tioslash
Member
Well, that actually works.
Wonder what it disables though, didn´t see any differences in IQ but only did a quick test so far.
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
Witcher 3 PC Performance Thread
- Thread starter batman2million
- Start date
Well, that actually works.
Wonder what it disables though, didn´t see any differences in IQ but only did a quick test so far.
Gumbie
Member
Holy shit. Can't wait to try this when I get home.
I've just tested that Kepler fix and it seems to work and from what I can tell not actually change anything in game. I think it may just fall back to running on the CPU when it can't run it on the GPU? Perhaps there's just a bug with the library on Kepler GPUs for now.
I'm pretty sure the clothing includes things like nets, sheets, flags waving in the wind, etc, but I could be wrong on that.
I'm pretty sure the clothing includes things like nets, sheets, flags waving in the wind, etc, but I could be wrong on that.
Shit that's amazing. If it's not fixed on NVidia's end by the time I get home Saturday I'm going to try that.
This is interesting if actually true.
If the game doesn't crash if you simply delete the dll, then this means one of two things:
In case of 2, it could be that the system dll is a different version which shows better performance in TW3.
Still strange that it would make such a big difference, my first instinct with something like this is always user error
If the game doesn't crash if you simply delete the dll, then this means one of two things:
- It didn't actually use it, in which case it also can't have a performance impact.
- It uses the dll in a system folder (installed by a driver) instead.
In case of 2, it could be that the system dll is a different version which shows better performance in TW3.
Still strange that it would make such a big difference, my first instinct with something like this is always user error
Steve Youngblood
Member
Thinking about it a bit and with minor research, I'm wondering if inadequate power may be the issue. I bought a 970 about a month ago to replace a 770, and put the 770 that was in it into my old secondary PC to replace a 560 Ti. Well, the old machine only had a 500W power supply. I didn't even think about that. Before I replace the PSU, does anyone think I'm barking up the right tree in eyeing that as the culprit?
For reference, here was my last post:
For reference, here was my last post:
CarbonFire
Member
Awesome, going to try this as soon as I get home. Makes me wonder if there are other "fixes" to be found.
Sevenfold
Member
Cross posting for more reads...
Any help greatly appreciated. Geralt is fine. Just horses.
RandomSeed
Member
Well I'll be a dumbass. Returned to the last official AMD drivers (Dec I think) and the fucked up meshes are gone. Drivers would usually be one of the first things I'd think of, but I haven't changed anything since I've been playing the game. Strange.
tioslash
Member
This is interesting if actually true.
If the game doesn't crash if you simply delete the dll, then this means one of two things:
- It didn't actually use it, in which case it also can't have a performance impact.
- It uses the dll in a system folder (installed by a driver) instead.
In case of 2, it could be that the system dll is a different version which shows better performance in TW3.
Still strange that it would make such a big difference, my first instinct with something like this is always user error
Certainly interesting. There were several areas in the game where my fps would inexplicably drop (including the one used in video from that person) from 30 to 25-20 and even less.
And there wasn´t anything particularly GPU intensive going on in these scenes that justified this drop (like just passing by a single hut in a certain place, or entering an area with 5 NPCs and experience a huge drop) while in another huge place, with tons of NPCs, houses, keeping a solid FPS.
After deleting this file it keeps the fps locked at 30 on all those areas.
Until recently, I was powering a 980 build with a 500W PSU and had no problems whatsoever. I upgraded about two days ago because I just wanted to be safe, and give myself the necessary room for overclocking purposes down the line, but I played TW3 pretty extensively before doing so and didn't have any stability issues.
So, I'd say it's worth upgrading, but I'm not so sure you'll see improvements.
s_mirage
Member
It's not just you, I have the same problem if Precision X is running. Freeze + driver crash.
Welp. There it is.
It seems to be in certain areas rather than an fps gain across the board. Look forward to trying this tomorrow.
Also want to know what exactly Nvidia's "fix" was going to be...
JudgmentJay
Member
Nice, can't wait to try this when I get home.
Steve Youngblood
Member
Yeah, but -- and I'll note in advance that there are people here who know a hell of a lot more than I do about PC hardware -- the specification for the 970 and 980 only call for a 500W. Meanwhile, the 770 specification recommends a 600W power supply. Are they full of it? I don't know. Like I said, I'm not really in my comfort zone here. All I know is that I didn't even think about the PSUs when swapping cards around, and I'm hopeful that the fact that my current PSU is simultaneously old (got it six years ago) and 100W below the recommended spec might explain crashes every 5 minutes.
http://www.geforce.com/hardware/desktop-gpus/geforce-gtx-980/specifications
http://www.geforce.com/hardware/desktop-gpus/geforce-gtx-770/specifications
I'm having some weird cross stitch effect on character models while playing this. It happened with the witcher 2 also.
During gameplay when watching a conversation play out Geralt and co. Will often have a weird blocky texture look to certain parts of their face.
I'm playing on Ultra at 1080p.
Is it a monitor or resolution thing?
During gameplay when watching a conversation play out Geralt and co. Will often have a weird blocky texture look to certain parts of their face.
I'm playing on Ultra at 1080p.
Is it a monitor or resolution thing?
Okay, unless I'm being stupid, I think it has to do with GPU PhysX. I reverted change of the file back to default (so the game loads it) and then went to the NVIDIA control panel and set PhysX processing to CPU.
Screenshot (PhysX - CPU) 71fps:
Screenshot (PhysX - Automatic (auto-selected my 2nd 780)) 62fps:
fps counter on the second is a bit hard to see, especially with imgur compression, but I've double checked, it's 62fps. Keep in mind, I'm running SLI also, so I'm not sure if that'll change anything in relation to your testing - but the guy in the video is running a single card.
