Mesa Continues With More Optimizations For Workstation OpenGL Performance

[bridgman]

Is there any movement by AMD in the direction of making radeonsi work on Windows and then throwing out the closed OpenGL one? It will also make Windows gamers more happy about OpenGL situation there.

No specific plans but it has been discussed as a possibility. Getting workstation functionality and performance in place for radeonsi would be a pre-requisite anyways.

I don't think anyone expected Minecraft to stay on OpenGL and optimized only for NVidia OpenGL as long as it has. It seems strange that emulators are being ported to newer APIs before Minecraft - I would have at least expected a DirectX port of Minecraft by now.

[Guest]

AMD has a LOT more jingle to throw around these days and Lisa Sue seems to be hell bent on fixing the back log of problems AMD has had rather that just doing stock buy backs like so many companies these days. The big problem that money hasn't been able to solve so far is fab capacity.

I doubt that, they haven't bothered implementing fundamental things like current, voltage, power monitoring for Ryzen consumer CPUs. The only place where they've done great work recently is for Zen 3 APUs, most likely due to Chromebooks. They don't care that much about the Linux desktop, it's definitely nowhere as good as Intel's support.

[stiiixy]

It's more like 2 GL drivers than 3 - one open-source code base in Mesa and a separate closed-source driver that runs on Windows and on Linux as part of AMDGPU-PRO.

Considering eho owns MineCraft, a bizzarre situation indeed. But, I guess OpenGL is more portable across Android, iPhone etc where you find MineCraft also.

[niner]

I doubt that, they haven't bothered implementing fundamental things like current, voltage, power monitoring for Ryzen consumer CPUs.

"Fundamental" things that interest only a select few. Everyone else just uses those CPUs and is happy with the massive performance (and at least in my case exactly 0 issues from the first boot)

[zboszor]

Getting workstation functionality and performance in place for radeonsi would be a pre-requisite anyways.

Other pre-requisite functionality is implementing the missing Win32 extensions in Mesa: GL_EXT_memory_object_win32, GL_EXT_semaphore_win32.
On the Vulkan side: VK_KHR_external_fence_win32, VK_KHR_external_memory_win32 and VK_KHR_external_semaphore_win32, VK_KHR_win32_keyed_mutex and VK_KHR_win32_surface

[Guest]

"Fundamental" things that interest only a select few. Everyone else just uses those CPUs and is happy with the massive performance (and at least in my case exactly 0 issues from the first boot)

There's a difference between dumb consumers vs power users/devs. Dumb consumers don't know or care what CPU they're using, and the ones you're talking about are dumb consumers, who don't care about Intel vs AMD, x86 vs ARM.

[niner]

There's a difference between dumb consumers vs power users/devs. Dumb consumers don't know or care what CPU they're using, and the ones you're talking about are dumb consumers, who don't care about Intel vs AMD, x86 vs ARM.

So by your definition I'm a dumb consumer, even though I have almost 30 years of experience as a software developer and bought my 5950X (and the 1800X before that) explicitly to speed up compilation and test runs. I have become measurably more productive thanks to the CPU that I very much care about.

So what exactly does constitute a developer or power user in your eyes? I've known a lot of people who fiddled around for hours with BIOS settings, cooling and whatnot to squeeze out the last few benchmark points. Heck, I used to do that as well. But that was because the computer was an interest by itself. Overclocking was simply fun. Nowadays I care more about getting my actual work done.

[MadeUpName]

I doubt that, they haven't bothered implementing fundamental things like current, voltage, power monitoring for Ryzen consumer CPUs. The only place where they've done great work recently is for Zen 3 APUs, most likely due to Chromebooks. They don't care that much about the Linux desktop, it's definitely nowhere as good as Intel's support.

Well the big moves on graphics cards in recent years have been in compute. You haven't been able to complain about Intel's because they didn't have any. Now that they have sort of released some graphics cards that have some they suck. But it's early days so we need to give them a break. You don't just snap your fingers and stuff happens like you seem to think.

[baryluk]

I did some runs of viewperf11 a few days ago, and indeed, performance in some tests increased A LOT. For example all tests in snx-01 suite, improved between 5 and 70 times. (8 times on average). In fact on my GPU, these test suite was really executing poorly (sometimes ~1.3fps) and not smooth. Now it just works beautiful (30fps minimum, some 105fps). All the other tests also improved by about 10%, which is a nice bump (however, that could be attributed to slightly different compiler options used in my tests).

[qarium]

Would this mean someone at AMD's top brass had a stroke and was replaced with someone with more the 3 working neurons that realized it was idiotic to have 3 drivers codebases instead of optimizing the living bejesus out of one?

yes you are fully right... this was long time overdue
but better late than never.

but i think now it will not only replace the closed source workstation linux driver but also the closed source windows openGL driver.