Dxvk: The Witcher 3: Known issues

[doitsujin]

The Witcher 3 on DXVK has gotten some attention as of late, and while it does render correctly for the most part, there are still several issues left which may break the immersion or make the game unplayable:

• [X] Critical: Toussaint does not load for some players.
• [X] Rotfiends etc. are distorted or invisible (#135). Fixed in v0.90.
• [X] Hairworks does not work (#135). Fixed in v0.90.
• [X] Some geometry may be missing due to incomplete Tessellation support (fixed in master)
• [X] AMD RADV: Water surfaces have incorrect geometry. Fixed in Mesa 18.1.
• [X] AMD RADV: Game is broken on 17.3.x. Use Mesa 18.0 or later.
• [X] Nvidia: Terrain flickers with the new driver. Fixed in 396.24.02.

If you find any more issues with this game, please post a comment on this thread.

[shmerl]

An example of distorted rotfiends:

[shmerl]

Overall performance is also lower than with last working buffer pool hack in Wine. Around 30 fps with RX 480, 1920x1200, vs 40 fps with OpenGL + buffer pool.

[volca02]

All these were experienced with nvidia GTX 1080, so might be NVidia specific:

Focusing out of the game and back in will freeze the game (will try to backtrack git here, it didn't happen before the tesselation was added, but I didn't find the exact commit yet).

I also got a spontaneous freeze while in crafting menu:

Exception in CreateObject("red.game.witcher3.menus.inventory_menu.ItemSectionData"):
    Exception in CreateObject("red.game.witcher3.menus.inventory_menu.ItemSectionData"):
    err:   DxvkDevice: waitForIdle: Operation failed
terminate called after throwing an instance of 'dxvk::DxvkError'

abnormal program termination

with a matching line in witcher3_d3d11.log

err:   DxvkDevice: waitForIdle: Operation failed

Edit:
Another bug I found - setting any other ambient occlusion than None will freeze the game.

[shmerl]

Regarding framerate, it helps to set CPU governor to performance (at least it helped a lot on my Ryzen 7 1700X, I got 40 fps!).

sudo cpupower frequency-set -g performance

You can switch it back to ondemand after playing the game:

sudo cpupower frequency-set -g ondemand

[shmerl]

setting any other ambient occlusion than None will freeze the game.

That must be Nvidia specific, since I have no problem with HBAO+ with radv.

[doitsujin]

@volca02 can you please post the full d3d11.log? This sounds like the device gets randomly lost for some reason. SSAO and HBAO+ do indeed work fine on RADV.

[MartinPL]

I dont have any problem with ambient occlusion on GTX 1070
390.25 / Antergos (Arch Linux)
https://youtu.be/jaR7sMMms4k?t=4m7s

[roothorick]

Device randomly gets lots you say? Could it be a similar issue as #111?

[volca02]

I just re-tested the AO freeze and can't reproduce. Will update after I find more about that. I got the freeze just before entering the main menu.

@doitsujin Unfortunately I already overwritten the log file for that issue - but it seemed totally normal aside of the last line in that file, which is in the previous post (DxvkDevice: waitForIdle: Operation failed).

[shmerl]

By the way, is water broken only with radv, or it also happens with Nvidia? I.e. does it make sense to open a Mesa bug about it?

[Yardanico]

@shmerl even if it's radv only you'll need to provide a pure-vulkan example (without dxvk and wine)

[shmerl]

Why so? As long as it's radv bug, they should be interested in it. radeonsi developers are OK with potential bugs that are exposed through Wine, so why should radv developers be against it?

[Yardanico]

@shmerl they need to be sure that this is not a dxvk bug

[shmerl]

May be, but reproducing it outside of that scenario might be too hard. So if it's narrowed down enough as potential driver bug, it's in their interest to test it.

[aqxa1]

By the way, is water broken only with radv, or it also happens with Nvidia?

Water is fine for me on Nvidia (980ti) and broken with radv (rx550).

[shmerl]

OK, I'll wait until upstream Wine will get working Vulkan support, and if it will still happen by that time, I'll open a Mesa bug. That would reduce the burden of testing this scenario for Mesa developers.

[volca02]

There are actually at least two kinds of freezes.

One is refocus freeze, which goes away when using compositor (after running compton I can switch away from the game and back just fine).

The second freeze so far seems to be related to inventory screens/trading. It seems to happen when dragging items around.

I tried running with DXVK_DEBUG_LAYERS=1, and this is what I got that was unusual:

DS(ERROR): object: 0x0 type: 9 location: 2405 msgCode: 11: Cannot submit cmd buffer using deleted buffer 0x0.

Repeated many times.

Edit: This was tested on f152359f309d542cd04d8bdcf6024dc1993a967a (direct-image-mapping branch).

[MartinPL]

Fullscreen doesn't work if you set resolution below native.
Newest Nvidia prop diver / Wine-staging 3.4

[doitsujin]

@MartinPL That's not Witcher 3-specific.

[shmerl]

So, does it make sense to report water distortions in Mesa? They still happen with upstream Wine.

[doitsujin]

I already reported a bunch of tessellation-related bugs (see this) and it's being worked on. With latest llvm-git and mesa-git, It's still not perfect but at least one problem has been solved.

[shmerl]

Just tested with latest Mesa master and llvm 6.0.0. Water animation is smooth now (no jagged movements), but "stitching" issue is still present i.e. water surfaces elements don't connect properly leaving see through holes.

[shmerl]

Just for the reference, that's how it looks with Vega 56 at present (Mesa master, Wine master, dxvk master, kernel 4.15.4):

Is there a point to report it to Mesa bug tracker? Let me know if I can help troubleshooting it somehow.

[shmerl]

I see a bunch of errors in d3d11.log:

err:   D3D11: Invalid blend factor: 1042108203
err:   D3D11: Invalid blend factor: 1526616704
err:   D3D11: Invalid blend op: 4100
err:   D3D11: Invalid blend factor: 1073889088
err:   D3D11: Invalid blend op: 1042108203
err:   D3D11: Invalid blend factor: 1049700610
err:   D3D11: Invalid blend op: 1046135251
err:   D3D11: Invalid blend factor: 1086918619
err:   D3D11: Invalid blend op: 1040187392
err:   D3D11: Invalid blend op: 1966821600
err:   D3D11: Invalid blend factor: 32530
err:   D3D11: Invalid blend factor: 41301646
err:   D3D11: Invalid blend factor: 1073889088
err:   D3D11: Invalid blend op: 1042108203
err:   D3D11: Invalid blend factor: 1049700610
...
err:   DxbcCompiler: Unsupported operand type declaration: DxbcOperandType::InputPrimitiveId
err:   DxbcCompiler: Unhandled operand type: DxbcOperandType::InputPrimitiveId
err:   DxbcCompiler: Unhandled PS SV input: DxbcSystemValue::PrimitiveId
warn:  D3D11Device: No matching border color found for (1e+010,1e+010,1e+010,1e+010)

And these in dxgi.log:

warn:  DxgiAdapter: DXGI_FORMAT_D24_UNORM_S8_UINT -> VK_FORMAT_D32_SFLOAT_S8_UINT
warn:  Vulkan extension VK_KHR_maintenance2 not supported

[doitsujin]

None of these errors are Vega-specific. They are aware of those Vega issues so I don't think there's a need to report this.

It should run with mesa-git and RADV_DEBUG=nohiz set.

[MartinPL]

There is some regression with newest 5eedbc0, also ~-3FPS, Nvidia:
https://i.imgur.com/7lDMwrY.jpg
https://i.imgur.com/KjUfEHM.jpg
https://i.imgur.com/USFzqDJ.jpg

[doitsujin]

Are you sure it's not 53d557c2db9e51c98696840e19aced94687be0ea that introduced the regression? Because that's the commit that broke it for me.

I'm aware that it doesn't work at the moment and I think it'll stay like that for a while until I figure out what is actually wrong. The game stops working when mapping an image with D3D11_MAP_FLAG_DO_NOT_WAIT fails - and it is allowed to fail - which may actually indicate a game bug.

[MartinPL]

Just tested 53d557c 100% cause non ground stuff, but its looks like 5eedbc0 stole 2-3 FPS.
Some one should retest fps drop

[doitsujin]

The FPS drop is probably to be expected since DXVK now uses a different format for depth buffers on Nvidia and Intel hardware, but using 24-bit depth buffers causes issues in other games.

[doitsujin]

Added workaround for the missing terrain in d9b1995cf05c9bd8ab1b6bbcb884d37fbebf05d1. Even after I fixed a bunch of rather severe bugs, it was still broken, which makes a game bug all the more likely.

[shmerl]

It should run with mesa-git and RADV_DEBUG=nohiz set.

