Closed Bug 1210683 Opened 10 years ago Closed 10 years ago

20% performance regression in Unity3D Dead Trigger 2 after recent ANGLE update(?)

Categories

(Core :: Graphics: CanvasWebGL, defect)

Product:
Core
Component:
Graphics: CanvasWebGL
Platform:
x86_64
Windows 8.1
Type:
defect
Priority:
Not set
Severity:
normal

Tracking

()

Status:
RESOLVED WORKSFORME
Tracking Flags:
Tracking Status
firefox44 - ---

People

(Reporter: jujjyl, Unassigned)

Details

(Whiteboard: [gfx-noted])

Running Dead Trigger 2 Unity3D WebGL demo in emunittest 0.4.1 suite, between ANGLE update from bug 1179280 (five runs): User agent: Mozilla/5.0 (Windows NT 6.3; Win64; x64; rv:43.0) Gecko/20100101 Firefox/43.0 buildID: 20150920030217 Test Name | Total time (lower is better) | FPS | CPU Time | CPU Idle | Page load time | # of janked frames --------------------------------------------------------------------------------------------------------------------------------------- Dead Trigger 2-2014-03-13 | 42536ms | 47.02 | 38216ms | 10.16% | 9590.00ms | 41 Dead Trigger 2-2014-03-13 | 42464ms | 47.10 | 38213ms | 10.01% | 2806.84ms | 31 Dead Trigger 2-2014-03-13 | 42203ms | 47.39 | 37780ms | 10.48% | 2806.54ms | 35 Dead Trigger 2-2014-03-13 | 43813ms | 45.65 | 39907ms | 8.91% | 2738.42ms | 47 Dead Trigger 2-2014-03-13 | 43111ms | 46.39 | 39083ms | 9.34% | 3101.40ms | 38 and after ANGLE update: User agent: Mozilla/5.0 (Windows NT 6.3; Win64; x64; rv:44.0) Gecko/20100101 Firefox/44.0 buildID: 20151001030236 Test Name | Total time (lower is better) | FPS | CPU Time | CPU Idle | Page load time | # of janked frames --------------------------------------------------------------------------------------------------------------------------------------- Dead Trigger 2-2014-03-13 | 51524ms | 38.82 | 47993ms | 6.85% | 9357.58ms | 434 Dead Trigger 2-2014-03-13 | 52549ms | 38.06 | 49034ms | 6.69% | 2853.51ms | 427 Dead Trigger 2-2014-03-13 | 50947ms | 39.26 | 47574ms | 6.62% | 3145.94ms | 423 Dead Trigger 2-2014-03-13 | 50167ms | 39.87 | 46637ms | 7.04% | 3188.16ms | 415 Dead Trigger 2-2014-03-13 | 52494ms | 38.10 | 48983ms | 6.69% | 2933.82ms | 446 which amounts to -10fps of performance, or ~10 seconds of extra CPU time spent in the demo (CPU Time column). Test was run on the following system: HASWELL ------- Custom built desktop PC Windows 8.1 64-bit 3.0 GHz Intel 8-Core i7-5960X 16GB of RAM 3840x2160 pixels display System DirectX version: DirectX 11.0 NVIDIA GeForce GTX 980, 12GB of VRAM, driver version 355.82 - nvd3dumx.dll, nvwgf2umx.dll: - version 10.18.13.5582 - date 8/25/2015 - D3D Version 11.1, WDDM 1.3, WHQL approved Firefox Nightly 41.0a1 (2015-06-29) Several other tests in the suite also carry a CPU performance penalty as well between 2-8 seconds of total CPU score, but the impact is most visible in Dead Trigger 2 (at least on the above tested system). After Dead Trigger 2, Voxatron carries the second biggest CPU runtime impact (+ ~8 seconds).
In Voxatron, the relative impact looks actually quite high. Before the timedemo took 16 seconds of CPU time to run, after the update it is on the order of 24 seconds, which amounts to a +50% increase.
Err sorry, in comment 0 at the very end, the line "Firefox Nightly 41.0a1 (2015-06-29)" is a remnant copypasted line from a previous bug report description, please ignore that line.
Hi Jukka, Could you tell me how to dump these profiling data for WebGL games?
[Tracking Requested - why for this release]: We want the ANGLE update (because we need it for WebGL2 among other things), but we need to understand why we have this performance degredation. Benoit, can you look at this with Jeff next week?
tracking-firefox44: --- → ?
Flags: needinfo?(bgirard)
Whiteboard: [gfx-noted]
(In reply to Daosheng Mu[:daoshengmu] from comment #3) > Hi Jukka, > > Could you tell me how to dump these profiling data for WebGL games? The Dead Trigger 2 demo is publicly available at http://blogs.unity3d.com/2014/04/29/on-the-future-of-web-publishing-in-unity/ . The printed profile sheet in comment 0 is something else from my custom profiling overview tool (geckoprofiler is much better for detailed profiling). I sent you an email about that separately.
