AMDGPU vs. Radeon DRM On Linux 4.13 For AMD GCN 1.0/1.1 GPUs

[Holograph]

Are there any serious Windows games that would benefit from a faster OpenGL these days ? My impression was that all the OpenGL users of old had hopped across to Vulkan, so the closest thing to a "legacy API" these days was DX11.

It's really frustrating how fast you hardware/driver guys want to get into terms like "legacy"

It feels like you'd stop supporting any games running on your GPUs within 30 days if you could.

You realize that many gamers play games that are a few years old, right?

I don't want to argue this on a technicality as I know that the very most recent games will almost always have priority of support but do you seriously need to ask if any of us still play OpenGL games? Do you really need an answer to that? If so, YES OF COURSE WE DO. I'm so frustrated that the question even has to be asked. (If you want to argue the point of your question was if we needed it to be "faster" then unless you demonstrate that no OpenGL game EVER has any slowdowns on any computer which meets said game's minimum requirements, then yes, we literally always need "faster")

I play games like 7 Days to Die which is similar to Minecraft and uses the Unity engine (which does have an OpenGL driver)... It loads quite a large area and stresses my Haswell + 290x somewhat still. Planning a CPU upgrade next (whether that is Ryzen or not depends on when and how you guys address the GCC segfault issue that has been mostly ignored so far from what I've seen). 7DtD first came out a few years ago but still gets significant updates - it got a big one within the past few weeks.

It just boggles my mind that you would want to stop caring about OpenGL already.

I won't forget this post next time I buy a GPU, because I don't care if that's 5 years from now, yes, I'll still expect OpenGL.

One of the benefit of a PC is that you're supposed to retain compatibility with most software for a long time... I really don't appreciate when hardware vendors think games are a reasonable exception there. They aren't. Hell I was PISSED to learn that you guys got your stupid third party extension idea accepted to Vulkan, which all but guarantees a lack of future compatibility with games.

[Holograph]

i think you miss some points.
first of all think for 1 minutes about amd drop everything what is not vulkan do you even realize the point that you can run OpenGL over Vulkan?
It is a FACT that you only need Vulkan as a Driver you can run DX9 or DX11 or Open'GL over Vulkan.

The GPUs work internally however they do based on which GPU you're talking about. The drivers take in some sort of pseudo-standard ("pseudo" due to extensions which will eventually become non-supported when the hardware vendors arbitrarily decide they don't want to do so anymore) language like OpenGL or Vulkan and then instruct the hardware what to do with it. The drivers do either need to support your 3d API, or you need a translation layer like VK9. The hardware has to support a certain level of features in order to implement a 3d API (with reasonable performance) and the driver also has to speak that language. And you can use whatever userspace libraries to convert APIs if they exist, at the cost of performance.

He's definitely asking why they should continue to care about OpenGL. Based on the post he made he literally seems to think that people have already abandoned OpenGL for gaming, leaving CAD as the sole user, even though there are current game engines out there without any Vulkan driver, and even more games out there which are themselves still current, but which have not updated their engine to take advantage of any possible Vulkan driver.

Maybe they should stick to consoles where planned obsolescence is expected/okay if that's how they feel. Or if that's not how they feel, I'd recommend clarifying because I insist that there is no other way to interpret what he wrote. I mean, maybe you could argue he's only talking about Windows, but I would say it's ridiculous to segment such a statement by OS. There are sill plenty of supported (by their author) games using OpenGL without a Vulkan option on both Windows and Linux.

If I'm wrong, let me know, but GPUs do not speak Vulkan internally and you cannot just run any API over Vulkan without significant additional effort, so I don't believe I'm wrong on the point you called me out on.

[bridgman]

It's really frustrating how fast you hardware/driver guys want to get into terms like "legacy"
It feels like you'd stop supporting any games running on your GPUs within 30 days if you could.
You realize that many gamers play games that are a few years old, right?

With respect, that was not my question. I was asking if it was likely to be worth restructuring the Windows driver stack starting a year or more from now (after adding compatibility support) just for a slight increase in OpenGL performance. My feeling was "probably not" because existing applications could probably be improved more effectively via working on the current Windows GL driver..

I say "slight" because some of the Mesa performance advantage on Linux comes from the fact it has been getting the performance tuning effort on Linux games which goes into the closed-source driver on Windows games... so we would initially be replacing a driver which already has a fair amount of Windows app optimization with one whose only optimization had been for Linux apps. We're talking about a *lot* of work which could be applied in other ways to benefit users, ie there would be a very high opportunity cost.

I don't want to argue this on a technicality as I know that the very most recent games will almost always have priority of support but do you seriously need to ask if any of us still play OpenGL games? Do you really need an answer to that? If so, YES OF COURSE WE DO. I'm so frustrated that the question even has to be asked. (If you want to argue the point of your question was if we needed it to be "faster" then unless you demonstrate that no OpenGL game EVER has any slowdowns on any computer which meets said game's minimum requirements, then yes, we literally always need "faster")

I can help you with this. You are frustrated because "the question even has to be asked" but I DIDN'T ASK IT - I ASKED A COMPLETELY DIFFERENT QUESTION.

My question was whether the trends in the gaming market would support moving to a new Windows GL driver at this point (or actually a year or two from now because significant work on compatibility profiles would be required before it even made sense to start).

I play games like 7 Days to Die which is similar to Minecraft and uses the Unity engine (which does have an OpenGL driver)... It loads quite a large area and stresses my Haswell + 290x somewhat still. Planning a CPU upgrade next (whether that is Ryzen or not depends on when and how you guys address the GCC segfault issue that has been mostly ignored so far from what I've seen). 7DtD first came out a few years ago but still gets significant updates - it got a big one within the past few weeks.