That helped, thanks! Is this related? https://bugs.freedesktop.org/show_bug.cgi?id=105396

[SirBubbles]

I've got a pretty nasty issue relating to Endrega and Arachas varieties in the game. They look like they've got a whole mass of red triangles coming off them. It looks pretty different to Smerl's picture of Rotfiends, and I don't think I get that problem.

Is this the same bug, or have I found something new? Or maybe screwed something up somewhere? I'm running Ubuntu 17.10, latest dxvk, wine-staging 3.4 and latest NVIDIA drivers on a 970.
(EDIT) In spite of what it says in the DXVK hud, I've updated nvidia drivers since then.

[shmerl]

@SirBubbles : In general, don't test with staging, test with upstream wine first. You don't need staging for TW3.

Check the log. It probably complains about missing stream output support. So it's a missing feature in dxvk that wasn't implemented yet. That's why I'm waiting with playing TW3 for now, until it will work.

[SirBubbles]

@shmerl
There's a lot of spam about stream support, true, but also stuff about invalid blend factor and invalid blend op. Like so:
err: D3D11Device::CreateBuffer: D3D11_BIND_STREAM_OUTPUT not supported
err: D3D11: Invalid blend factor: -1507022288
err: D3D11: Invalid blend op: 32703

Is that pretty much as expected?
(EDIT) Looking back, I see you posted about something similar, yes?

[doitsujin]

@SirBubbles that's the exact same issue as with the rotfiends. Won't be fixed until there's some way to (somewhat) efficiently implement Stream Output in Vulkan.

[shmerl]

Won't be fixed until there's some way to (somewhat) efficiently implement Stream Output in Vulkan.

Is there any way to use the idea from wined3d, or it's too OpenGL specific?

[doitsujin]

Their OpenGL implementation uses transform feedback, Vulkan does not have anything like transform feedback yet. We'll see how things work out though.

[shmerl]

Just for the reference, I tested the patch from https://bugs.freedesktop.org/show_bug.cgi?id=105396 with Vega, and that didn't change it. I guess it's a different bug.

[artivision]

@doitsujin you can implement it like Transform_feedback is implemented upon Gallium. It is almost certain that for every Gallium function there is a Vulkan alternative. You don't need documents for every D3D11 function, just take ideas from the OpenGL state tracker.

[doitsujin]

@artivision Gallium is not Vulkan, and no, Vulkan does not have transform feedback. I don't want to get into details right now, but emulating transform feedback accurately with the only alternative that exists (namely, SSBO stores) is pretty much impossible.

[sebastianlacuesta]

@doitsujin Maybe it's like throwing a bottle with a message to the sea, wait and see. Filing an issue here about transform feedback needed?: https://github.com/KhronosGroup/Vulkan-Docs/issues

[artivision]

@doitsujin neither has Gallium. You need futures equivalent to PIPE_CAP_TEXTURE_BUFFER_OBJECTS. Compute Shaders is a good alternative. For example proprietary AMD does implement D3D11 and OGL4 upon their own Low Level API and IL, but not with any direct equivalents. This is called implementation and not emulation.

[doitsujin]

@artivision Gallium does have native support. They just call it stream output, just like D3D11.

Compute Shaders is a good alternative.

Stream output is part of the rendering pipeline, compute shaders are not, so no, they are not an alternative.

This is called implementation and not emulation.

Feel free to implement it yourself and submit a pull request then.

@sebastianlacuesta I've been talking to other people on other channels about that already. Luckily, DXVK is not the only project that would need it, vkd3d could use it as well for their D3D12 implementation.

[sebastianlacuesta]

Got it, thanks. Keep the good work!

[shmerl]

I'm not sure whether it's TW3 specific or not. But I noticed that there are some lighting / shadows issues. NPC faces sometimes turn almost black. Here is an example:

Look at that old woman's face. It's related to the shadows that are cast from the clouds supposedly.

I can open a separate bug if it's not specific to TW3.

[doitsujin]

Is there an easy way to reproduce this? I've never seen this before.

[shmerl]

I can make a save in such area, and you can check if you get the same issue.

I observed it a bit more closely, and it looks very much like this, except it often happens on faces, rather than walls.

[shmerl]

Save in the area where that happens (version: GOTY/GOG). Walk around and look at NPCs. Weather conditions in the save have clouds and shadows falling on the ground.

tw3_save_dark_surface.zip

[shmerl]

@doitsujin : did you manage to reproduce it, or may be it's Vega specific issue / something to report to Mesa?

[shmerl]

I recorded a small video of the problem (look at merchant's face): https://uploadfiles.io/akbb0

[shmerl]

Tested with newer kernel, in case it's some issue with amdgpu (4.16~rc6). No change really.

[mac1202]

The game freeze after the cutscene where you meet with yennefer at White Orchard.
d3d11.log : https://pastebin.com/teGqZAxQ
dxgi.log : https://pastebin.com/53FSABHw

[doitsujin]

@mac1202 can you please attach a save so i can test that? The logs don't contain anything useful.

@shmerl haven't checked yet, there's a bunch of other issues that i've been trying to solve.

[mac1202]

here the save before the cutscene.
https://drive.google.com/open?id=1bRmczleuQWuLKX0inhjaB_pvtDEyes9A

[volca02]

There's a glitchy water problem present on nvidia GPUs (can't confirm/deny it happening on AMD).

This save file has glitchy water surface, which jumps around erraticaly and "blinks". This happens with both current git (99d9a5df0b6b52f92845529c6cd6cc6c9f92e14c) and v0.40.

When running with DXVK_DEBUG_LAYERS=1, I get these:
witcher3_dxgi.log
witcher3_d3d11.log

vulkanlog.txt stays empty.

In stderr output, I get the following line spammed:
DS(ERROR): object: 0x0 type: 9 location: 2405 msgCode: 11: Cannot submit cmd buffer using deleted buffer 0x0.

[shmerl]

I noticed that the issue with black faces has a counterpart in wined3d + radeonsi. Except instead of blackness, some faces lose features at some distance (like that merchant in the save losing a beard). I didn't see this happening before with wined3d. Setting glsl_correct_derivatives_after_discard=true doesn't help it.

[rohit-n]

I'm also noticing a strange dithering around people's skin that go away when the camera is close enough.

[volca02]

Regarding the debug layer validation error - Cannot submit... - it seems it is actually a bug in the validation layers: https://github.com/KhronosGroup/Vulkan-LoaderAndValidationLayers/issues/2371

[shmerl]

By the way, the water looks OK with latest dxvk and radv (Mesa master).

[shmerl]

Just for the reference, I opened this, may be someone from Khronos can comment about their plans for transform feedback / stream output like support in Vulkan.

[Vash63]

I'm seeing some glitchy water in the shallow moving streams in Blood and Wine on Nvidia 390.48. Happening with current git branch and one I built from last week. It's not bad and looks perfect most of the time but erratically flickers with large black triangles, seems angle/distance dependent.

Also, I'm unable to duplicate the issues with ambient occlusion - seems to work perfectly for me.

[volca02]

@Vash63 since I reported this I tried to sort out this confusion - there was an ambient occlusion related freeze, which isn't present any more. The game is generally unstable, though - with rising FPS the instability gets worse. I use frame limiter at 60 FPS, but still get freezes.

With my recent save I get freeze pretty much guaranteed after no more than a minute. I don't think this is a problem of DXVK, though. It seems it is actually a problem of nvidia driver (nvidia 390.48-2 currently). I got this behavior in other games (directly wine vulkan - f.ex. Doom) before.

[Vash63]

It's actually really stable for me since I locked the framerate at 120FPS with an XML hack. I was getting freezes in the menus without this though as it would get like 900FPS. With the framerate limited to 120 it doesn't crash at all for me.

Same Nvidia drivers, I'm using wine staging 3.5 right now.

[shmerl]

Regarding dark faces, quite similar thing happens with radeonsi, I opened this bug for the reference for stock wined3d (see there also attached save).

[ekianjo]

I'd like to add that the botchling in the red Baron mission (where you look for the deceased baby corpse) is not properly rendered (weird geometry issue) in certain scenes, while it appears correctly rendered in others for some reason.

[shmerl]

Can you post a screenshot please? It could be related to missing stream output.

[volca02]

It almost certainly is, it looks like the same problem as most rotfiends, certain ghouls, some endregas, most earth elementals and some special cases - like the whispering hillock. Not sure if this enlists all cases but it's what I remember seeing broken.

[ekianjo]

missing stream input is something that can be fixed or is that related to a
lack of features from vulkan?

On Wednesday, April 11, 2018, shmerl [e-mail] wrote:

Can you post a screenshot please? It could be related to missing stream
output.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/doitsujin/dxvk/issues/138#issuecomment-380467371, or mute
the thread
https://github.com/notifications/unsubscribe-auth/ACWMmASl0wjFtpW7nOYPqlpaF1m3WWPbks5tng_EgaJpZM4Sj9Z7
.

