1) NVidia drivers had a lot of secret sauce to give high performance.
2) NVidia for machine learning still has a lot of platform lock-in (although fading gradually), and cross-compatibility doesn't help them
3) Quite often, if you've licensed something from a third-party, you can't legally open-source. Proprietary codebases sometimes get... messy.
I'm jumping ship as soon as Intel drivers are good enough. I don't trust AMD to have anything working -- too many bad experiences -- but Intel has a good track record. Arc A770 gives 16gb for <$300. That's as much as I have on my >$1000 NVidia card. I don't need maximum FLOPS. So long as deep learning models run, and 3d apps are accelerated, I'm happy.
1) NVidia drivers had a lot of secret sauce to give high performance.
2) NVidia for machine learning still has a lot of platform lock-in (although fading gradually), and cross-compatibility doesn't help them
3) Quite often, if you've licensed something from a third-party, you can't legally open-source. Proprietary codebases sometimes get... messy.
I'm jumping ship as soon as Intel drivers are good enough. I don't trust AMD to have anything working -- too many bad experiences -- but Intel has a good track record. Arc A770 gives 16gb for <$300. That's as much as I have on my >$1000 NVidia card. I don't need maximum FLOPS. So long as deep learning models run, and 3d apps are accelerated, I'm happy.