Radeon OpenGL Linux Driver Massively Improves 3D Texturing Performance For Older GPUs

[cutterjohn]

They are taking consequences of their greed right now. Their driver for linux is not in great state as well, especially for games which are using XeSS. So no, thank you. (For now)

it's worse than than that, they're splitting the drivers with the current GPU left in limbo.

One vocal phoronix member wants us to think that the Xe driver is 'experimental' 'specifically' for Alchemist! Meanwhile IIRC Intel has repeatedly stated that for 'best' 'perf' use the Xe driver once available! HOWEVER that loses HWENCODE as hwencode is teh too hard to implement for Alchemist cards, so if you want hwencode take lower perf and go back to i915...

The above is what that vocal idiot missed several months ago, and he had this weird fixation on Xe being 'experimental' PERMANENTLY for alchemist which was NEVER my understanding of the Xe driver for Alchemist. It's true that it IS CURRENTLY experimental, but NVER that it was going to be PERMANENTLY exoerimental for Alchemist... ah well... welcome to phoronix I guess, the slightly less toxic version of reddit... nowadays...

I must say that I could not believe what affront that guy took to mentioning that Intel was too lazy to support hwencode on their new driver because by their own words it was teh too hard, and remember these are intel guys writing these drivers not random drivebyes... makes me wonder if the guy was a fired former intel GPU drier developer who couldn't cut the mustard... as he was rather incensed, and in my research entirely incorrect other than apparently trying to drive his own agenda...(which ended up w/insulting me, mentioning that I have NOT obviously contributed to intel driver development(unsurprisingly given that it's primarily intel employees much like AMD and nVidia drivers are also comppany employees), etc.

ah well... I feel better semi venting about that guy now and whatever his misguided point was? If he had one?

One last thing that I was NEVER clear on w/the Xe driver was does it provide better compute perf b. i915? I would presume that it DOES, which is another reason that I cannot believe that they couldn't get even a q&d hack for hwenc going for alchemist on Xe, when they're already effectively recommending it when available... and no they do not call it 'experimental' for Alchemist. Sure right NOW it IS 'experimental' but once released that is irrelevant! i915 was the stopgap hack driver for Alchemist...

So unless I completely read this wrong, that bastard was completely incorrect and read like a lazy Intel employed(more likely formerly employed) engineer that was too lazy to provide a hwencode hack for Xe and alchemist... which is sad as they're quite happy to provide all sorts of hacks for BOTH Xe and i915 for other functionality...

[EDIT]

IOW I think that Intel hired the WRONG people to work on their drivers given how far behind schedule they are, that or understaffed the linux drivers, probably a combination of both... the windows drivers do NOT suffer from these problems, and at a low level there should not be much of a difference...

[/EDIT]

[smitty3268]

Ahh, yes your response straight from google shows a strong understanding of the subject matter.

Why did Intel have to patch AMDs CPUs, if AMD is so commited to supporting their hardware on linux?

And why was an official Intel contributor to the linux project unable to get any documentation for AMD CPUs even when contacting people at AMD directly?

That was years ago. I don't think anyone would argue that AMD's support was good back then. The company was almost dead, and linux support wasn't their highest priority.

I would actually agree with you if you just said that Intel supports linux better than AMD does. I don't think that's particularly controversial. It's just kind of silly to throw out asking about a well-understood bit on some old hardware and acting like people don't know how to program the current AMD cpus. If you have something more recent, please use that as an example instead of just picking something based off a funny name.

I certainly don't love their slow support of new compiler targets, or the poor support for stuff like their temperature monitoring/etc. Those are way better examples and way more recent.

[Jorgp2]

That was years ago. I don't think anyone would argue that AMD's support was good back then. The company was almost dead, and linux support wasn't their highest priority.

I would actually agree with you if you just said that Intel supports linux better than AMD does. I don't think that's particularly controversial. It's just kind of silly to throw out asking about a well-understood bit on some old hardware and acting like people don't know how to program the current AMD cpus. If you have something more recent, please use that as an example instead of just picking something based off a funny name.

I certainly don't love their slow support of new compiler targets, or the poor support for stuff like their temperature monitoring/etc. Those are way better examples and way more recent.

It doesn't matter how old it is, the situation has not changed.

Right now you can read up Intels documentation, and it outlines features and changes for upcoming products so that developers can make sure all their software works for it.

Or you know, you can diagnose, report, and patch bugs based on what a system is upposed to do.

