Comment by shmerl | Hacker News Reader

Comment by shmerl | Hacker News Reader

Parent Full thread

> It doesn't at all.

If you didn't follow this, it doesn't to you. It does, to those who pay attention and constantly see people switching to AMD and commenting about their reasons. And these reasons are quite obvious really. Nvidia did nothing to improve their integration with Linux (the meain issue is their unwillingness to upstream their kernel driver), while AMD did a lot to improve their drivers. It's only natural to expect Nvidia Linux gaming usage to continuously drop as a result of the above.

> Source?

https://www.techpowerup.com/215663/lack-of-async-compute-on-...

AMD hardware is known to be better for GPGPU for a long time already.

> There doesn't have to be a problem with the hardware, adoption will be zero as long as alternative OpenCL/Vulkan compute based libraries don't exist.

If developers are using some closed source libraries, they are at mercy of those vendors who might be in the pocket of Nvidia. And no one stops open source libraries from not being CUDA only and from using Vulkan and OpenCL for compute needs. Given Vulkan is still rather new, it can take time for libraries to pick it up though for compute scenarios. But it will happen either way. Overall, I don't see good future for CUDA, if it will remain Nvidia only. Same as with gaming, open solutions will catch up, and Nvidia lock-in will crumble.

> If you didn't follow this, it doesn't to you. It does, to those who pay attention and constantly see people switching to AMD and commenting about their reasons.

I'm talking about this purely from a data point of view. I'm referring to the graph, and it has too many holes to be used as a sweeping generalization about the market and Linux users as a whole.

To be clear, I'm not arguing for or against NVIDIA/AMD. I'm just trying to point out the issues with using that graph as definitive evidence of anything.

> Nvidia did nothing to improve their integration with Linux, while AMD did a lot to improve their drivers.

And I applaud AMD for that. They've come a long way. However, (and this is personal opinion based off of experience), I don't see Wayland as being as a particularly massive reason. Wayland is what this thread is about anyways. There are still plenty of people using XOrg that are happy with that and don't run into any issues, myself included. That's not to say I don't support Wayland development and I hope NVIDIA never adds support for it in the future, but it's such a 'young' piece of tech that still requires some more maturing.

> AMD hardware is known to be better for GPGPU for a long time already.

Raw power is only a piece the pie. Some people take issue with that statement, but it's reality.

To the point in your link, Pascal fixes the async problem present in Maxwell.

https://www.anandtech.com/show/10325/the-nvidia-geforce-gtx-...

> And no one stops open source libraries from not being CUDA only and from using Vulkan and OpenCL for compute needs.

And yet they primarily are CUDA only. It's a conscious and intentional decision by the developers of those libraries to start with CUDA and stick to it. And the community not adding OpenCL support shows that AMD hardware isn't even in the space for that application.

I think that we could go back and forth with this, providing counter examples for each point the other brings up. I merely wanted to offer up an enterprise oriented view of the picture, when most people are looking at individual usage.

> I'm referring to the graph

Graph makes sense if you analyze the context. It's as I said above - increasing number of users are switching to AMD. Just check any thread in Linux gaming forums on the topic of "what should be my next GPU".

> I don't see Wayland as being as a particularly massive reason. <...> don't run into any issues,

It's one of the reasons. There are many issues with Nvidia. Abysmal integration due to lack of upstream driver such as broken vsync, no standard hardware sensors, no PRIME support and so and so forth are all well known Nvidia problems. To address them, Nvidia should either open up their driver, or to support Nouveau to begin with. So far they showed no interest in either of those cases.

> It's a conscious and intentional decision by the developers of those libraries to start with CUDA and stick to it.

That's too bad. Avoid libraries which proliferate lock-in. If their developers don't know any better, find those who do and support their efforts instead. It should be in the interest of actual developers who use said libraries, not to be locked into single hardware vendor.

> AMD hardware is known to be better for GPGPU for a long time already.

Maybe, maybe not, but the async compute argument you are making only applies to graphics applications that use compute shaders on maxwell - it doesn't apply to pure gpgpu without graphics.

Reply on news.ycombinator.com