Open Bug 1099354 Opened 10 years ago Updated 2 years ago

Very slow 3D graphics performance on machines with AMD+Intel GPUs (switchable graphics)

Categories

(Core :: Graphics, defect, P3)

40 Branch
x86_64
Windows 10
defect

Tracking

()

REOPENED
Tracking Status
platform-rel --- -

People

(Reporter: raff, Unassigned)

References

(Blocks 1 open bug)

Details

(Keywords: regression, testcase, Whiteboard: [gfx-noted][platform-rel-Intel][platform-rel-AMD][platform-rel-nVidia])

User Agent: Mozilla/5.0 (Windows NT 6.3; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/38.0.2125.111 Safari/537.36 Steps to reproduce: Start FireFox, visit http://threejs.org/examples/#webgl_animation_cloth Using I5 Laptop with AMD 8800M/Intel 4000 switchable graphics. Even on Intel should run at 60 FPS. Actual results: ThreeJS examples run very slowly,typically 20 FPS compared to 60 FPS on Chrome/IE11 and previous versions of FireFox Expected results: Performance should be around 60FPS as before.
Please provide the graphics section of about:support . Have you tried updating your drivers? It sounds like we used to support HWA on your chipset and now dropped it because of driver/GPU issues. Checking the drivers are up-to-date would be the first step to figuring this out.
Component: Untriaged → Graphics
Flags: needinfo?(raff)
Product: Firefox → Core
Resolving this as INCOMPLETE due to lack of information. @raff, please reopen this bug report and provide the information requested in comment 1 if you're still able to reproduce this in the latest Firefox version.
Status: UNCONFIRMED → RESOLVED
Closed: 9 years ago
Resolution: --- → INCOMPLETE
Flags: needinfo?(raff)
OS: Windows 8.1 → Windows 10
Hardware: x86 → x86_64
Version: 33 Branch → 40 Branch
Tried this page on two different laptops: http://threejs.org/examples/#webgl_performance Chrome gets 30-45 FPS, FireFox 40.0.3 10-15 FPS Seems better than it was on previous versions of FireFox (for some ThreeJS examples), but still poor compared to Chrome. Here's the video info dump of one of the laptops: Adapter Description Intel(R) HD Graphics 4000 Adapter Description (GPU #2) AMD Radeon R9 M200X Series Adapter Drivers igdumdim64 igd10iumd64 igd10iumd64 igdumdim32 igd10iumd32 igd10iumd32 Adapter Drivers (GPU #2) aticfx64 aticfx64 aticfx64 amdxc64 aticfx32 aticfx32 aticfx32 amdxc32 atiumd64 atidxx64 atidxx64 atiumdag atidxx32 atidxx32 atiumdva atiumd6a atitmm64 Adapter RAM Unknown Adapter RAM (GPU #2) 2048 Asynchronous Pan/Zoom none ClearType Parameters Gamma: 3600 Pixel Structure: R ClearType Level: 50 Enhanced Contrast: 300 Device ID 0x0166 Device ID (GPU #2) 0x6821 Direct2D Enabled true DirectWrite Enabled true (10.0.10240.16430) Driver Date 6-16-2015 Driver Date (GPU #2) 8-3-2015 Driver Version 10.18.10.4242 Driver Version (GPU #2) 15.200.1062.1004 GPU #2 Active false GPU Accelerated Windows 1/1 Direct3D 11 (OMTC) Subsys ID 00000000 Subsys ID (GPU #2) 0000000c Supports Hardware H264 Decoding true Vendor ID 0x8086 Vendor ID (GPU #2) 0x1002 WebGL Renderer Google Inc. -- ANGLE (Intel(R) HD Graphics 4000 Direct3D11 vs_5_0 ps_5_0) windowLayerManagerRemote true AzureCanvasBackend direct2d 1.1 AzureContentBackend direct2d 1.1 AzureFallbackCanvasBackend cairo AzureSkiaAccelerated 0
-> milan?
Status: RESOLVED → REOPENED
Ever confirmed: true
Flags: needinfo?(milan)
Resolution: INCOMPLETE → ---
Depends on: webgl-perf-parity
Flags: needinfo?(milan)
depends on bug 975132 ?
There's a couple systems in Toronto with the same Intel GPU: https://wiki.mozilla.org/QA/Platform/Graphics/Inventory#Generation_7 Could we give those a shot at finding the regression window, if this is indeed a regression?
When was it better? The comments indicate that 40 is better than before, and that version from comment 0 is even better.
I get 60 FPS @ http://threejs.org/examples/#webgl_performance at 4K on my MBP. Looking into the windows performance.
