Meta: AMD Issues/Workarounds

[mirh]

Follows #1508 and hrydgard/ppsspp#8698 Gs dump is here.

Bissected up to either 16c2baa0df2d7859619d51d3995b78f057a8e965 or 29c97a9bf21a985e1524e0b428ff97aa678adcc4 Happens in Ace Combat 5 after "press start" screen only after Blending Unit Accuracy has been set to none in OGL hw.

I'd just complain over at AMD but I'd like to get a more straightforward testcase for them. Wouldn't be bad if somebody added "Upstream | External" label

List of AMD issues:

• ~SSO is broken. SSO is partially disabled in GSdx due to this issue.~
• ~Disabling blending causes a BSOD. This is related to SSO as the issue does not happen on v1.4 where SSO isn't mandatory and is disabled on AMD GPUs.~
• Dreadful performance. OpenGL can be slower by about 80% compared to Direct3D with same effects emulated.

Links to AMD forum issue threads: https://community.amd.com/message/2748362 https://community.amd.com/message/2756964

Possible BSOD Citra Workaround Merged from issue #2362 As Gregory requested so we don't forget about it.

Currently Citra added a workaround for the amdfail driver that fixes the crashing caused by SSO. The commit is located here https://github.com/citra-emu/citra/pull/3499/commits/0cf6793622b01f3941fbc77fe04c3b68476004ca

Reddit post: https://www.reddit.com/r/emulation/comments/88vva4/citra_on_twitter_new_update_to_the_hardware/

Idea would be for this to be checked out and maybe implemented.

Some useful info

You unbind everything so you pay extra invalidation What we need to do is Create a pipeline by shader combination So you only bind once stage to a pipeline And then we bind and rebind pipeline But I'm not sure it will fix the crash Potentially citra workaround might not work on ours side.

[Nucleoprotein]

@gregory38 Yes, but AMD do not burn they own GPUs by broken drivers like NVIDIA ;)

I also agree we need to wait for 16.12.x ... or 17.01.x

[avih]

I'm not registered to their forums, but could someone ask for clarification on this?

[FlatOutPS2]

I'm not registered to their forums, but could someone ask for clarification on this?

I'd doubt you'd get a clearer responce. The developers usually don't know, and even when they do, it seems that fixes get postponed to later releases fairly regularly.

[lightningterror]

AMD driver 16.11.5 Bug is still present.

[mirh]

Oh FFS. I guess "release" is just a silly euphemism to mean the specific branch then.

[Nezarn]

http://wccftech.com/amd-radeon-software-crimson-relive-driver-leak/

Maybe they meant this new release? (I hope they will finally include the fix...) This comes out tomorrow.

[Nezarn]

Installed this new driver, still BSOD\driver crash....

[Nucleoprotein]

I expect that because this release is their "year release", expect fix in 16.12.x hotfix or 17.x drivers.

[mirh]

Release number has nothing to do with internal builds one or branches rolling out. Aside of this, too bad it doesn't even fix ogl performance.

[Nezarn]

Also from what i've seen, this new driver is worse than older crimson drivers, settings app always crashes, and i get microlags in counter-strike global offensive lol

[Nucleoprotein]

Use "Clean Install" option.

[avih]

Can someone who's registered to their forums (where they posted their reply) please ask them what they meant by next release, and when, approximately, can we expect it?

[Nezarn]

@Nucleoprotein I always install drivers after cleaning with DDU. (and i always use custom install)

@avih I've asked them but got no reply. (maybe even they don't know what "next release" means) I've asked it on 11.29....

[lightningterror]

This is such a great driver. My only complaint is why they didn't implement this fix. I guess we can't have everything.

Soon™

[Nucleoprotein]

@avih Like @FlatOutPS2 said - the developers usually don't know when drivers will be released.

@Nezarn From this driver release you do not need DDU, they finally added "Clean Install" option to installer! I got Watman enabled for my R7 360 which is little strange, but setting app and drivers seems to work fine, just like 16.11.x Not tested ReLive yet.

[avih]

@avih I've asked them

Ah, thanks.

but got no reply.

:/

[lightningterror]

@Nucleoprotein That's because they enabled wattman for 260/360+ gpus

[Nezarn]

did anyone try 16.12.2 Optional yet?

[lightningterror]

@Nezarn Still not fixed.

[FlatOutPS2]

Small update from the AMD dev:

As stated, the issue has been fixed internally. I'm going to update this thread once it goes public.

So he still has no idea. xD

[mirh]

Showerthought: putting aside I damn hope next release will be massive (unless there just weren't many releases due to holidays..), I wonder if GSDumpGUI/pcsx2 run through wine on linux causes a kernel panic too.

