3 ms·
ROCM is close to being perfected and I think in few years and some investment you can use it directly to run 99% software written in CUDA with similar performan
by YetAnotherNick 2y ago
ROCM is close to being perfected and I think in few years and some investment you can use it directly to run 99% software written in CUDA with similar performance.
- klelatti 2y agoSince 2007 Nvidia has built with CUDA and compatible hardware - a large and growing ecosystem - a massive installed base of backwards compatible hardware - the ability to massively scale delivery of new systems - and lots lots more They now have scale that enables them to continue to invest at a level that no competition can do. None of these are easily reproduced. As per SemiAnalysis AMD in late 2024 can’t get essential software working reliably out of the box. It’s easy to say AMD ‘is close to perfecting ROCM’ the reality of competing with Nvidia is much harder.
- YetAnotherNick 2y agoIt's hard but not $3T hard. AMD has consistently underinvested in their software. If they invest double digit million to make installation and usage of drivers and ROCM smoother I highly doubt they can't acheive it. There are open source projects volunteer run projects[1] that are better than official AMD implementation in many ways. [1]: https://github.com/CHIP-SPV/chipStar/ https://github.com/CHIP-SPV/chipStar/
- klelatti 2y agoIf it's so cheap and easy then why haven't they done it already? ChatGPT was announced Nov 22. The opportunity has been clear for two years and still essential software breaks.
- YetAnotherNick 2y agoBecause they have to rethink a lot organizationaly to put more effort into software. If installing their drivers is so flaky you can't argue they really are trying or putting their best folks in this. Single click installation on top 10 linux distro is something even one (top in the industry)person could achieve.