Chorvix

When Your Muse Has No GPU: Two Compute Workarounds Worth Knowing

Not every Muse gets a machine with a graphics card in it — ours runs on a small shared box with no GPU at all. When the heavy work still has to happen, there are two honest ways to borrow compute: a free cloud GPU opened in a phone’s browser, and a spare Android phone lending its GPU to your PC. What each route is, when it helps, and exactly where it stops helping.

Here is the short answer first. If your Muse runs on a machine with no graphics card — a small shared box, an old laptop, a cheap server — you do not have to buy hardware before the heavy work can move. There are two honest ways to borrow compute.

The first is a free cloud GPU: services like Google Colab, opened in the browser on an Android phone, will run your notebooks, AI scripts, and media processing on Google's hardware instead of yours. The second is the opposite direction: a spare Android phone lending its GPU to your PC, running llama.cpp under Termux over a USB/ADB connection, so the phone carries a few layers of a local AI model your PC cannot quite fit.

Neither one is magic, and this article will tell you exactly where each one stops helping. Chorvix is the workspace this house builds for people working with Muse, and our own Muse works on a box with no GPU in it at all — so these are not tips collected from the internet. They are the doors we reach for ourselves, written down where other Muse users can find them.

Workaround one: a free cloud GPU, opened from an Android phone

Google Colab is a notebook service: you write or upload a notebook — code in cells, run one cell at a time — and it executes on a machine in Google's data center, not on the device in your hand. The free tier includes access to a GPU, the graphics processor that AI work and media processing are hungry for and that a GPU-less box simply does not have.

The part that surprises people is the door in. You do not need a computer to reach it. Colab runs in a mobile browser, so an Android phone is a perfectly good window: open Colab in the phone's browser, sign in, start a notebook, set the runtime to use the GPU, and run your work. The phone stays cool and its battery stays out of it — the lifting happens on Google's hardware, and your phone is just the screen and keyboard.

That makes this the right reach for bursts of heavy work: running an AI script that would crawl on a shared CPU, testing a model, or pushing media processing through a GPU for an afternoon. Work that would take all day — or never finish — on a small box can finish in a sitting, because for that sitting you are borrowing a real GPU.

Now the honest limits, because they decide how you should use it:

Used inside those lines, it is genuinely useful — it is the difference between "this render test runs tonight" and "this render test runs never" for a Muse working without a card.

Workaround two: an Android phone's GPU, lent to a PC

The second route never leaves your desk. Connect an Android phone to a PC with a USB cable, and the phone can take part of a local AI model's work off the PC's hands.

The pieces are these. Termux is a terminal app that gives the phone a real Linux-style command line. Inside Termux you can build and run llama.cpp, the widely used open-source tool for running AI models locally. Connect the phone to the PC over USB/ADB, and the PC can offload small model layers — slices of the model — to the phone, spreading one model across two devices.

This helps in one specific situation: your PC's VRAM and RAM are both maxed out, and a local model almost fits. Almost is the word that matters. A model that misses fitting by a little can miss entirely — it simply refuses to run. A spare phone carrying the overflow layers can be the difference between a model that runs and a model that doesn't.

And here the limits are not fine print; they are the headline:

Set up with those expectations, it is a real workaround: hardware most people already have in a drawer, pressed into service instead of a purchase.

Which one do you reach for?

The two routes solve different problems, so the choice is quick once you name yours:

That is the whole of it. A Muse with no GPU is a real constraint, and it is also a solvable one for a surprising amount of work — a borrowed GPU through a phone's browser for the bursts, and a borrowed phone for the model that almost fits. We keep this note in the Chorvix journal because every Muse user who hits this wall deserves to find the doors already marked.

Keep reading

New to Chorvix? What Chorvix is explains the workspace itself, and the journal's other pieces are below.

More from the article series:

All articles