--
Raphael

[shmerl]

Mmissing stream input is something that can be fixed or is that related to a lack of features from vulkan?

See #135. Hopefully some solution would be found, or Vulkan working group will add some features to the API to make it more feasible.

[shmerl]

@doitsujin : any news on those dark faces? It's interesting, that behavior is somewhat similar to wined3d.

[doitsujin]

Haven't found the time to look into it yet. And with the insane number of bug reports I'm getting per day I don't think I'll be doing that any time soon.

[shmerl]

Apparently that distance dependent rendering behavior is related to some LOD features in the game, and disappearing beards aren't really a bug but happen on Windows too. However dxvk is still doing something incorrectly. It's just that LOD makes it more visible, since that incorrect dark mesh disappears at a close distance.

[ekianjo]

A couple of stuff I noticed as I progress in the game on Linux with DXVK (0.42 currently - nVidia 396.18 drivers, with nextgen compiler off)

• Whispering Hillock: the monster at the heart of the tree is just a mess of polygons.
• When entering Novigrad for the first time you get a cutsene where 2 people are burned on the public square. The woman appears correctly, but the doppler's face is not rendered (while its hair is).
• Rotfiends are still not rendered properly.

Still, overall, graphical glitches are relatively few!

[shmerl]

That is likely related to the lack of stream output, so that's expected (except may be the face issue).

[ekianjo]

yes probably. does not affect much in the game honestly.

On Wednesday, April 25, 2018, shmerl [e-mail] wrote:

All that is likely relate to the lack of stream output.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub, or mute the thread.<
https://ci5.googleusercontent.com/proxy/nagDW2O_tgz6RGVzZPJDepkGTGAK0uGUqzmCOwVBwUHxe8lAkAavzq1v06iz5Qffrcemtd3HhICvAa_m_VBtfws70leCM8PLA6ZGUJyg5XY5JfALlEJ_SQdehmh5dl1LAYo3le8_uUajXd7Menkpvh0edPaJSg=s0-d-e1-ft#https://github.com/notifications/beacon/ACWMmESWyd72vwjr6ZsrACTyeWY53TNdks5tr7CGgaJpZM4Sj9Z7.gif

--
Raphael

[shmerl]

For it's rather immersion breaking, that's why I'm not playing it further for now. Hopefully Khronos Vulkan working group will come up with some good solution to help this.

[volca02]

While listing the things that are visually broken (and all probably related to lack of stream output) - water surface is broken when in storm. It is okay however when the weather is nice.

[shmerl]

@volca02: Can you post a save with a screenshot please? I can test it with radv.

[j3ky]

Endegra and doppler have the same rendering issue of Rotfiend! haha

rm -rf /dev/sda6 (windows partition)

[jarrard]

I can confirm the Endergra Warrior appears to have the triangle-hell _'as I call it'_ rendering problem. Pretty funny.

Also the game seems to have some issues going back into fullscreen under KDE, you can right-click on the witcher3 task and select more-actions than untick fullscreen, then retick. You will then need to alt-tab into witcher 3 for keyboard control, bit of a weird workaround but it works.

PS. I don't have new compiler enabled.

Screenshot for fun. This beast is SERIOUS!

[ekianjo]

I confirm also that the character Hubert Rejk in Witcher 3 appears with the
triangle hell as well, but interestingly only for the lower part of his
body including his hands. his face and upper torso look fine. I am using
DXVK 0.51 along with the "OLD" compiler option at launch.

On Wed, May 16, 2018 at 10:34 PM, jarrard [e-mail] wrote:

I can confirm the Warrier Endergra appears to have the triangle hell as I
call it rendering problem. Pretty funny.

Also the game seems to have some issues going back into fullscreen under
KDE, you can right-click on the witcher3 task and select more-actions than
untick fullscreen, then retick. You will then need to alt-tab into witcher
3 for keyboard control, bit of a weird workaround but it works.

PS. I don't have new compiler enabled.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/doitsujin/dxvk/issues/138#issuecomment-389520186, or mute
the thread
https://github.com/notifications/unsubscribe-auth/ACWMmANFCFLv54NPnBDyNnQsuHDacXJbks5tzCrggaJpZM4Sj9Z7
.

--
Raphael

[shmerl]

FYI: The dark faces bug is gone in the latest dxvk master / Mesa master. I didn't test it in a while, so I don't know when exactly it got fixed and what fixed it.

[shmerl]

Also, there is no need for RADV_DEBUG=nohiz for Vega anymore. radv works fine without it now in TW3.

[SirBubbles]

I'm happy to report that using the latest dxvk, plus wine 3.8 and latest nvidia drivers 396.28? (whoops, it's 396.24) that my issue with Endrega and Arachas types seems to be fixed. I've loaded up an area with examples of both creature types (Poison Arachas in this case) and it seems to be correctly rendered, as do the Endrega.
Awesome! Means I can get back to Witcher 3!

[jarrard]

Will test this out myself since I haven't had a chance since last time. Thought I have only .24 drivers

Yep seems to resolve the issue for me also.

[doitsujin]

@SirBubbles This could be a random side effect of a39b9cb131f453d34f8498090413157565377931, which prevents buffers from containing random garbage.

[shmerl]

Probably related to the above too - rotfiends now became simply invisible vs previous mess of polygons. It looks a bit cleaner immersion wise at least.

[ekianjo]

oh... invisible is not good.

On Friday, May 25, 2018, shmerl [e-mail] wrote:

Probably related to the above too - rotfiends now became simply invisible
vs previous mess of polygons.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/doitsujin/dxvk/issues/138#issuecomment-391951900, or mute
the thread
https://github.com/notifications/unsubscribe-auth/ACWMmOy8x3k4mion1YInbSL6LwaxRk2Xks5t16AAgaJpZM4Sj9Z7
.

--
Raphael

[jarrard]

Pretty sure I have faced some rotfiends and their not invisible for me.

[SirBubbles]

Weird. I just ran into a Temple Guard in Novigrad who has an invisible face. The eyes and mouth still stick around, so it looks really freaky, but the head textures just aren't there.

[jarrard]

Well I'm now using the new beta 396.24.02 which might help. Have not noticed missing faces or anything yet, will keep a eye out for next time.

[shmerl]

No idea, may be it manifests differently in Mesa and Nvidia blob now. Stream output is not implemented so far, so something must be missing still visually either way.

[ximik1987]

witcher3_d3d11.log
witcher3_dxgi.log

gpu: r290x
dxvk: 0.51-0.52
wine: 3.8, 3.9 stading

the image is not visible, but the game has a problem. branches of trees move unnaturally. they do not move with the tact of the wind, but as if they have 2 -3 positions and between them switches animation. without a smooth animation.

[shmerl]

You can record a video of that and upload it somewhere. It can be short just a few seconds in order not to be huge.

[ximik1987]

by the link of the video with the running witch, it is seen that the bug is manifested and in pause mode.

https://mega.nz/#!Pr5nmYTQ!uGrPyzSW32-Ln60x2jUxrvtW3VH9rG2b2uTgC1iwe18

[doitsujin]

Haven't seen that on Polaris. Can you upload a save file where the issue occurs and post your graphics settings?

[ximik1987]

1.tar.gz

any saving, any graphics settings. even a new game does not fix the problem.

[Yardanico]

@ximik1987 screenshot is a bit small :)

[shmerl]

Yeah, save would be useful. I haven't seen such bug with Vega either.

[ximik1987]

save file in tar archive

[shmerl]

Branches look OK to me in this save. So it can be hardware specific issue with radv? Are you using Mesa master?

[ximik1987]

yes, mesa driver 18.2 - git arch

[shmerl]

I suppose it's some problem with radv on older GCN cards. You can report it to Mesa bug tracker.

[ximik1987]

good, thank you.

[dimgel]

Hello doitsujin! Fantastic job, my favorite Skyrim runs ideally. :) But Witcher3 is still unplayable for me: http://dimgel.me/drop/witcher3-wine3.9-dxvk0.53-nvidia396.24.ogv Should I provide more details as you describe in the new issue template?

[dimgel]

UPD. Edited non-working link above. Card is GeForce GTX 1060 6GB.

[doitsujin]

@dimgel Please update your driver to 396.24.02, the issue should be fixed there.

[dimgel]

Thank you for answering! Gonna wait a bit until it (or later one) arrives in gentoo portages.

[pchome]

@dimgel
There is no vulkan beta drivers in portage tree, but you can use your local repo.