Jeff, testing those two builds, I don't see a difference in performance. (Btw those builds are 32bit, I originally tested with 64bit) As a matter of fact, re-testing again with 64-bit builds of current Nightly channel (2015-10-03) against the same Nightly 64bit 2015-09-20, I am no longer seeing the performance difference: emunittest-0.4.1 Run date: 2015-10-04 06:24 User agent: Mozilla/5.0 (Windows NT 6.3; Win64; x64; rv:44.0) Gecko/20100101 Firefox/44.0 buildID: 20151003030225 Test Name | Total time (lower is better) | FPS | CPU Time | CPU Idle | Page load time | # of janked frames --------------------------------------------------------------------------------------------------------------------------------------- Dead Trigger 2-2014-03-13 | 44110ms | 45.34 | 39590ms | 10.25% | 8732.42ms | 64 Dead Trigger 2-2014-03-13 | 43290ms | 46.20 | 38720ms | 10.56% | 2684.94ms | 34 Dead Trigger 2-2014-03-13 | 43692ms | 45.77 | 39656ms | 9.24% | 2759.01ms | 34 Dead Trigger 2-2014-03-13 | 43612ms | 45.86 | 39276ms | 9.94% | 2909.98ms | 34 Dead Trigger 2-2014-03-13 | 43445ms | 46.04 | 39142ms | 9.91% | 2802.38ms | 35 vs emunittest-0.4.1 Run date: 2015-10-04 06:29 User agent: Mozilla/5.0 (Windows NT 6.3; Win64; x64; rv:43.0) Gecko/20100101 Firefox/43.0 buildID: 20150920030217 Test Name | Total time (lower is better) | FPS | CPU Time | CPU Idle | Page load time | # of janked frames --------------------------------------------------------------------------------------------------------------------------------------- Dead Trigger 2-2014-03-13 | 43195ms | 46.30 | 39207ms | 9.23% | 8383.58ms | 58 Dead Trigger 2-2014-03-13 | 42098ms | 47.51 | 37788ms | 10.24% | 2806.20ms | 31 Dead Trigger 2-2014-03-13 | 41903ms | 47.73 | 37521ms | 10.46% | 2883.47ms | 42 Dead Trigger 2-2014-03-13 | 42785ms | 46.75 | 38860ms | 9.17% | 2830.18ms | 51 Dead Trigger 2-2014-03-13 | 42571ms | 46.98 | 38451ms | 9.68% | 2783.05ms | 49 which suggests that the original timings were a fluke. (technically this system is not the same though, as I installed a brand new M.2 SSD drive to this computer and reinstalled Windows from scratch) Jeff/Benoit, can you try running a performance difference on your systems? If there is no reproducible difference, lean on calling this invalid (in which case, very sorry for the noise!)
Flags: needinfo?(jujjyl)
I downloaded the test suite Friday (thanks) and we're going to look into it on Monday. > technically this system is not the same though, as I installed a brand new M.2 SSD drive to this computer and reinstalled Windows from scratch But it's the same CPU/GPU/RAM? Same windows version? I'd hope we don't do IO during the benchmark so the SSD shouldn't be relevant. Perhaps the regression was due to a particular driver version? That's the most likely bad change after a re-install of the same OS version. Looks like you posted the previous driver info, it would be nice to get it again after the re-install. In that case then we might just want to see if we notice a regression on several machines just to be safe.
Flags: needinfo?(bgirard)
Same CPU, GPU, RAM, and Windows version. Checked and GPU driver version is the same as well. If the regression is not reproducible elsewhere, I can only think that I either had some prefs set on one of the runs that I didn't have on the other, or that there was some background process activity going on during the bad run that affected the overall performance.
Given that no interesting variables were changed and particularly the OS and driver version is the same I don't think it's worth testing on several machine. Jukka I'll leave the call up to you if you think it's worth pursing this further. If not please resolve WORKSFORME.
Tried another time today, but could not measure a difference.
Status: NEW → RESOLVED
Closed: 10 years ago
Resolution: --- → WORKSFORME
No need to track this as it probably wasn't a real issue based on comment 9 and 11.
tracking-firefox44: ? → -
You need to log in before you can comment on or make changes to this bug.