AMDGPU-PRO 16.50 vs. Mesa 13.1-dev + Linux 4.9 Radeon OpenGL

[pjezek]

I have noticed very impressive results of relatively lazy RX 460. This is realy a live water for these cheap chips in linux environment.

[agaman]

They don't have proper automated benchmark modes, thus not tested regularly and only rarely for premium readers.

I think Tomb Raider has added a command line benchmark mode not too long ago...
I have found this PTS test for Tomb Rider (I have not used it, so I can't say if works): http://openbenchmarking.org/test/pts/tomb-raider
If it works, would be nice to add that game to the normal benchmarks you do periodically when new driver versions came out

[bridgman]

You know that, coming from the green camp, all I care is "the driver". Whatever you decide to call bits and pieces of it, is no concern of mine

I'm trying to reconcile the above statement with the fact you explicitly called out one of the bits and pieces (DAL/DC) by name

[bug77]

I'm trying to reconcile the above statement with the fact you explicitly called out one of the bits and pieces (DAL/DC) by name

It's hard to ignore names when Michael keeps posting 2-3 articles each day about every check in :P
I know there are several parts, I know names, but I have no idea what they do or how they work with each other. That's the part I haven't grasped.

[Michael]

I think Tomb Raider has added a command line benchmark mode not too long ago...
I have found this PTS test for Tomb Rider (I have not used it, so I can't say if works): http://openbenchmarking.org/test/pts/tomb-raider
If it works, would be nice to add that game to the normal benchmarks you do periodically when new driver versions came out

I added that PTS test, but as written on that page "- Initial commit, setting options seem to have problems currently." at last check, it still wasn't obeying settings behaviors properly and thus can't be used reliably for benchmarking.

[agaman]

I added that PTS test, but as written on that page "- Initial commit, setting options seem to have problems currently." at last check, it still wasn't obeying settings behaviors properly and thus can't be used reliably for benchmarking.

Ok, sorry. I didn't noticed that
Then, forget what I said.
Anyway, thanks for all your hard work!

[Shevchen]

To the "green team cheating" part:
Under Direct X, its known that Nvidia exchanges "shader paths" (or however they are called) with their driver to optimize performance. They also include some sort of "invisible quality cutoff" that cuts off graphics quality in areas that the human eye cannot perceive anymore. But I wouldn't call it cheating per se, its more about "intelligent optimization". If the user doesn't notice the difference, but you get more performance, where is the problem?

Its like the jpeg: Nobody complains about the jpeg, even thou its a lossy format, but the advantages far outplay putting in bmps on every website. AMD would need a dedicated driver-optimization crew to do the same, for every game - which is somewhat a waste of time, if you can very similar results with just a low level overhead API - you save yourself the hassle to optimize your driver for every game (and leaving out unpopular ones at whole)

The thing is: Once Vulkan arrives, this advantage from Nvidia gets close to zero, making raw performance and good drivers more important. Seeing the improvements on the OS drivers for AMD cards, I might guess its about 1 year until we reach parity with Nvidia and then its about price/performance or just raw performance. That being said, I'm waiting for the Linux version of Star Citizen before I buy a new card and this card should be a *big* one. So please AMD, would you mind putting out a new Vega card with 6k/8k shader units please? Or, when Navi arrives: 12/16k. Something that puts pressure on the Titan-line. I'm willing to pay you appropriately. (Not 1 grand thou)

[bug77]

To the "green team cheating" part:
Under Direct X, its known that Nvidia exchanges "shader paths" (or however they are called) with their driver to optimize performance. They also include some sort of "invisible quality cutoff" that cuts off graphics quality in areas that the human eye cannot perceive anymore. But I wouldn't call it cheating per se, its more about "intelligent optimization". If the user doesn't notice the difference, but you get more performance, where is the problem?

What you are describing here is very similar to Z culling or Mipmapping. Remove stuff that isn't seen/perceived anyway.

[bridgman]

Mipmapping yes, Z culling not so much. Z culling is about removing stuff that absolutely can not be seen/perceived, while Mipmapping & shader replacement are about removing stuff that arguably can not be seen/perceived by someone with average vision.

Just to be clear, when people talk about "cheating" they are generally not talking about things like shader replacement (except back when it was done for the first time and so was obviously a terrible thing ).

The big issue that gets discussed is relaxed API compliance checking relative to the API spec, which (a) reduces CPU overhead allowing higher frame rates on CPU-limited games and (b) when combined with an aggressive developer relations program can result in games with "not quite right" code that runs fine on one vendor's HW but fails on other drivers.

Formalizing lower overhead APIs with optional API checking go a long way to dealing with the first issue (there are a few remaining gaps, example below), while having game developers test on multiple vendors during development usually avoids the second.

http://stackoverflow.com/questions/3...to-clear-depth

[Max Payne]

Its hard to call mesa the definitive winner I think as it lacked behind on ET and DeusEx, plus it doesn't have Freesync option (a gaming feature). So hopefully the issues with those titles can be resolved. I would have liked to see Tomb Raider and Mordor compared at 1080p and 2160p also!

Gotta agree there, also mesa loses on CS:GO at 1920x1080, by some 30fps, it's very playable, but on esports more fps is always important, mesa probably has more overhead than the proprietary driver.

It would be interesting to know how they compare to each other in regards to image quality, my windows gaming buddies tell me the new generation of NVIDIA cards brings more quality to the screen when compared to AMD's last 28nm generation of cards, so I guess AMD devs also need to look into that once they address more pressing performance and rendering issues on some games.