[Bug 56167] New: Unigine Heaven 4.0 DX11 benchmark performance loss on Nvidia gpu
https://bugs.winehq.org/show_bug.cgi?id=56167 Bug ID: 56167 Summary: Unigine Heaven 4.0 DX11 benchmark performance loss on Nvidia gpu Product: Wine Version: 7.2 Hardware: x86-64 URL: https://assets.unigine.com/d/Unigine_Heaven-4.0.exe OS: Linux Status: NEW Keywords: download, performance, regression Severity: normal Priority: P2 Component: d3d Assignee: wine-bugs(a)winehq.org Reporter: gyebro69(a)gmail.com CC: hverbeet(a)gmail.com, z.figura12(a)gmail.com Regression SHA1: 7c844cd3c9c465bdeaf3d98d6da5f2f5e460e650 Distribution: ArchLinux I'm filing this bug report because the proposed patch by Zeb Figura from bug #54223 doesn't resolve the performance regression for me in Unigine Heaven DX11 benchmark: https://gitlab.winehq.org/wine/wine/-/merge_requests/4821 The problem: with current Wine (wine-9.0-rc4-25-g44aa651dc54) there is a roughly 30% performance drop in the DX11 benchmark compared to that of Wine-7.1. The DX9 and OpenGL benchmarks are not affected. The first commit where the problem appeared for me: commit 7c844cd3c9c465bdeaf3d98d6da5f2f5e460e650 wined3d: Use the chunk allocator for GL indirect draw buffers. The settings I'm using in Unigine Heaven 4.0: Preset: custom API: DirectX 11 Quality: Low Tesselation: Disabled Stereo 3D: disabled Multi-Monitor: disabled Anti-aliasing: Off Full Screen: unchecked Resolution: 1280X720 This is the result of the DX11 benchmark in wine-9.0-rc4-25-g44aa651dc54: FPS: 126 Score: 3174 Min FPS: 11.9 Max FPS: 362.3 The benchmark with commit 7c844cd3^: FPS: 175 Score: 4390 Min FPS: 10.9 Max FPS: 362.9 Tested and reproduced with the following Nvidia drivers: 470.223.02 (legacy branch) 535.43.22 (Vulkan dev) 545.29.06 Please let me know if I should provide debug logs or further tests. About my current system: Kernel: Linux 6.6.10-zen1-1-zen [x86_64] @ Arch Linux X.Org X Server 1.21.1.10 XFCE 4.18 (compositor disabled) GPU: NVIDIA GeForce GTX 1660 SUPER/PCIe/SSE2 (6 GB) Memory: 32 GB CPU: AMD Ryzen 5 5600X 6-Core Processor Unigine_Heaven-4.0.exe md5: 2477dcb2d7072b96abc6a1ae447a960c -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
https://bugs.winehq.org/show_bug.cgi?id=56167 mirh <mirh(a)protonmail.ch> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |mirh(a)protonmail.ch -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
http://bugs.winehq.org/show_bug.cgi?id=56167 --- Comment #2 from Zeb Figura <z.figura12@gmail.com> --- (In reply to Matteo Bruni from comment #1)
I'm attaching a patch that puts UBOs into their own suballocation pool. With that performance improves 20%+ for me on Nvidia. I haven't tested AMD yet but I'm not sure if this approach would be okay for upstream.