[Nucleoprotein]

17.1.1 - still no fix

[gregory38]

ReLive driver ought to be called Zombie drier ;) AMD QA seems around 6 months. I'm afraid that it is still a matter of months. If you're lucky guy, you might even have an extra GL speed optimization, few months later.

[Nucleoprotein]

17.1.1 "Optional" drivers are for Resident Evil 7 so yep, Zombie driver ;)

[lightningterror]

17.1.2 No fix.

[gregory38]

So next release in AMD world means next release after you've waited an half-year.

[Nucleoprotein]

All 2017 releases so far are Non-WHQL, I expect this fixed in first WHQL one.

[lightningterror]

[FlatOutPS2]

@lightningterror Why? Do you desperately want to set Blending Unit Accuracy to "None"? :P

[lightningterror]

Perhaps it will offer better performance and I can switch to opengl for some games.

Tho it's not just blending accuracy i wait for but it's a good start.

https://community.amd.com/message/2756964#comment-2756964 is my main concern.

[gregory38]

Note: I tried my best to enable the "basic" level only when it is free (CPU) or really required. It is true that it can kill perf on a couple of games (sometimes it is even faster on basic, likely due to less blending state switch).

IMHO, you can wait 2018 if you want a release that improve the perf.

If you want to use openGL, sell your GPU to a poor guy. And buy an Nvidia GPU (you can reconsider AMD in 3/5 years when games are mostly Vulkan/Dx12).

[Nucleoprotein]

Performance of AMD OpenGL will not change even in few years, AMD stated that they will not open their OpenGL implementation (I think because of some patents) and their OpenGL remembers ATI ... I think they need to rewrite it from scratch to improve it and they have no money for that. RadeonSI on Linux already is faster than their ICD. ( http://www.phoronix.com/scan.php?page=article&item=mesa171-pro60-nvidia&num=1 )

[mirh]

They actually should have already fixed the performance bug in internal builds. We'll see how that will play then.

And read comments to your commits 🙃

[gregory38]

OMG Only remain to wait 6 months to get a public release. ;)

[FlatOutPS2]

They actually should have already fixed the performance bug in internal builds.

*They actually should have fixed a performance bug in internal builds. :P

[pixelherodev]

On Tuesday, January 31, 2017 1:36:19 PM EST FlatOutPS2 wrote:

They actually should have already fixed the performance bug in internal

builds.

*They actually should have fixed a performance bug in internal builds.

:P

Writes in Reminders: Never buy an AMD GPU.

[avih]

Writes in Reminders: Never buy an AMD GPU.

That's arguable. Other than PCSX2 with OpenGL, at least on windows, AMD GPUs provide good value for the most part.

[pixelherodev]

On Tuesday, January 31, 2017 2:05:28 PM EST avih wrote:

Writes in Reminders: Never buy an AMD GPU.

That's arguable. Other than PCSX2 with OpenGL, at least on windows, AMD GPUs

provide good value for the most part.

I'm a Linux user - from what I've seen, NVidia are the best, but I haven't owned a desktop in almost a decade, and both of my laptops have Intel GPUs.

[Nucleoprotein]

Today for AMD, Linux Mesa RadeonSI is better than AMD official OpenGL on Windows ...

[gregory38]

Hopefully Mesa will get soon a multi thread gl implementation (I pulled some strings on IRC). (I got a 30% increase on SotC rendering).

However it seems they have a mechanism to enable multi thread on the proprietary driver. Likely based on the app name. Maybe we just need to rename PCSX2 to get a speed boost.

[mirh]

Something I did try in the past (rage.exe iirc) But (putting aside blb format is a mess) I guess multi-threading or not, if we hit that other bug it's useless.

[gregory38]

Which others bugs ?

[Nucleoprotein]

I think @mirh is referring to TDR bug (ie. that related to this issue)

[mirh]

Oh god. This. Which I have already said [somewhere] they PMed me saying they had "found an optimization to the OpenGL Program Pipeline implementation [that] should get rolled into a release soon"

Where of course soon means by next year, but whatever.

[gregory38]

Oh. It isn't a bug. It just an extra validation cost for the driver. If you put the driver is a separate thread, it will get available time to do useless validation.

[lightningterror]

Guess what? AMD totally messed up the drivers. This is how the display looks now on 17.2.1

[gregory38]

Did you try 1.4 without SSO?

[lightningterror]

Works fine , I knew this would work. Nice work amd you managed to bug sso again.

On the testcase provided by mirh I get yellow instead of green again.

[gregory38]

Lol I was right when I said the BSOD was related to the fix. Seriously I don't see the point of 6 months of QA for that !

[lightningterror]

Sigh ...