Mass Effect 1 - Insane Glitching / Artifacting on AMD hardware

you can't test for how a wrapper would change things in a benchmark that is purely testing gpu speed.

 

Then how you estimate the boost in performance from using the DXVK? Benchmark can give FPS difference with and without DXVK, if you people claim noticeable difference in overheads.

 

Ah, a fellow Heaven Benchmark user (it's my fav). Yup, planned to do something like that actually.

Will give it a go, so far have to understand if it actually runs properly on my end.

 

The Navi10's or the 5000 series has some behavior where CPU and GPU utilization can be a severe bottleneck as GPU clock speeds just drop to low or even idle even with a 3D engine going which should never happen, Mass Effect 1 I believe is old enough to run on D3D9 so single threading and thus even worse performance characteristics.

AMD has improved this but even some D3D11 titles are infamously sensitive to poor performance scaling and DXVK (Now includes D9VK) resolves it even if as a wrapper it should incur a CPU overhead performance penalty meaning a 10 - 20% loss assuming everything works as intended.


Instead there's a sigh and this is a fun one....30, 60 or even 500% performance gain depending on game and throttling it's kinda weird but I have no real theories as to what this issue is or why AMD has not made any serious solutions to it even workarounds.
(Early powerplay info held that the way the Navi GPU scaled it kinda had issues with trying to force some minimum state like earlier cards could but that could have improved or even changed through driver software changes.)

https://imgur.com/a/qjI3X8n

Borderlands 2 D3D9 67 FPS to 89 FPS.
~32%

https://www.reddit.com/r/Amd/comments/ek9zvd/potential_fix_for_sekiros_performance_issues_with/

Sekiro D3D11 51 FPS to 73 FPS
~44%


Witcher 2 well the info here said it used to run at single digit FPS and then DXVK made it run at the refresh rate so that's like 60 FPS up to 120 or 144 possibly.
~500% or more using 9 for the single digit figure and then 55 or so.
This one I've heard AMD actually did fix in a later driver around 20.3.x or possibly 20.4.x so that sort of problematic performance scaling is hopefully going to be a rare exception.



EDIT: NVIDIA GPU's also clock down but not where it actively harms performance.
Though Ubisoft has *something* in Assassin's Creed Odyssey and also Origins so DXVK benefits NVIDIA GPU's here too in some manner whatever the bottleneck is. (Severe draw calls and general GPU bottlenecks?)

Also benefits NVIDIA under D3D9 though less than AMD but some PC ports probably still see solid gains from using DXVK.


https://github.com/doitsujin/dxvk/releases
And the latest compiled builds.
https://git.froggi.es/doitsujin/dxvk/-/jobs

Currently the extensions drawing on 1.2.140+ won't work on the Windows AMD or NVIDIA drivers just to mention that which can be a bit of a compatibility thing or give additional performance when supported.

AMD uses 1.2.139.0 and NVIDIA I think is at 1.2.133 on the main drivers now?
(AMD bundles the 1.2.132.2 runtime for whatever reason though but the latest one is available separately and backwards compatible. https://vulkan.lunarg.com/sdk/home just the runtime don't grab the full SDK it's not needed outside of development work.)

Vulkan development driver from NVIDIA might be 1.2.145 ready though.
https://github.com/doitsujin/dxvk/commit/9f3fb028b0d8ff170a411480b28838f97fc78fdf

https://www.phoronix.com/scan.php?page=news_item&px=Vulkan-1.2.145-Released

 

And after all that text on DXVK and performance scaling and back to the actual topic itself I don't know as to why DXVK would resolve this glitch here, maybe just something is skipped or avoided wrapping the D3D9 to Vulkan calls and if it improves performance that's a win.
Layers can be a bit trickier for hooking up though, ReShade 4.7.0+ is pretty compatible but not with everything and there's some potential issues with having more programs hooked in (Additional layers.) plus possibly programs such as some recording software being known as highly incompatible even without other software and their layers.

DXVK itself works on Windows and is really good as a Navi GPU user but kinda unsupported and has additional compatibility problems but overall it tends to be functional and pretty handy.
(Easy enough to just try and compare and see how the results are and how the GPU operates and advanced info such as frame time consistency and performance scaling as well.)