• Manually download driver from https://developer.nvidia.com/vulkan-driver and place it to your DISTDIR (see $ emerge --info | grep DISTDIR)
• If you use tools use flag for x11-drivers/nvidia-drivers than unpack nvidia-settings-396.24.tar.bz2 from your DISTDIR, rename directory to match driver version (396.24.02) and pack in nvidia-settings-396.24.02.tar.bz2 in DISTDIR
• Than place content from gentoo's .../x11-drivers/nvidia-drivers into your local repo, copy nvidia-drivers-396.24.ebuild to nvidia-drivers-396.24.02.ebuild and use ebuild nvidia-drivers-396.24.02.ebuild digest command

if you have no local repos you can create one in e.g. /etc/portage/repos.conf/local.conf

[local]
location = /var/db/repos/local
masters = gentoo
auto-sync = no
priority = 9999

# mkdir -p /var/db/repos/local/profiles && echo "local" > /var/db/repos/local/profiles/repo_name
# mkdir -p /var/db/repos/local/metadata && echo "masters = gentoo" > /var/db/repos/local/metadata/layout.conf

[dimgel]

Thank you. I'll probably try it when tired of waiting. :) I have local repo but had problem finding that latest driver on nvidia site, thanks for link.

[ekianjo]

Some new reports from me - on Nvidia and DXVK 0.53, I noticed the following as I progressed in the game:

• The Crones (as young women) appear fine for 2 of them, but the 3rd one is a mess of polygons (when you confront them later in the game).
• Sirens on the sea are a triangles mess as well
• One quest involving a doppler who transformed into a cat shows leads to a scene where the face of the doppler is only made of his eyes and teeth, everything else missing.

[dimgel]

@pchome Hurrah! That worked. :) Thanks again. Also had to delete "amd64-fbsd?" line from copied ebuild - it complains about missing FreeBSD file even though i'm on linux.

[bobwya]

@dimgel Simple... The Nvidia Vulkan beta drivers don't have a BSD release tarball (they are Linux and Windows only), 64-bit only. The modified ebuild will try to download a non-exist BSD tarball to build the ebuild Manifest file.

[ptkato]

The Hound of the Wild Hunt, like the Rotfiend, is also invisible.

[shmerl]

That's expected, they also use stream output.

[ekianjo]

not for me on nvidia.

On Friday, June 8, 2018, Patrick Augusto [e-mail] wrote:

The Wild Hunt Hounds, like the Rotfiends, are also invisible.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/doitsujin/dxvk/issues/138#issuecomment-395592008, or mute
the thread
https://github.com/notifications/unsubscribe-auth/ACWMmKaeSfW84FexLozgFWTX5mpM-qzcks5t6bBfgaJpZM4Sj9Z7
.

--
Raphael

[ptkato]

I just noticed that elementals are also invisible, but not always, sometimes they render just fine.

[dante429]

Witcher 3 runs impressively good on my rig. However, everytime I do a manual save or the game do it automatically, the game freezes. I can hear the audio still. When pressing Ctrl+Alt+Del and the Cancel, Logout, Shutdown Menu comes up, the game comes back to full life again in the background. When I select Cancel everything locks up again. So I have no choice to close witcher 3 and restart it.
What's causing this?

I'm running:
-Arch Linux, Gnome 3, wine-staging 3.10, latest dxvk, Mesa 18.1 with a AMD R9 290X with 4GB VRAM.

[doitsujin]

No idea, doesn't happen on my end.

[dante429]

Found the bug! It's something with Gnome 3. Don't know the exact error, but I installed XFCE, started Witcher 3 and it runs perfect!

[rinalds14]

i found that sometimes in cutscene there is someone with invisible head

[ptkato]

@rinalds14 that's likely to be related to #135.

[rinalds14]

ok

[ionian-wind]

I don't know is it wine problem or it's in dxvk, but i cannot get into Toussaint - game is freezing at load screen

[doitsujin]

Toussaint works fine here.

[rafalcieslak]

I can confirm there is some issue with Toussaint. I was able to complete the main game and first expansion, but as soon as Geralt travels to Toussaint for the first time, the game gets stuck at (or right after) loading screen.

I believe this issue is related to dxvk, because when I disable dxvk and use bare wine with the same configuration & save file, Toussaint does load correctly and game does not freeze (though, obviously, it runs at < 2 FPS).

My configuration is wine-3.11 (Staging), NVIDIA driver 396.24.02, dxvk current master (4d1a70bd89675f9551aefee2a52699791c228078), Witcher 3 GOTY 1.31 (no mods).

To my eye, the logs do not contain anything suspicious:
witcher3_d3d11.log,
witcher3_dxgi.log.

[doitsujin]

Can you provide a savegame? I cannot reproduce the issue and Toussaint loads correctly for me, as I've said before, and I need something to work with...

Also, what exactly are you experiencing? "game gets stuck" is rather ambiguous.

[rafalcieslak]

Here's my savefile: witcher3_savefile_before_toussaint.zip. Speak to knights nearby, agree to take the contract, and they take Geralt to Toussaint.

The exact symptom is that the map loading screen occurs, a background picture of the map being loaded (Toussaint) appears and the 'loading' progress bar moves slowly. Once it reaches 100%, the background fades to black and the 'loading' text with a spinning circle above it are the only visible part of the interface. The circle spins for a second and two, then it stops and the game becomes completely unresponsive, with 0% CPU usage, and stays this way for at least 20 minutes. When any other map loads, this is the exact moment when the loading screen would disappear and game world or a cutscene would become visible again.

I have performed one more test (to check whether the first Toussaint cutscene could be causing problems): I have played through that cutscene up to the first checkpoint in Toussaint with dxvk disabled, saved the game there, re-enabled dxvk and attempted to load it again - the result was identical, a freeze at the map loading screen.

I have attempted to prepare an apitrace, but the game seems to behave differently when running under apitrace (it freezes as soon as any loading screen appears, not just Toussaint, instead of freezing after loading completes), so I assume the resulting apitrace would be misleading.

[doitsujin]

Thanks for the savegame. Unfortunately I still cannot reproduce the problem, it loads just fine:

Not sure what's going on. I also don't see any Vulkan validation errors which could potentially cause such issues. What are your ingame settings?

[ionian-wind]

By the way, I have same problem with game freezing when I launch game in fullscrean mode with dxvk right before game menu - music is playing but no picture. Everything is fine without dxvk or in window mode with dxvk.

I'm playing GOG version if it's matters

[rafalcieslak]

I have tested a bunch of different setting combinations, including these minimal settings. I have also tried removing all config files to reset all settings to defaults. In all cases the problem persists.

Another observation: when the game breaks on the loading screen, it creates a *.crashinfo file (example here) in the bin/x86directory, which describes the exact crash reason (memory access violation) and seemingly useless internal stack-traces of all game threads. This suggests there maybe be some naughty undefined behaviour in play, which would make it hard to reproduce this issue... Naturally, such .crashinfo file does not appear when running vanilla wine w/o dxvk.

Standard validation layers do not report any problems on my end as well.

[MartinPL]

Hmm don't have that problem on my setup, i got to the boss fight. Witcher from Steam, ultra, DXVK - master, wine - wine-esync-staging.m4-pba-SMT-TkG, i7 3770, GTX 1070 (396.24.02), Antergos, GNOME.
BTW i had before lock on loading screen but pressing meta / windows and back to game fix that problem, at least for me.

[triffid]

The rotfiends seem ok for me, however Sirens, Ekhidnas, the hands of the novigrad coroner and the head of the giant cat doppler are invisible.

Curiously, the sirens and ekhidnas become visible when they're killed, and other similar creatures like harpies are visible.

I assume this is all related to D3D11_BIND_STREAM_OUTPUT discussed in #135 .

Using nvidia 396.45, wine 3.13, dxvk 0.63 binary release on Mint.

Setting up dxvk and vulkan took a little finagling, but well worth it! The jump in FPS was _amazing_, from ~5fps to basically smooth with mild choppiness in novigrad. Keep up the good work!

[Veske]

It seems that after rain, NPC's have some of their parts flying around randomly while generating weird rays over the screen.

In this example I have an NPC sitting who looks bald... actually his hair is floating above him.

DXVK 0.64
Mesa 18.1.5-1
RX 480

I can get rid of this only by restarting the game.

[doitsujin]

Can this be reproduced regularly and are you sure that it's not a game bug (there are a few of those as well)?

I've never seen this before.

[ptkato]

Is it possible that dxvk makes the game behave differently? I often notice hairs floating around weirdly, like, Yen's/Geralt's hairs move all over the place, and there's barely any movement to the character or wind at all; I also feel that it rains all the time in Velen, like, non-stop, that didn't happen in native.

[Veske]

Playing casually right now. Weird things start happening after a couple of hours of playing. Right now I have a wall in front of me wherever I walk. This appeared near the mountain with the huge tree (Crones cook people there)

Guess I will let you know if I find a more reliable way to reproduce this stuff.

[Veske]

@doitsujin Also it happens sometimes that the game freezes completely, cannot alt+tab out of it nor go to another TTY to close the wine server. I cannot even use Alt+SysRq+R to get input back. Only thing that works is Alt+SysRq+B to reboot the system.

This is what Gnome system logs tool seems to show for the previous boot. GPU fault sounds like something that could cause such an issue. I have only encountered such hard freezes with Vulkan games.

