatically detects every NVIDIA and AMD card on your system. Surfaces VRAM, utilization, temperature, and NVLink topology in real time.
Smart GPU routing
Pin any application to specific GPUs. GPUCtrl handles
CUDA_VISIBLE_DEVICES,
HIP_VISIBLE_DEVICES, and venv
activation so apps see exactly the GPUs you assigned.
Profile management
Save unlimited launch profiles per app. Templates for ComfyUI, A1111, Fooocus, Ollama, vLLM, Blender — clone, edit, share, launch in one click.
Live VRAM & utilization
Color-coded bars, time-series charts, fragmentation detection. Know what's running on every GPU without leaving the app.
Argument helper
Visual builder for 14 app types and 300+ command-line flags. Search, hover for docs, live-preview the launch command. No more grepping through READMEs.
Container integration
Launch profiles inside Docker, LXD, or systemd-nspawn with GPU passthrough already wired up. Reproducible, isolated, painless.
From install to launched in 60 seconds
Detect
Open GPUCtrl. Every GPU on your system shows up automatically — name, VRAM, NVLink groups, the works.
Configure
Build a profile: pick the app, pick the GPUs, pick a venv. Use the argument helper to add flags without touching the docs.
Launch
One click. Live console output, process tracking, clean shutdown when you're done. Star it as a favorite for next time.
Build launch commands without the docs
300+ flags across 14 apps. Categorised, searchable, with inline tooltips and a live command preview. Click. Configure. Launch.
$ python main.pySame builder for ComfyUI, A1111, Forge, Fooocus, InvokeAI, Kohya-ss, Blender, Ollama, vLLM, LM Studio, Text Gen WebUI, OBS, DaVinci Resolve, and any custom executable.
Anyone living on the GPU
AI builders
Switch between ComfyUI, A1111, Fooocus, and InvokeAI without re-typing CLI flags. Run multiple instances on separate GPUs simultaneously.
ComfyUI · A1111 · Fooocus · Forge · InvokeAI3D artists
Launch Blender with optimal GPU + backend selection. Dedicate cards to rendering while leaving one for viewport & UI.
Blender · DaVinci ResolveLLM developers
One profile per quantisation, model, and tensor-parallel layout. Spin up Ollama, vLLM, or LM Studio with the right flags every time.
Ollama · vLLM · LM Studio · Text Gen WebUIMulti-GPU workstations
Pin renders to one GPU, training to another, inference to a third. NVLink groups, VRAM fragmentation alerts, history charts.
NVIDIA · AMD · Mixed-vendor isolationSimple, fair, one-time
Try every feature for 7 days. Buy once, own it forever.
Trial
Full access. No card required.
Pro
For builders who live on the GPU.
Payments processed securely by Lemon Squeezy. EU VAT & tax handled for you.
Quick answers
Which operating systems are supported?
Linux is fully supported on all modern distros (Ubuntu, Debian, Fedora, Arch). Windows 10/11 is in public beta — every feature is implemented but we're still hardening it across hardware. macOS is not supported — Apple Silicon doesn't expose the GPU control surface GPUCtrl needs.
What's the current version?
GPUCtrl v0.8.0. Linux build is production-ready; Windows is in beta and improving rapidly. All 14 supported app types, container integration, workflow automation, and the auth/license system are live.
Why does Windows show a SmartScreen warning when I install?
The Windows build isn't code-signed yet — while it's in beta you'll see a Microsoft Defender SmartScreen prompt on first launch. Click More info → Run anyway to proceed. We'll ship a signed release alongside v1.0.0 once the Windows build is out of beta, which removes the warning. Linux installs are unaffected.
Does the trial really include everything?
Yes. The 7-day trial unlocks every Pro feature, including container launching and workflow automation. We want you to test the entire product before paying.
How do you bill — subscription or one-time?
One-time payment for lifetime access and updates. No recurring charges, no per-seat pricing. Pay once, own it.
How many devices can I install on?
Up to 2 active devices per license. You can swap devices anytime from your dashboard — sign out of one to register another.
What about offline use?
GPUCtrl works fully offline. The desktop app keeps a 7-day grace period after the last successful license check, so a flaky internet connection won't lock you out mid-render.
Refunds?
30 days, no questions. If GPUCtrl doesn't earn its place in your workflow, email gpuctrl@gmail.com and we'll refund the order.
Stop fighting your GPUs.
Get GPUCtrl on your machine in two minutes. Free for seven days.