Jensen Huang says even free AI chips from his competitors can't beat Nvidia's GPUs
Jensen Huang says even free AI chips from his competitors can't beat Nvidia's GPUs
Jensen says free 'isn't cheap enough' to stand a chance against the green team in data center AI.
You're viewing a single thread.
I'm free to choose any laptop I want for work. This means, that for me, the GPU and other processors are free. It turns out that I still avoid Nvidia like the plague. I don't care if it is free, if the drivers are horrible.
You should try working with their CUDA tool chains. Good. Lord.
And AMD or Intel are better? Everyone complains about the drivers.
For Linux it is a huge difference. AMD and Intel have great open source drivers, while Nvidia have binary drivers with a lot of issues.
I find this strange, because I had nothing but trouble getting my R9 390 working with any Linux distro, but my RTX 3060 hasn't given me a single issue on like 6 different distros.
Are you talking about a laptop or a desktop? If desktop, using offload or something like that?
Desktop, both Intel and AMD builds. And no, I just had multiple SSDs that I played with distros on.
Nvidia works flawlessly most of the time on desktops. The suckyness of their drivers shows its ugly head mostly on laptops.
And it's bad, crash your computer bad some times.
I guess I've been lucky with my laptop, although it is fairly old at this point.
Three laptops ago it mostly worked with FOSS solutions around the driver. Nvidia mostly killed them, and intreduced their own unstable solution. As far as I care, if they won't fix it before my hardware dies, the next laptop will have a GPU by another manufacture.
And for AI at home? Since this is a story about AI DataCenters
I want to get an AMD but the integration of Nvidia GPUs for processing ML/AI stuff is much higher. So if I want to mess with running AI at home I only have 1 choice.
I hope AMD release something that competes on that front, and can still play games on the weekend, but currently, he is right there is no competition
I run AI stuff just fine on my AMD GPU using HIP. At least LLMs and Stable Diffusion work perfectly fine but that's the only things I've tested.
All the benchmarks put the equivalent Nvidia cards almost 2x more in Stable Diffusion https://www.tomshardware.com/pc-components/gpus/stable-diffusion-benchmarks
I hope that is an old benchmark and times have changed, but I can't find anything like it that is more recent.
For a bit of my soul I get a lot more ai power.
AI works on AMD but the speed doesn't seem to be anywhere near Nvidia.
I still won't consider an AMD GPU because all 3 I've had throughout my life have had horrible driver experiences. Even the FirePro I had at work at one point required a special driver build that AMD eventually gave me to work even half decent. Never had any major issues with NVidia drivers.
I've had the opposite experience here on GNU/Linux. The AMD drivers are phenomenal.