[Eirikr1848]

Working in the test lab for a developer like this would be a remarkable experience, dang.

[chithanh]

Bruh's been playing solitaire on his Intel iGPU and thinks that enough for an average gamer.💀💀

Drivers on linux are subpar to window's ones. Games on windows perform up to 30% better. Include here that Intel said fuck off to archmage users declining media engine

It is also a problem with iGPUs. Try running Bazzite on an MSI Claw A1M, it is a miserable experience, way worse than on Windows (which is already not very good). The driver is lacking important Vulkan features. No convenient way of controlling TDP. Etc.

Ahh, yes your response straight from google shows a strong understanding of the subject matter.

I don't think your posts demonstrate great knowledge or understanding either, or at least suffer from very selective memory.

how many years it took AMD to actually implement power management for their CPUs.

On the Intel side, they actually work on hardware support before launch. And the community can actually work on the code themselves, since anyone can download their documentation without signing an NDA.

Why did Intel have to patch AMDs CPUs, if AMD is so commited to supporting their hardware on linux?

If Intel is so committed to getting their hardware work on Linux then what about Poulsbo and its successors? Phoronix has been reporting for many years on this sad story. When a volunteer (inside Intel) got fed up with the situation and started to write a reverse engineered driver, they were even instructed to keep it outside of i915 despite considerable amounts of code duplication.

And it is not that the situation was perfect outside Poulsbo. If you had a Bay Trail system, then maybe you came across the over a thousand comments long kernel bugzilla report how power management would lock up the CPU, which got almost no attention from Intel for years.

The successor Cherry Trail was supported only partially as well. Again a volunteer (this time from Red Hat) had to come work on bringing essential functionality to these devices.

I'm all for calling out AMD when they fall short. But portraying Intel as some kind of role model is completely not warranted. AMD at least aims for full open source support of all their CPUs and GPUs, something which can't be said from Intel.

[NeoMorpheus]

The best solution for regular consumers are intel gpus, actually. Better value for money. And the fuzzy feeling that you promote competition in the gpu market to fight off the duopoly that has inflated prices for so long.

I guess that you forgot how big intel still is, compared to AMD.

Also forgot how dirty intel still is (explain how dell still magically only use intel in their lucrative optiplex, latitude lines) or how they treated all of us when they had no competition.

Nah, fuck intel and ngreedia, both sucks from a consumer perspective.

[ET3D]

Although this fix is great, it feels more like "oops, our drivers slowed things down by mistake" than "wow, we some magical performance." Still, well done and it's great that Marek and the Linux drivers aren't bound by the deprecation policy from the Windows size.

The best solution for regular consumers are intel gpus, actually. Better value for money. And the fuzzy feeling that you promote competition in the gpu market to fight off the duopoly that has inflated prices for so long.

Not that much better value for money, and obviously if Intel ever gets competitive in GPU space it will raise prices. Intel won't want to keep the low margin it has due to Alchemist being large and inefficient. Intel's 406 mm2 chip performs about the same as AMD's Navi 33 that's half the size on the same process. Hopefully Battlemage will be better, but still, I see no reason for Intel to really push prices down once it becomes competitive.

[coder]

this shows that radeon cards for regular consumers are superior choice for now (at least while not just new hardware gets updates and attention from an open driver team).

Tell me: how dependent are "regular consumer" games & GPU-accelerated apps on 3D texture performance?

https://gamedev.stackexchange.com/questions/9668/what-are-3d-textures

Surely, you didn't think AMD left so much untapped performance for regular texture mapping!

[nuetzel]

Someone did post figures for polaris (rx580) in the merge request. I cant seem to understand what those figures mean though.

Hu?
Read them, again.

[nuetzel]

So? I agree he probably has access to at least one Polaris card, but again, maybe he didn't have a system with one ready to go, or he didn't have time, or he felt that the Tonga results were representative enough for gfx8 cards that he didn't bother repeating the experiment on Polaris.



They seem meaningless to me without anything to compare them to.

Read them, again.
And do the math - Marek and Pierre-Eric could do it, so it was Acked.

For your pleasure:

energy-03
Window 1900 x 1600, fps

	orig	patch
Composite Score	3,75	7,85
Blake Ridge volume	3,43	7,31
F3 Netherlands	9,61	16,28
Opunake	0,99	2,57
BRV with	4,97	11,09
F3N with	12,72	20,94
OV with	1,34	3,30