Let me know if I can produce some better logs for you somehow... and maybe you also know of a trick on how to recover from such case without reboot?

[doitsujin]

@Veske which version of LLVM are you using? Is there a (somewhat) reliable way to reproduce it?

System freezes are indeed a side effect of GPU faults, and they can be caused by driver bugs or DXVK bugs, so finding a way to reproduce this is important.

and maybe you also know of a trick on how to recover from such case without reboot?

No, that's impossible.

[shmerl]

No, that's impossible.

At least for now, until GPU reset will be implemented properly. It is quite annoying at present.

It seems that after rain, NPC's have some of their parts flying around randomly while generating weird rays over the screen.

@Veske: A save with such conditions would be useful.

[Veske]

pacman -Q llvm-libs 
llvm-libs 6.0.1-2

This is the LLVM package I assume?

It's really hard to give reliable way to do it. I am just able to produce this in a couple of hours gaming session. Saving when the issue happens and then loading back in after game restart means the issue is gone. Could be that alt-tabbing a lot helps to the cause of such anomalies also?

Today I managed to play until such a scene appeared:

witcher3_d3d11.log
witcher3_dxgi.log

[doitsujin]

LLVM 6 is known to be notoriously broken with DXVK in various ways, that includes causing GPU hangs in games that otherwise work fine. Might be worth testing with LLVM 7-svn, but you'll have to build that yourself, and you'll also have to build Mesa against it.

[Veske]

Alright, sounds like something I have little idea of. Will see if I want to go down that rabbit hole. You have any idea when LLVM 7 it self will be released?

Nvm. found it on their homepage.. 5th of September :)

[shmerl]

For some distros, development llvm snapshots are available as packages, so you don't need to build llvm itself at least (which can be quite annoying). But you'd still need to build Mesa against that llvm.

I did that for Debian testing, and it's not that hard.

[dimgel]

Starting with DXVK 0.70, I face significant FPS drop. Cannot say in numbers (DXVK_HUD stopped working too), but Geralt now runs really slowly. Face it with 0.70 / wine 3.13 / nvidia 396.*, 0.71 / wine 3.15 / nvidia 396.54. Had to revert to DXVK 0.65 (with manual deletion of files DXVK 0.70 installer copied to wine dir) few times.

[dimgel]

P.S. GeForce GTX 1060 6GB

[Capostrophic]

You didn't install 0.70 properly so the game runs on pure Wine. No issues on the side of DXVK.

[dimgel]

Hm, haven't thought about that. But how to install it properly then? I uninstalled 0.65 before running 0.71 installer, and that installer didn't complain about anything: it copied files, performed library overrides (winecfg shows 7 files in "Libraries" tab compared to 1 file for 0.65). Where else can I look?

[dimgel]

Solved: had too old winetricks, upgraded to latest available. Sorry for disturbing.

[antkir]

Can confirm Toussaint doesn't load using Nvidia GTX 760 with the latest 396.54.05 driver.
Using __GL_NextGenCompiler=0 makes it load without any issues, but the graphics is obviously corrupted.

[AccountOneOff]

Can confirm Toussaint doesn't load using Nvidia GTX 760 with the latest 396.54.05 driver.
Using __GL_NextGenCompiler=0 makes it load without any issues, but the graphics is obviously corrupted.

@antkir Works here with my 960 and 396.54.05

[triffid]

Toussaint seems to work fine on my GTX1060 with both 396.45 and 396.54,
perhaps the nvidia suite is using a different shader compiler (or different
codepath) for older cards, or the problem is elsewhere in your system?

Add earth elementals to invisible things

On Mon, 17 Sep 2018 at 08:07, cxf [e-mail] wrote:

Can confirm Toussaint doesn't load using Nvidia GTX 760 with the latest
396.54.05 driver.
Using __GL_NextGenCompiler=0 makes it load without any issues, but the
graphics is obviously corrupted.

@antkir https://github.com/antkir Works here with my 960 and 396.54.05

[image: screenshot from 2018-09-16 17-04-02]
https://user-images.githubusercontent.com/15083452/45602427-f1744600-b9d2-11e8-85a3-7db07956251b.png

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/doitsujin/dxvk/issues/138#issuecomment-421857229, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AAKBGucCUPFvpbX4zc8bUJyEgf6w8H4qks5ubufYgaJpZM4Sj9Z7
.

[icemanx7]

So the performance of TW3 in wine and dxvk is perfect but it keeps freezing up my entire system 4min when I continue from my current save. I have force restart my pc to get out of it. And played past that point on my windows partition directly after I tried switching the dxvk version around so its definitely not a hardware problem. Is there anyway to diagnose why the freeze is happening? My system specs are:

OS tried: Arch Linux and Ubuntu 18.04 lts

CPU: AMD 2400g

RAM : 16G

GPU: GTX 1060 6GB

GPU Drivers: 410.57-2

Resolutions Tried: 1920x1080, 1400x900

DE Tried: KDE, Gnome and XFCE4

Using the Lutris GOG TW3 wine script and tried dxvk 0.72 and dxvk 0.80.

[jarrard]

CPU: AMD 2400g

These APU's have been known to still have stability issues under Linux. I would try and get the bleeding edge version of kernel for your system (4.19/20 release candidate or something). ON TOP of this the 410.57-2 drivers have had known crash issues in them so downgrade to 396.54.05/6 version.

[icemanx7]

CPU: AMD 2400g

These APU's have been known to still have stability issues under Linux. I would try and get the bleeding edge version of kernel for your system (4.19/20 release candidate or something). ON TOP of this the 410.57-2 drivers have had known crash issues in them so downgrade to 396.54.05/6 version.

I'll the try bleeding edge kernel because When I tested on Ubuntu I used the 396.54.05/6 version and it still froze.

[jarrard]

Might pay to run a temp logger or something in the background for your GPu/CPU also just to be sure because sometimes Linux does stress hardware out in such a way that it becomes unstable compared to Windows (The Ryzen segfault issue for example).

Also there might be some system logs able to give you a idea on what happened at point of crash, on top of that you can attempt to TTY into your system via another device connected on the network and see if it really is a FULL system crash or not.

[icemanx7]

Thanks @jarrard I will checkout my temps although I highly doubt its that since I have a beefy cooling cooling setup and I stress tested both the cpu and gpu where the temps for both didn't go beyond 65c.

Also there might be some system logs able to give you a idea on what happened at point of crash,

Which logs would that be because I have only been checking the logs with gets printed out with the journal b-1 command

on top of that you can attempt to TTY into your system via another device connected on the network and see if it really is a FULL system crash or not.

Oh this is a good idea I can try this. Thanks

[icemanx7]

CPU: AMD 2400g

These APU's have been known to still have stability issues under Linux. I would try and get the bleeding edge version of kernel for your system (4.19/20 release candidate or something). ON TOP of this the 410.57-2 drivers have had known crash issues in them so downgrade to 396.54.05/6 version.

@jarrard
I tested with running kernel 4.19-rc5 and the same froze happened. Although it took much longer to froze up than on the other kernels. And yes its a system freeze since I can't ssh into the machine. It's weird since Metal Gear Solid 5 Phantom Pain works amazingly.

[doitsujin]

@icemanx7 have you tried running the game with an older driver (396.54)? Wouldn't be the first time that people report issues with the 410 drivers, although a GPU hang seems new.

[icemanx7]

@icemanx7 have you tried running the game with an older driver (396.54)? Wouldn't be the first time that people report issues with the 410 drivers, although a GPU hang seems new.

I'll try that and report back if there is any new development.

[edmondo]

Witcher 3 with latest DXVK and wine 3.18 + radv patched for VK_EXT_transform_feedback looks good. Thanks for the awesome job.

info:  Game: witcher3.exe
info:  DXVK: v0.90+
warn:  OpenVR: Failed to locate module
info:  Enabled instance extensions:
info:    VK_KHR_get_physical_device_properties2
info:    VK_KHR_surface
info:    VK_KHR_win32_surface
info:  AMD RADV POLARIS10 (LLVM 8.0.0):
info:    Driver: 18.2.99
info:    Vulkan: 1.1.70
info:    Memory Heap[0]: 
info:      Size: 3840 MiB
info:      Flags: 0x1
info:      Memory Type[0]: Property Flags = 0x1
info:    Memory Heap[1]: 
info:      Size: 256 MiB
info:      Flags: 0x1
info:      Memory Type[2]: Property Flags = 0x7
info:    Memory Heap[2]: 
info:      Size: 4096 MiB
info:      Flags: 0x0
info:      Memory Type[1]: Property Flags = 0x6
info:      Memory Type[3]: Property Flags = 0xe
warn:  DXGI: VK_FORMAT_D24_UNORM_S8_UINT -> VK_FORMAT_D32_SFLOAT_S8_UINT
info:  Enabled device extensions:
info:    VK_EXT_shader_viewport_index_layer
info:    VK_EXT_transform_feedback
info:    VK_EXT_vertex_attribute_divisor
info:    VK_KHR_dedicated_allocation
info:    VK_KHR_descriptor_update_template
info:    VK_KHR_get_memory_requirements2
info:    VK_KHR_image_format_list
info:    VK_KHR_maintenance1
info:    VK_KHR_maintenance2
info:    VK_KHR_sampler_mirror_clamp_to_edge
info:    VK_KHR_shader_draw_parameters
info:    VK_KHR_swapchain
info:  DXVK: Read 896 valid state cache entries
info:  DXVK: Using 9 compiler threads

[shmerl]

Thanks for wrapping things up quickly after the spec was ready!

