What NEURAX knows about this computer
Detected, not chosen from a list.
NEURAX reads the machine it is installed on when the studio opens, and every figure is computed against what it finds. The toolbar names it, and Target shows it in full.
| What it reads | Why it matters |
|---|---|
| Processor | Model, physical cores, threads, and the instruction sets it supports — avx2 and fma are most of the difference between a fast kernel and a slow one. |
| Memory | What is genuinely available, not what is unused. Most of what Linux reports as used is reclaimable cache, and sizing against the free figure refuses runs that would have been fine. |
| Accelerators | Every card, with the memory each has free right now. An integrated processor is reported as sharing system memory rather than as a card with nothing left. |
| Disk | Free space where a run would write its checkpoints. |
| Runtimes | Whether Python and PyTorch are there, whether a driver answered, and which precisions the runtime will actually accept. |
Measuring#
The specification database covers datacenter accelerators. Most machines are not one, so NEURAX measures instead — a matrix multiply, a memory sweep, then three small training steps that time PyTorch's own cost; a few seconds, remembered afterwards. Press Measure this machine in Target.
Page 12 of 33. This guide is in English for now. It is the same guide that is built into the NEURAX desktop app (the ? button, or F1). Found a mistake? Tell us.