Looks perfectly reasonable to me; please feel free to submit it when you're able. -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
http://bugs.winehq.org/show_bug.cgi?id=56167 --- Comment #3 from Béla Gyebrószki <gyebro69@gmail.com> --- I revisited this bug and ran the benchmarks again in vanilla Wine-11.0-rc3. I also tried the proposed patch on top of 11.0-rc3: https://gitlab.winehq.org/wine/wine/-/merge_requests/9811 The Nvidia drivers I tested this time: 590.48.01 (open), 580.105.08 (proprietary), 570.207 (proprietary). The hardware and the settings used in Unigine Heaven were the same as in my original post. GPU: NVIDIA GeForce GTX 1660 SUPER - 6 GB VRAM (Turing family) CPU: AMD Ryzen 5 5600X (12) @ 3.70 GHz (Cpu boost disabled) Preset: custom API: DirectX 11 Quality: Low Tesselation: Disabled Stereo 3D: disabled Multi-Monitor: disabled Anti-aliasing: Off Full Screen: unchecked Resolution: 1280X720 I'm mainly still using the "old" WOW64 mode and it is shocking that there is a massive 40-45% performance loss with EGL compared to GLX in 11.0-rc3. This affects all 3 benchmarks in Unigine Heaven: DX11, OpenGL and DX9. With the new WOW64 there is only 5-6 % performance loss with EGL in the DX11 benchmark, a mere 2-3 % loss in the OpenGl benchmark and 10-11% loss in DX9 (compared to GLX). About the patch (MR9811): In the old WOW64 mode it barely gains 4-5% with EGL, but gives more than 60% boost in the DX11 benchmark with GLX. In the new WOW64 the patch gives about 50% more fps both with EGL and GLX. -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
http://bugs.winehq.org/show_bug.cgi?id=56167 --- Comment #4 from Zeb Figura <z.figura12@gmail.com> --- (In reply to Béla Gyebrószki from comment #3)
I'm mainly still using the "old" WOW64 mode and it is shocking that there is a massive 40-45% performance loss with EGL compared to GLX in 11.0-rc3. This affects all 3 benchmarks in Unigine Heaven: DX11, OpenGL and DX9. With the new WOW64 there is only 5-6 % performance loss with EGL in the DX11 benchmark, a mere 2-3 % loss in the OpenGl benchmark and 10-11% loss in DX9 (compared to GLX).
About the patch (MR9811): In the old WOW64 mode it barely gains 4-5% with EGL, but gives more than 60% boost in the DX11 benchmark with GLX. In the new WOW64 the patch gives about 50% more fps both with EGL and GLX.
That makes me think that EGL is bottlenecked by something else entirely, enough that fixing d3d performance in this way doesn't help. So that should probably be a new bug, and you could even call it a regression from 7abf4432a7dfdc044f0458f2457829038dc497f0. -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
http://bugs.winehq.org/show_bug.cgi?id=56167 Béla Gyebrószki <gyebro69@gmail.com> changed: What |Removed |Added ---------------------------------------------------------------------------- Resolution|--- |FIXED Status|NEW |RESOLVED Fixed by SHA1| |3d9b48bc5e443f83653faf48653 | |573da880d9008 --- Comment #5 from Béla Gyebrószki <gyebro69@gmail.com> --- With commit 3d9b48bc5e443f83653faf48653573da880d9008 merged-in I see a nice 50% gain in the benchmark result but only with GLX backend. The improvement when using EGL is only 5-6% Still remains the question why EGL performs so badly compared to GLX with Nvidia binary drivers. The final score is 65% less than with GLX. I tried the benchmark with nouveau vulkan driver and Mesa-25.3.2: the difference in the result of the benchmarks between EGL and GLX is almost zero. First I thought this could be due to the adaptive clocking feature, Powermizer. During the benchmark with EGL, GPU clock only momentarily touches 1000 MHz, it fluctuates between 350 and 740 MHz most of the time. With GLX in use, clock frequency is almost constantly near to the of maximum. Even if I set Maximum performance mode and lock the GPU frequency to the specified max. (around 1900 MHz) this doesn't change a dot in the benchmark result with EGL. I think this one can be closed as fixed. The remaining issue with EGL would be the subject of another bug report. -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
http://bugs.winehq.org/show_bug.cgi?id=56167 --- Comment #6 from mirh <mirh@protonmail.ch> --- Sorry for the roundabout question, just for the records what would be the expected performance level for GTX 1660 with the same settings on Windows? -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
http://bugs.winehq.org/show_bug.cgi?id=56167 Alexandre Julliard <julliard@winehq.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|RESOLVED |CLOSED --- Comment #7 from Alexandre Julliard <julliard@winehq.org> --- Closing bugs fixed in 11.0-rc4. -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.
participants (2)
-
WineHQ Bugzilla -
WineHQ Bugzilla