[shmerl]

Transform feedback patches didn't yet land in Mesa master, so if you want to build radv with them manually, download them, like this for example:

#!/bin/bash

patches=(
https://patchwork.freedesktop.org/patch/256575/raw/
https://patchwork.freedesktop.org/patch/256578/raw/
https://patchwork.freedesktop.org/patch/256579/raw/
https://patchwork.freedesktop.org/patch/256576/raw/
https://patchwork.freedesktop.org/patch/256577/raw/
https://patchwork.freedesktop.org/patch/256585/raw/
https://patchwork.freedesktop.org/patch/256582/raw/
https://patchwork.freedesktop.org/patch/256581/raw/
https://patchwork.freedesktop.org/patch/256583/raw/
https://patchwork.freedesktop.org/patch/256580/raw/
https://patchwork.freedesktop.org/patch/256586/raw/
https://patchwork.freedesktop.org/patch/256587/raw/
https://patchwork.freedesktop.org/patch/256584/raw/
https://patchwork.freedesktop.org/patch/256588/raw/
)

for patch in ${patches[@]}; do
    wget --content-disposition $patch
done

[shmerl]

@edmondo: how did you get this part?

info: VK_EXT_transform_feedback

Did you use some patches for Wine for it? It's not showing up for me (Wine 3.17.0, latest dxvk master, Mesa master + patches).

[shmerl]

Tried building Wine master, still getting:

err:   D3D11: CreateGeometryShaderWithStreamOutput:
  Transform feedback not supoorted by device

UPDATE: yep, Wine needs patching. You can take needed updates here:

• https://raw.githubusercontent.com/ValveSoftware/wine/266d11ab06d55dc4ca34e5842cc3ab945b6e9b0b/include/wine/vulkan.h
• https://raw.githubusercontent.com/ValveSoftware/wine/266d11ab06d55dc4ca34e5842cc3ab945b6e9b0b/include/wine/vulkan_driver.h
• https://raw.githubusercontent.com/ValveSoftware/wine/266d11ab06d55dc4ca34e5842cc3ab945b6e9b0b/dlls/winevulkan/make_vulkan
• https://raw.githubusercontent.com/ValveSoftware/wine/266d11ab06d55dc4ca34e5842cc3ab945b6e9b0b/dlls/winevulkan/vulkan_thunks.c
• https://raw.githubusercontent.com/ValveSoftware/wine/266d11ab06d55dc4ca34e5842cc3ab945b6e9b0b/dlls/winevulkan/vulkan_thunks.h
• https://raw.githubusercontent.com/ValveSoftware/wine/266d11ab06d55dc4ca34e5842cc3ab945b6e9b0b/dlls/winevulkan/vulkan.c
• https://raw.githubusercontent.com/ValveSoftware/wine/266d11ab06d55dc4ca34e5842cc3ab945b6e9b0b/dlls/winevulkan/vulkan_private.h
• https://raw.githubusercontent.com/ValveSoftware/wine/266d11ab06d55dc4ca34e5842cc3ab945b6e9b0b/dlls/winevulkan/winevulkan.spec

Now, I'm getting:

info:    VK_EXT_transform_feedback

[shmerl]

Rotfiends are working, and here is a patch for Wine master if anyone needs it (hopefully Wine developers will catch up to it upstream soon).

winevulkan_patch.zip

[Tk-Glitch]

@shmerl Doitsujin has been providing the required patch here : https://github.com/doitsujin/dxvk/issues/695

And I've been providing it here : https://github.com/lutris/wine/commit/3110fff2ec1e0d8b9862706f7e8dad34fbffc456 and there : https://github.com/Tk-Glitch/PKGBUILDS/blob/master/wine-tkg-git/winevulkan_transform-feedback_ext.patch

No need to reinvent the wheel :D

[shmerl]

Good, thanks, I didn't see that issue, so I did it myself :)

[shmerl]

@Tk-Glitch : why does Proton's fork have so many differences in winevulkan though (besides just for transform feedback)?

[aqxa1]

@shmerl I suspect because they forked wine-3.7 and winevulkan wasn't finished until wine-3.10.

[Tk-Glitch]

@shmerl The fullscreen hack is a pretty big part of it.

[antkir]

Toussaint now loads just fine with GTX 760, DXVK 0.90 and 396.54.09 driver, so it seems like this issue may have been fixed as well.

[doitsujin]

That's good to hear, thanks for the heads-up.

[Leopard1907]

@rafalcieslak

Can you try loading Toussaint with 396.54.09 drivers?

[rafalcieslak]

I confirm I can no longer reproduce the issue I experienced when loading Toussaint :+1:

[doitsujin]

Great, thanks for reporting back.

Since all issues should be fixed, I'm going to close this one. If you encounter any new issues, please open a new bug report.

[a-pushkin]

Witcher GOG GOTY seems to be broken on
Ubuntu 18.04 NVidia 396.54.9 Vulkan 1.1.85 DXVK 0.92 Wine 3.20 - wineserver crashes after the intro videos
Rolling back Wine to 3.19 fixes the issue - game runs no problem (at least the initial Kaer Morhen level.
I also tried rolling back to Wine 3.10 (earliest that supports DXVK) - game runs but NVidia hairworks doesn't work - Geralt gets has a short haircut.

[doitsujin]

Wine 3.10 doesn't support the vulkan extension for transform feedback, but that should be irrelevant now since recent versions support it.
Wine 3.20 is known to be broken with this game, but that's not a DXVK bug.

[shmerl]

Witcher GOG GOTY seems to be broken on
Ubuntu 18.04 NVidia 396.54.9 Vulkan 1.1.85 DXVK 0.92 Wine 3.20 - wineserver crashes after the intro videos

That's not a dxvk bug. Use Wine master for now, it's already fixed there. Wine 3.21 will have the fix.

See here and here.

[dimgel]

Not sure who's this old bug is, but: from the top of Beauclair Palace I usually cannot see Mont Crane Castle's LOD.

To reprodure reliably, have to walk around Toussaint for some time (for many models and textures to get loaded, I believe). These steps worked for me: started at Ardaiso Quarry, fast travelled to my house, walked around it, then fast travelled to Gran'Place and from there on foot to top of the castle, then turn back and see: http://dimgel.me/drop/witcher3-castle-lod-bug.ogv Don't know how to take screenshots, so I recorded video, haha! :)) I stand where I talked to Guillaume and Vivienne.

LODs get visible after reload from save (also demonstrated in video).

Seen on wine-vanilla-3.16...3.19 at least, wine-staging-3.19; DXVK is 0.92 and before, nvidia-drivers 396.54.09 and before. GeForce GTX 1060 6GB.

[shmerl]

@dimgel: Can you post a save please?

[dimgel]

No problem:

http://dimgel.me/drop/witcher3-save/QuickSave_1286f6_7e256000_4b95f6f.sav
http://dimgel.me/drop/witcher3-save/QuickSave_1286f6_7e256000_4b95f6f.png

Although no idea how this can help: as I demonstrated, LOD is there after save is loaded.

Mods: auto loot, manual trophy equip, sort everything, stack your items, clean merchant reset, no story (on load screens).

[K0bin]

Thats probably a game bug.

[shmerl]

I'll give it a try, but yeah, try finding someone to test it on Windows.

[dimgel]

If I could I would already. :)

[shmerl]

You can try here. Some might actually be familiar with this, if it's indeed a game bug.

[shmerl]

I tried the save, I can see the castle. But if it's indeed something to do with LOD, it's a game issue.

[aldonogueira]

As requested by @shmerl here is the save for the armless dwarf. Tell me if you have any trouble.
armless_dwarf_save.zip

Original comment with screenshot:
https://github.com/ValveSoftware/Proton/issues/607#issuecomment-440835134

[shmerl]

@aldonogueira: Hm, what version are you using? It told me "the save can't be loaded, because it was made by the newer version of the game". I'm using GOG GOTY 1.31. I had no idea there were newer versions. At least GOG doesn't list anything newer.