Assignee: nobody → bgirard
I get 40 FPS on windows with Chrome & Firefox (release or nightly). ~20 FPS with Edge. Quadra 600, D3D11 + D2D 1.1
Alright, I can reproduce on a somewhat similar machine: Firefox Release or Nightly is getting 8 FPS, Chrome is getting 30 FPS Machine: GCN 1.0 - Oland AMD Radeon HD 8570M 0x0166 Intel HD Graphics 4000 Thinkpad G500
So we're disabling surface sharing on this configuration of 'Unexpected Intel/AMD dual-GPU setup'. This halves our FPS rate. By setting layers.acceleration.force-enabled;true I get 15 FPS. Still half of chrome. The time is spent in uniform updates.
Bug 1207170 will double the performance (8 FPS -> 16 FPS). The rest is CPU Bound because of the draw calls. On chrome it's done in the GPU process so it gets a bit more time to run because it doesn't need to run JS. But I'm not sure why it would get double the frame rate. I'd expect it to be more like 20-50% faster, not 100% faster.
Sorry I meant bug 1097321.
Summary: Very slow 3D graphics performance → Very slow 3D graphics performance on machines with AMD+Intel GPUs (switchable graphics)
Depends on: 1097321
No longer depends on: webgl-perf-parity
Whiteboard: [gfx-noted]
JFYI, the support of AMD+Intel GPUs (switchable graphics) literally doesn't exist on my computer. Setup: AMD Radeon HD7400M + Intel HD Graphics 3000 + Win 10 x64 + Firefox beta (43) x64 I used the modified driver because AMD's official one locks browsers to energy saving GPU. I set both firefox.exe and plugin-container.exe to "high performance" (AMD GPU). After that, I can't even start my main profile, NOR profiler: they both can not even render the user interface (totally black/white). To be honest I consider this is a crucial bug that needs to be fixed ASAP..
What you're describing is a different issue. Can you file another bug for it?
(In reply to Benoit Girard (:BenWa) from comment #15) > What you're describing is a different issue. Can you file another bug for it? I filed that with bug 1225285; However I have another problem that may related to this bug: when I check about:support, I can never see my AMD GPU, only intel's. Is that normal? I filed another bug about that anyway: Bug 1225284
Just my 5 cents.. Firefox 43.0 Gentoo Linux, AMD HD7970M+Intel, fglrx driver 15.12. X server set up to always use ATI card. For http://threejs.org/examples/#webgl_performance Chrome: 33 FPS Firefox: 20 FPS For http://threejs.org/examples/#webgl_animation_cloth Chrome: 60 FPS Firefox: 37 FPS
Whiteboard: [gfx-noted] → [gfx-noted] [platform-rel-Intel]
platform-rel: --- → ?
Assignee: bgirard → nobody
Whiteboard: [gfx-noted] [platform-rel-Intel] → [gfx-noted][platform-rel-Intel][platform-rel-AMD]
Just tested on Intel hd4000 For http://threejs.org/examples/#webgl_performance Chrome: 17 FPS Firefox: 15 FPS For http://threejs.org/examples/#webgl_animation_cloth Chrome: 60 FPS Firefox: 52 FPS
This is NOT limited to AMD CPUs and Intel GPUS. Just FYI. My data on intel Xeon 6-core 4.2GHz w/ nVidia Geforce GTX 1080. Firefox 49.0.2 (release channel): For http://threejs.org/examples/#webgl_performance Chrome: steady 59-61 FPS Firefox: max 21 FPS For http://threejs.org/examples/#webgl_animation_cloth Chrome: steady 60 FPS Firefox: steady 59 FPS Just FYI.
Whiteboard: [gfx-noted][platform-rel-Intel][platform-rel-AMD] → [gfx-noted][platform-rel-Intel][platform-rel-AMD][platform-rel-nVidia]
platform-rel: ? → +
There are a few issues in this bug for sure. Dual systems vs. different performance on different hardware. And probably a few more. We could treat this under the "webgl performance compared to chrome".
A WebGL performance issue that is visible across all vendors. Not sure it's a kind of thing that we want to track as platform-rel.
Severity: normal → S3
You need to log in before you can comment on or make changes to this bug.