Isn't 7 Days to Die a DX10 game on Windows, not OpenGL ? Are you saying that you play it on OpenGL on Windows, and if so why ? (serious question)

It just boggles my mind that you would want to stop caring about OpenGL already.

Again, you are totally mis-interpreting my question to a degree that boggles the mind. The topic under discussion was whether it would make sense to do a significant re-structuring of the Windows driver stack in order to use the Mesa driver rather than the current closed-source GL driver.

I won't forget this post next time I buy a GPU, because I don't care if that's 5 years from now, yes, I'll still expect OpenGL.

Of course you will - never a question. If you are going to remember the post though, please make sure you understand it first. I think that is a fair request.

One of the benefit of a PC is that you're supposed to retain compatibility with most software for a long time... I really don't appreciate when hardware vendors think games are a reasonable exception there. They aren't. Hell I was PISSED to learn that you guys got your stupid third party extension idea accepted to Vulkan, which all but guarantees a lack of future compatibility with games.

Again, you are responding to questions I didn't ask and sentiments/beliefs I didn't think. It makes it hard to carry on a discussion.

[bridgman]

He's definitely asking why they should continue to care about OpenGL. Based on the post he made he literally seems to think that people have already abandoned OpenGL for gaming, leaving CAD as the sole user, even though there are current game engines out there without any Vulkan driver, and even more games out there which are themselves still current, but which have not updated their engine to take advantage of any possible Vulkan driver.

Nope, that is not at all what I am asking, and after re-reading my post a number of times it's still not clear why you would believe it is.

My question related to whether we will see a significant stream of new OpenGL games arriving on Windows (which we could potentially influence to take advantage of a driver designed for core profiles only) or whether the OpenGL games we play a year from now will have largely been written for and optimized around our current driver.

[bridgman]

yes,,, but it is really smart to follow Nvidia ?
and in the end to support Nvidia by supporting Nvidias tactic to encourage Independent Software Vendor's to support Nvidia-style driver's only.

in my point of view it is maybe smarter to drop any Nvidia-only(by vendor-specific design decision) workstation software
It maybe motivates the Independent Software Vendor's to write generic software instead of Nvidia-only software.

Sorry, maybe I wasn't clear. It's not that compatibility profile support is *inherently* "NVidia-style", it's that compatibility profile support tends to end up having vendor-specific behaviour because the specification is not as strict in terms of interaction between new and deprecated API functions as it is for the core profile behaviour, and as a result vendor-specific behaviour automatically creeps into every implementation without any explicit design decision. This tends to favour "whoever gets there first" because expectation of that behavior (whether correct or not) gets hard-wired into application code.

I hate to say it, but if any vendor (including NVidia) decided unilaterally that they would stop supporting existing CAD workstation software the more likely outcome would be for the ISV to simply remove that vendor from their certified HW lists and carry on. If they were writing new applications or even planning to make significant investments in their current applications we would not be having this discussion.

[Holograph]

Nope, that is not at all what I am asking, and after re-reading my post a number of times it's still not clear why you would believe it is.

My question related to whether we will see a significant stream of new OpenGL games arriving on Windows (which we could potentially influence to take advantage of a driver designed for core profiles only) or whether the OpenGL games we play a year from now will have largely been written for and optimized around our current driver.

"My impression was that all the OpenGL users of old had hopped across to Vulkan"

If you meant that NEW SOFTWARE doesn't use it then perhaps you could have worded it in a better way. "OpenGL users" are "users who use software that runs on OpenGL" regardless of what you actually intended to express. I definitely reacted to what you wrote as far as I see it. You guys already got your extensions accepted into Vulkan, but you can't extend the English language the same way

Anyway, I hear you that you didn't mean that.


"Are there any serious Windows games that would benefit from a faster OpenGL these days ?"

Always work necessary on minimizing frame times. Literally always.

[bridgman]

"My impression was that all the OpenGL users of old had hopped across to Vulkan"

If you meant that NEW SOFTWARE doesn't use it then perhaps you could have worded it in a better way. "OpenGL users" are "users who use software that runs on OpenGL" regardless of what you actually intended to express. I definitely reacted to what you wrote as far as I see it. You guys already got your extensions accepted into Vulkan, but you can't extend the English language the same way

Ahh, OK... I guess I meant both but it wasn't very clear... I was talking about users running recently developed OpenGL games which would be likely to make use of new GL functionality rather than old (and now deprecated from core profile) functionality... but since it's game developers doing the initial hopping to Vulkan and users following as the games get released I really should have said "developers" instead of "users".

Part of the challenge here is that we all respond to posts with an implicit context from the ongoing discussion (which was about app developers and old vs new GL functionality). Making each post stand alone without assuming context would make each message a lot longer...

... and make posting so time consuming we probably wouldn't have time to post in the first place

[bridgman]

Are you trying to run a miner on upstream Clover CL ? I didn't think that worked... you would probably have more success with our OpenCL driver.

[haagch]

Doesn't work great, but without X it does work somewhat https://bugs.freedesktop.org/show_bug.cgi?id=101584

As the bug report says, you only need in libethash-cl/ethash_cl_miner_kernel.cl

#pragma OPENCL EXTENSION cl_clang_storage_class_specifiers : enable

[bridgman]

I was talking about the OpenCL driver we ship with AMDGPU-PRO, which should work fine with your 8320 and RX470.