[aldonogueira]

It says 1.32 in the game. It's not a GOTY edition. It's this one here
https://store.steampowered.com/app/292030/The_Witcher_3_Wild_Hunt/

Here is another save near the dwarf, but was created using version 1.31
near_armless_dwarf.zip

To get there, follow the instructions in this page:
http://www.gosunoob.com/witcher-3/bank-location-in-novigrad/

[shmerl]

Interesting, I had no idea it had 1.32 version, I guess it's different numbering for non GOTY releases? Though I have both on GOG (regular and GOTY) and they are both numbered as 1.31. Let me try your second save.

[shmerl]

@aldonogueira: I found the bank. And I don't see this bug. Sapphire Pulse Vega 56, Mesa master / llvm 8.0, all settings on max, hairworks off.

So it could be Nvidia blob specific.

Also, try yourself using latest Wine+dxvk instead of Proton to narrow down the differences, not sure if it would make a difference though.

[Vash63]

1.32 was released about a month ago on Steam, added simplified Chinese.
Maybe they didn't patch the GOG version? Wouldn't be the first time,
Witcher 2 was frequently ahead on Steam.

On Sun, 9 Dec 2018, 23:32 shmerl <[email protected] wrote:

@aldonogueira https://github.com/aldonogueira: I found the bank. And I
don't see this bug (Sapphire Pulse Vega 56, Mesa master / llvm 8.0:

[image: tw3_dxvk_dwarf]
https://user-images.githubusercontent.com/310927/49703848-51e0e100-fbd8-11e8-9dd8-9b77f441037f.jpg

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/doitsujin/dxvk/issues/138#issuecomment-445578193, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AC8WwbC6sKYHo5mX8iPfSwTNiXLEEkMNks5u3Y9hgaJpZM4Sj9Z7
.

[shmerl]

Surprising, given that GOG is their own store. But I suppose language addition shouldn't change the actual renderer, strange that it couldn't load older save. Anyway, this could be an Nvidia bug.

[shmerl]

Comparing those two images, it's interesting how different they look. Texture sharpness especially. For some reason my looks less sharp on the walls, but more detailed on the character.

Here is the one from the blob:

Also take a look and candles on the small table on the right.

[aldonogueira]

Why are you talking about nvidia? I'm using a AMD RX 470 on Ubuntu 18.14.1, padoka PPA drivers, which gives me Mesa 19.0.0-devel - padoka PPA and LLVM 8.0.0. Proton is 3.16-4

The settings for the image above is medium, which means that focus-something effect is not active (that thing that blurs the scene behind the characters)

[shmerl]

Ah, I got it mixed up then, sorry. It's interesting then that there is a difference between Polaris and Vega. Did you try it on max settings?

My Mesa is 19.0.0-devel (git-cc6a5e937b), built with llvm 8.0.
Kernel: 4.19.5
OS: Debian testing.

I usually build Mesa myself, pulling latest llvm snapshot from the Debian targeted llvm repo.

[shmerl]

Btw, if this is useful to anyone, I figured out what's going on with the version. GOG did release 1.32 but only through Galaxy. Standalone installers are still at 1.31 (they usually lag behind, but I didn't expect them to lag that much).

lgogdownloader --galaxy-show-builds 1495134320
0: Version 1.32 - 2018-11-06T10:32:48+0000 (Gen 2)
1: Version 1.31 (A) - 2017-02-08T13:55:27+0000 (Gen 2)
2: Version 1.31 - 2016-08-25T08:18:33+0000 (Gen 2)

[aldonogueira]

Ah, I got it mixed up then, sorry. It's interesting then that there is a difference between Polaris and Vega. Did you try it on max settings?

Sorry for the delay. I tested on various settings, from low to the highest, but made no difference to the visibility of the arm.

By the way, were you able to load on 1.32 that first game save I uploaded? I don't remember seeing this bug before. Maybe they changed something in the dwarf from 1.31 to 1.32...

[shmerl]

I'm still waiting for GOG to update their standalone version to 1.32. I'll try it then, though it's annoying since it reportedly breaks some mods despite being just a language pack update.

[Pangaea666]

Recently bought and started playing the game (on Linux Mint). Most things are working very well, but I ran into a graphical glitch issue with Keira Metz at the end of the "tower mice" quest sequence. In the screenshots she is standing outside the tower, after walking down the stairs, during a cutscene. I've not seen something similar anywhere else, before or after. At the time this issue also reproduced, because I reloaded to make a non-dumb choice, and the same thing happened again.

Another minor thing I'm seeing is a blue-ish colour at the very edges of the screen, during or at the end of using Witcher senses (I've ticked for colour blind, if that matters). Don't have a screenshot of that, but can and grab one if wanted.

Have also had some sound issues (mostly crackling), but this appears to be resolved by adding tsched=0 to /etc/pulse/default.pa

[doitsujin]

You could at least mention which graphics card and driver you're using....

[Pangaea666]

Yeah, sorry about that. I'm using GTX 770 with the 415.27 drivers. Been posting about it in 3-4 places, so forgot some key details this time. Here is a savegame where the glitch happens, at least on my end: https://megaupload.nz/r683t4vdbf/Keira-bug-TW3.tar_gz

Here are some more details about my system (whitespace may get swallowed, but I hope it's somewhat readable nevertheless):

inxi -Fxz
System:
  Host: thunderstruck Kernel: 4.18.0-15-generic x86_64 bits: 64 
  compiler: gcc v: 7.3.0 Desktop: Cinnamon 4.0.9 
  Distro: Linux Mint 19.1 Tessa base: Ubuntu 18.04 bionic 
Machine:
  Type: Desktop Mobo: MSI model: Z87-G43 (MS-7816) v: 1.0 serial: <filter> 
  BIOS: American Megatrends v: 1.11 date: 05/09/2015 
CPU:
  Topology: Quad Core model: Intel Core i5-4670K bits: 64 type: MCP 
  arch: Haswell rev: 3 L2 cache: 6144 KiB 
  flags: lm nx pae sse sse2 sse3 sse4_1 sse4_2 ssse3 vmx bogomips: 27200 
  Speed: 3640 MHz min/max: 800/3800 MHz Core speeds (MHz): 1: 3584 2: 3669 
  3: 3777 4: 3708 
Graphics:
  Device-1: NVIDIA GK104 [GeForce GTX 770] vendor: Micro-Star MSI 
  driver: nvidia v: 415.27 bus ID: 01:00.0 
  Display: x11 server: X.Org 1.19.6 driver: nvidia 
  unloaded: fbdev,modesetting,nouveau,vesa resolution: 1920x1080~60Hz 
  OpenGL: renderer: GeForce GTX 770/PCIe/SSE2 v: 4.6.0 NVIDIA 415.27 
  direct render: Yes 
Audio:
  Device-1: Intel 8 Series/C220 Series High Definition Audio 
  vendor: Micro-Star MSI driver: snd_hda_intel v: kernel bus ID: 00:1b.0 
  Device-2: NVIDIA GK104 HDMI Audio vendor: Micro-Star MSI 
  driver: snd_hda_intel v: kernel bus ID: 01:00.1 
  Sound Server: ALSA v: k4.18.0-15-generic 
Network:
  Device-1: Realtek RTL8111/8168/8411 PCI Express Gigabit Ethernet 
  vendor: Micro-Star MSI driver: r8169 v: 2.3LK-NAPI port: d000 
  bus ID: 03:00.0 
  IF: enp3s0 state: up speed: 1000 Mbps duplex: full mac: <filter> 
Drives:
  Local Storage: total: 4.15 TiB used: 2.44 TiB (58.9%) 
  ID-1: /dev/sda vendor: Kingston model: SH103S3240G size: 223.57 GiB 
  ID-2: /dev/sdb vendor: Seagate model: ST3320620AS size: 298.09 GiB 
  ID-3: /dev/sdc vendor: Seagate model: ST2000VX000-1CU164 size: 1.82 TiB 
  ID-4: /dev/sdd vendor: Samsung model: SSD 860 EVO 1TB size: 931.51 GiB 
  ID-5: /dev/sde vendor: Samsung model: SSD 860 EVO 1TB size: 931.51 GiB 
Partition:
  ID-1: / size: 219.06 GiB used: 22.64 GiB (10.3%) fs: ext4 dev: /dev/sda1 
  ID-2: /home size: 915.89 GiB used: 152.48 GiB (16.6%) fs: ext4 
  dev: /dev/sdd1 
Sensors:
  System Temperatures: cpu: 52.0 C mobo: 27.8 C gpu: nvidia temp: 76 C 
  Fan Speeds (RPM): N/A gpu: nvidia fan: 48% 
Info:
  Processes: 214 Uptime: 5h 38m Memory: 7.74 GiB used: 4.82 GiB (62.4%) 
  Init: systemd runlevel: 5 Compilers: gcc: 7.3.0 Shell: bash v: 4.4.19 
  inxi: 3.0.27 

This was with the game running in the background, so some values are on the high side.

