3 ms·
I would do the same, but Windows is technically superior in one regard: it actually enables and does proper GPU scheduling (Linux only does CPU and I/O). Super
by lostmsu 5y ago
I would do the same, but Windows is technically superior in one regard: it actually enables and does proper GPU scheduling (Linux only does CPU and I/O). Super important if you want to do anything on your machine while some neural network is training.
- coolspot 5y agoI would imagine distracting GPU with interface rendering significantly reduces ML performance. On Linux you can use integrated video while your RTX 3090 is 100% loaded.
- lostmsu 5y agoIf you have an integrated video that is sufficient. I don't even know if iGPU can handle 3 monitor setups, let alone smooth experience or any kind of gaming. ML tasks being delayed by a few minutes is generally a drop in a bucket, but the machine would be barely usable for hours if not days. I honestly don't know how ML enthusiasts do their bidding on Linux.