EDIT: For my own uses I utilize it to work around GPU compatibility problems like the other infamous black screen lock-up much as AMD has also improved (But not fully fixed.) several of these whatever the root problem is issues.

Behavior like consistent performance scaling and the benefit of higher framerate in many cases is a nice bonus though as is having stuff like the fan speed ramp up to the target and not down from near 100% even if the overall effect is the same. (Just a lot less loud. )


More advanced testing and additional also advanced overrides could be done but I'm happy how it works, D3D11 is in a pretty good state but the various known DXGI, D3D and WDDM or other OS issues and driver problems remain a concern which I hope to see AMD put a solid effort into fixing ahead of Navi20 coming out.

Not the best at explaining and the newer drivers have several improvements and some of the worst performing issues and consistency problems or stability concerns have slowly been ironed out and mostly resolved but it's not 100% resolved yet either.


At least DXVK is a really easy one to test, drop the D3D11 and DXGI .dll files (Or the entire thing for 32 or 64 bit specifically.) and just compare with and without it for 10 - 20 minutes or so.

ReShade and other stuff can be added on later if wanted I'd recommend testing one by one avoiding multiple hooks through Vulkan when wrapped like this to ensure it works and nothing bugs or breaks and making it easy to see what acted up before adding on everything. (Pretty standard stuff I suppose.)



EDIT: Oh right!

32-bit has a maximum 4 GB address space or less if it's not large address aware.

2 GB I think is where it's at then for non-LAA exe files but this is a one byte flip and can be done via tools like this one.
https://ntcore.com/?page_id=371


DXVK uses more memory and you might run into a out of memory crash as a result for 32-bit applications even with Large Address Awareness bit patched in just for something to be aware of.

 

THX guys.

DXVK - This is awesome for clunky UE3/4 Games (or other that lacks of opti.)
Tested so far Heroes VII & Senuas HB, it's worth noting that You need to run the game and let it create VLK cache then it just working great.
First run is always stuttery, 2nd is smooth.

In Senuas i have less heat - was 75deg now it's 69deg

 

-> https://cookieplmonster.github.io/2020/07/19/silentpatch-mass-effect/#introduction

 

Yeah i've been talking to MSFT about possibly using detours in a AppCompat patch to perform this without a dx injector

 

So @cryohellinc have you tried that silent patch, linked by OnnA?

 

Haven't had a chance yet - was really busy for the past days.

I plan to give it a go this weekend. However - I used a mod from Mass Effect 1 Nexus website, that covered a fix for this. I assume they differ in implementation, meanwhile have the same outcome?

Also, have to get back to benchmarking the DXVR.

 

@cryohellinc

I am quite curious about DXVK benchmarking. And about silent patch too.

 

This topic made me to try DXVK on WoW and it made an impressive difference, 20-30% up in the frames! However... everything else I've tried resulted in worse performance

 

Yet, when I linked to you the fix, you went apeshit on me.

Cool.

 

I will insist you on quoting me where I have gone "apeshit".

And once again - that wasn't the problem I had, nor was it fixed in any way by this solution.

 

This doesn't sound friendly to me.
In the way you worded it.

Still, all i wanted to do was help.

 

Sure, can be a grumpy slav sometimes. "slowly puts on a tracksuit"

 

Are you for real in touch with them?
Can you please for the love of god get them to drop the run as admin appfix? They are breaking other games with that half-ass workaround.
And when I tried to contact even their second-tier support, they just told me "it's an old game so we can't touch it".

 

only the DX team!

im not sure what you're seeing get broken by the app fix?

 

Borderlands 2 scores up to 50% higher CPU bound performance with it, and at least on Nvidia, also Mirror's Edge is much faster with recent git-master, even when GPU bound (AMD might have issues with MSAA performance).
Though that might also be because Microsoft has D3D9 performance let regress with Windows 10. Who knows if also Battlefield 3 D3D11, which at least temporarily regressed, runs as fast as it used to.

Edit: Anyhow, the Navi D3D9 driver is phenomenally bad and you're basically forced to use DXVK anyway.

 

I am not. DX games run without issues on my rig (and on previous one too).