[doitsujin]

Thanks for posting the savegame. This is a game bug, same thing happens on Windows 10.

[jrugia]

I tested out and tried to replicate @Pangaea666 issue on a few of my systems, can confirm this is a game bug. This Video: https://www.youtube.com/watch?v=idadlPlGBLY @ around 2:59 shows the same bug on Windows.

[kergalym]

I have issue with disappeared texture. More on https://github.com/doitsujin/dxvk/issues/983

[shmerl]

By the way, I finished the Towerful of Mice quest, and didn't have that Keira bug (Mesa master / Vega 56).

However after recently building dxvk (master), I experienced already twice a weird severe hang (the motherboard indicated something was seriously weird - never saw such error on the LED before), both times around Lindenvale. That never happened in the past. Not sure if it's related to latest dxvk changes or not. I'll try rebuilding master again now.

Customized settings that I have:

dxgi.numBackBuffers = 4
dxvk.useEarlyDiscard = True

[jarrard]

Just roll back to 1.0.3 and test, lutris allows easy method of doing this, however keep in mind the bins included in lutris are NOT compiled on your system, if you do compile yourself then you could be getting issues with compiling process specific to your setup.

[shmerl]

I can build older tags, no problem. But this bug is somewhat random, so it won't be a very conclusive test. Lutris build isn't any better than any other build in the sense of "specific setup" issues.

[shmerl]

I rebuilt Mesa master, and it's not hanging anymore. May be it was related to this:

https://gitlab.freedesktop.org/mesa/mesa/commit/6ac10e07c2d6d7740f09a4e53c6b2fc88f4ecadd

[doitsujin]

Witcher 3 does not use Predication (and DXVK doesn't support it anyway).

[Pangaea666]

I'll preface by saying I'm not sure if this is DXVK-related or something else, but will post about it just in case it might be.

I'm on Linux Mint 19.1 Cinnamon with a Nvidia GTX 770 card. A week or so ago the kernel got upgraded from 4.18.0-20 to 5.0.0-15 (or maybe it was ...-14 back then). After this upgrade I noticed the game started to crash more. The only issue I had before was the sound sometimes starting to crackle. I finished the game and soon later started over with a NG+. By now I had also upgraded the driver from 418.56 to 430.14 (via graphics-drivers PPA and Driver Manager).

Within a few hours of playing the game from the beginning, it had crashed about five times, and it kept crashing fairly regularly. Typically it would crash when transitioning from one cutscene to another, or from a cutscene and back to gameplay. But it also crash when using the Fast Travel system, or sometimes during normal gameplay. It crashes once when I was reading notes from a Notice Board. One time everything actually hard froze and I had to restart with a "Kill" shortcut (Ctrl-Alt-X in my case). Apart from the latter total freeze, when the game crashed it would instantly CTD.

Since then I rolled back to the 4.18.0-20 kernel. Had a kertuffle with Mint because I got an error message about "software rendering mode", but after puring Nvidia drivers in terminal and re-installing them things sorted itself out. I have played Witcher 3 for at least 10 hours since the kernel downgrade, and have had no more crashes.

You are better positioned to tell whether this _may_ be DXVK-related. I assume there could be something in the 5.0 kernel that have upset the game on my end, but that's just second guessing. If you need more info please holler, and I'll try to provide it.

As an aside, I also tried out the 1.2.1 version of DXVK, but the game felt more sluggish, so I rolled back to 1.0.3. The game has been installed using the following script: https://www.gog.com/forum/the_witcher_3_wild_hunt/the_witcher_3_for_linux, kindly created by a chap called adamhm.

[doitsujin]

Sounds like a driver problem to me, never had those issues.

That said, make sure to close everything that eats VRAM (web browser!) before playing, and don't use Ultra texture quality on a 2GB card. Nvidia doesn't really like running out of VRAM on Linux.

[Pangaea666]

Don't think so as the game crashed on the same driver while on the 5.0 kernel, and it hasn't crashed since I downgraded to 4.18 - and I've played at least 15 hours since. Seems a bit too coincidental to me. But I don't know exactly what the problem was, and maybe it has something to do with the mix of hardware and drivers. Thanks for the tip about the web browser. That said, I've had it open with a kajillion tabs after the change to 4.18 kernel as well, and the game hasn't crashed yet. Knock on wood and all that :P

[lieff]

@Pangaea666 Browser do not loads all tabs on load, only when you actually switch to them. So difference still possible, you can check actual vram usage using nvidia-smi or nvtop.

[shmerl]

Vespula has missing textures (Broken Flowers quest):

vespula_save.zip

Is it a game or dxvk / radv bug? Attached a save above.

DXVK: v1.2.1-4-g1e0fe36c
Mesa/radv: 19.2.0-devel (git-2285b93032) (LLVM 9.0.0)

[K0bin]

That's not a texture issue, it's geometry issue. It's probably a game bug though. We encountered something similar in the Tower of Mice quest which also turned out to be a game bug.

[shmerl]

Apparently, it's a known bug:

Nitpicker's patch mod fixes it: https://www.nexusmods.com/witcher3/mods/2117/

[Pangaea666]

Probably the same-ish deal again, that it's a game issue instead of DXVK, but have others noticed for instance trees and bushes coming through walls?

The game is in a pretty good spot when we can raise minor nitpicks like this instead of e.g. Geralt looking like a blob or the game game crashing at the sight of mobs :)

[Pangaea666]

This looks a little odd?

Came out a door from the theatre location and hit Escape to go into Options (there was a lighted torch on the left wall as I did this, with bright normal lighting). Then the menu bar to the left looks like that. Go back into the game, and Geralt plus other NPCs on the screen are flickering on and off. Never seen it before and perhaps just a very odd occurrence. Saved and went back into the game after a restart, and as expected things look right then.

[shmerl]

The menu looks like you fell back to wined3d. Are you still using dxvk? I remember menu looking like that a long time ago in wined3d, when dxvk wasn't out yet.

[Pangaea666]

Yeah, I'm on DXVK, otherwise everything would have crashed basically (I have accidentally tried without DXVK before). Restarted the game and everything has been fine now. Not sure what happened there, but possibly some outlier of one type or another.

[shmerl]

I encountered a distortion (running squares - zoom in on small trees in the middle in the background) which only shows up in the vicinity of Vegelbud Residence:

distortion_save.zip

dxvk master / Mesa master + llvm 9.0 nightly snapshot.

Any idea if it's an in-game issue, or something with dxvk?

[doitsujin]

Driver bug. Maybe AMD decided to push the optimization that introduced this exact issue to AMDVLK into mainline LLVM after fixing it in AMDVLK itself, who knows.

[shmerl]

Hm. Is it worth reporting it to Mesa or to llvm?

[doitsujin]

Mesa, but please attach a Renderdoc capture if you file a report.

BTW I can't reproduce this on mesa-git and llvm-git (both built a few minutes ago).

[shmerl]

My llvm came from Debian repo snapshot, so it was built on June 04 last:
http://apt.llvm.org/unstable/pool/main/l/llvm-toolchain-snapshot/

May be since then something has changed. I'll wait until next one appears, and re-test.

[Oschowa]

This was already reported and is fixed in llvm master:
https://bugs.freedesktop.org/show_bug.cgi?id=110861

[shmerl]

Just rebuilt Mesa with newer llvm snapshot, and can confirm that it fixes that bug.

[shmerl]

Latest Mesa master build cases a hang for TW3. GPU resets, but the weird errors after that show something related to audio in the kernel.

I'll try bisecting Mesa to see what broke it. It could be also latest llvm snapshot is messed up. I have one from June 26.

[shmerl]

Yeah, for sure llvm. I went back quite a bit in Mesa git, and it's still hanging. Trying to build now with llvm 8 (too bad Debian llvm repo only has one snapshot for specific version).

UPDATE: llvm bug confirmed. Just built Mesa master with llvm 8, and no hang.

UPDATE2: Opened Mesa bug: https://bugs.freedesktop.org/show_bug.cgi?id=111027

[doitsujin]

It's a known issue, r364296 (or 08e8cb576021ce493329bbc9fa29e31cb77bfbda on git) is the last working LLVM revision.

[shmerl]

I see. Last time I tried building llvm directly using their instructions, it failed after some very long testing sequence. And Debian llvm snapshot repo has only one :( I submitted a bug report for them to keep at least one before the current one as well.

[doitsujin]

Just don't enable tests then, they are always broken in development snapshots. You may want to look at the Arch PKGBUILD on how to set up a mesa-compatible LLVM build.

[shmerl]

New regression (likely llvm related): https://gitlab.freedesktop.org/mesa/mesa/issues/2147

[doitsujin]

I'm not even surprised at this point.

Anyway, i don't think there's much of a reason to post on this thread anymore as most people won't read it. If there's an issue, just report it on the appropriate bug tracker and be done with it.