Same answers, smaller card
The memory moves off the GPU into RAM, so a job too big for the card still runs — and the answers come out identical. Only the memory moves; the intelligence stays put.
Nadi
Run big AI jobs on a graphics card too small for them. Nadi moves the model's working memory into your computer's normal memory, so the job fits and the answers stay the same.
Big jobs need big GPUs. Nadi keeps your card and moves the memory instead: the answers stay identical, and you choose when to trade speed for room.
| What Nadi does | What it means for you |
|---|---|
| Fits a big job on a smaller GPU | Run long-context work the card couldn't hold before |
| Keeps the model's answers | Identical to the original — nothing gets dumber |
| Costs some speed, on your terms | Slower only while the memory sits in RAM; a switch you control |
Early results on real open-source models: the win is memory, the cost is speed, on your terms.
How much GPU memory the offload saves, and how fast it runs while it does.
When it becomes real
Nothing here is a product. No download, price, or date — a release exists only once there's a build you can run.
The usual way to run a bigger job is a bigger GPU. Nadi keeps your card and moves the memory instead.