Screenshot (PhysX - CPU) 71fps:
Screenshot (PhysX - Automatic (auto-selected my 2nd 780)) 62fps:
fps counter on the second is a bit hard to see, especially with imgur compression, but I've double checked, it's 62fps. Keep in mind, I'm running SLI also, so I'm not sure if that'll change anything in relation to your testing - but the guy in the video is running a single card.
Dude, it's probably the driver. I rolled back and haven't crashed since. Do the same and forget until Nvidia pull their finger out.
EDIT: Didn't read any of the stuff about that dll :O
RandomSeed
Member
The only tweak I did was the one to remove the KB/gamepad message, but when I reinstalled everything, I tested with zero tweaks. Was corrupt drivers after all...ah well, I'm happy I saved $300+.
That would explain why the problem is quite area-specific. Like, where there's lots of... PhysX... Stuffs...
Right...?
tioslash
Member
Yeah, I can confirm that changing the setting PhysX processing to CPU on the NVIDIA Control Panel have the exact same effect as deleting that file. A 10-15% increase in FPS in all specific areas where I was getting drops.
I´m running a single 760.
if all this is true then it may explain why the AMD counterparts were performing better. They just default PhysX to the CPU. I wonder if this affects performance on Maxwell cards. Has anyone tried?
Dictator93
Member
Hrm.. does that mean the game DOES use GPU accelerated PhysX?
The PhysX identifier says it is running on the CPU though...
You mean on Maxwell cards?
People already tested kepler.
Yup, before seeing your post, I thought of the exact same thing when I saw the original "fix." Earlier, I did the same test (runnning a GTX 770 4GB), first with the file deleted, the second time with the PhsyX set to "CPU."
Seems like both do the same thing, and only in certain areas; many areas, for me, show little to no improvement with either fix.
It looks as if the performance drop in affected areas are due to the PhysX processes intended to run solely on the CPU being offloaded to the GPU instead. Strange, seeing as it's been confirmed that all PhysX in The Witcher 3 run off the CPU.
Well Apex clothing can run on either the CPU or GPU and the name of this file is 'Apex_ClothingGPU' presumably meaning they're running it on the GPU when available.
If you look at my GPU usage on the screenshots, when PhysX is set to the CPU - both GPUs running at 98%. Whereas when I set it back to the GPU, the second GPU (which is the selected PhysX processor) is 98% whereas the first GPU is now 84%, presumably bottlenecked by the additional workload on the second GPU?
Has anyone tried with the file deleted AND PhysX set to CPU to see if there's a bigger improvement? Or do they just do the same thing?
yea, fixed it. lol.
tioslash
Member
Lovely Salsa
Banned
Would I double the fps if I bought another gtx 680?
Aww okay. Think I was being a bit greedy hoping for moreYes, I tried.
GHG
Member
Another 680 is not a good investment in any way at the moment due to the VRAM limitations on that card.
Put the money towards a proper GPU upgrade.
m_shortpants
Member
Interested in this as well.
Aww okay. Think I was being a bit greedy hoping for more
Yeah, I also just did another test in the abandoned village, and both methods (Apex file removal/PhysX on CPU) lowered GPU usage by 10%. I'll take it
JudgmentJay
Member
There are 4gb versions of the 680... he didn't state which he had.
I can't speak for 2 680s, but I have 3, and at 1080p with mostly ultra settings (foliage distance on high), I get 80-90 FPS in the most graphic-intensive area I've discovered so far. Probably would be higher if I offloaded PhysX to my CPU which I'll be trying when I get home! So yeah, the scaling seems pretty great. Not sure about double, but I'd wager at least an 80% increase.
Leatherface
Member
JudgmentJay
Member
Does anyone remember what post (or thread even) had the screenshot comparison of Ultra vs. Low shadow settings?
I can't find it
http://www.geforce.com/whats-new/gu...-guide#the-witcher-3-wild-hunt-shadow-quality
![[Asmodean]](/data/avatars/m/123/123312.jpg?1526393694){kind=avatar}
[Asmodean]
Member
I'm obviously not 100% on this, but re people deleting / renaming the GPU Physx library.
I'd imagine it's because, if the game doesn't detect the gpu lib on runtime, it reverts to the 'more simple' CPU physics they use on consoles.
Hence the increase in performance. Good tweak, though. If it has no major side effects.
I'd imagine it's because, if the game doesn't detect the gpu lib on runtime, it reverts to the 'more simple' CPU physics they use on consoles.
Hence the increase in performance. Good tweak, though. If it has no major side effects.
Mad Season
Banned
Yeah tried it as well no dice but at least I'm able to keep a consistent 30fps across the board. Using a 660ti here though.
I have a dedicated gpu as physx card(gtx 670) and its usage is 0 when playing witcher 3 when I check MSI's afterburner.
Funny enough I didn't think it was part of the nvidia guide but that's the exact screenshot comparison I was looking for. Thanks for the quick reply!
wanderjahr
Banned
This is interesting if actually true.
If the game doesn't crash if you simply delete the dll, then this means one of two things:
- It didn't actually use it, in which case it also can't have a performance impact.
- It uses the dll in a system folder (installed by a driver) instead.
In case of 2, it could be that the system dll is a different version which shows better performance in TW3.
Still strange that it would make such a big difference, my first instinct with something like this is always user error
What do you think it is in this case?
IIRC Physx includes fog right? Anytime fog is introduced in a scene my FPS goes to total shit yet when I've got tons of enemies on screen it holds solid.
Weird, I just dedicated one of my 780s to PhysX and I get 14-16% usage on it.
There are PhysX libraries for stuff like volumetric fog, but I don't think this game uses it.