Benchmarking Unity graphics performance in WebGL

[SchumannFrequency]

You can take the test on this page. https://files.unity3d.com/jonas/benchmark2015/
When you post, please include your specific hardware and software.

Hardware: Intel i3-3240 + 8GB RAM @1600MHz DDR3 dual-channel + NVIDIA GTX 650 1GB + EVO 850 500GB
Software: FreeBSD 13.1 + Chromium version 114.0.5735.198

Result:

 

[Athlonite]

this is what I get

reran it 3 times get the same message everytime at the exact same spot if I omit this test it gets the same message with Particles aswell

 

[SchumannFrequency]

this is what I get

reran it 3 times get the same message everytime at the exact same spot if I omit this test it gets the same message with Particles aswell

I used Chromium, uBlock Origin and Privacy Badger. It always worked for me.
I tested Firefox on FreeBSD as well. It could complete the test without an issue.
I tested on Brave and Firefox on Clear Linux and nouveau. Both tests completed without an issue.
Ofc it had lower performance on nouveau, only 60 percent of the performance of the proprietary driver on FreeBSD.

I can think of four reasons for your issue:
1. The Vivaldi Pict extension is incompatible with the test
2. Vivaldi is incompatible. But I think it is the only browser you use. But can you install Chromium or Brave? They are open source browsers.
3. Very new AMD GPU is less compatible with the test than old Nvidia GPU, it is an older test so it would make sense
4. Windows is less compatible with the test than BSD and Linux

Maybe other persons can help test on their system so we can see what the exact issue is.

 

[Athlonite]

Seems it's only the Animation and Skinning and Particles tests that fail everything else runs perfectly fine

 

[SchumannFrequency]

Hardware: Intel i3-3240 + 8GB DDR3 @1600MHz dual-channel + NVIDIA GTX 650 1GB + EVO 850 500GB
Software: ROSA Fresh 12.4 + LXQt + Chrome version 117.0.5911.2 + proprietary Nvidia driver

It seems that Linux is slower than FreeBSD in this test.

Hardware: Intel i3-3240 + 8GB DDR3 @1600MHz dual-channel + NVIDIA GTX 650 1GB + EVO 850 500GB
Software: ROSA Fresh 12.4 + LXQt + Yandex Browser 23.7.1.1148 beta + proprietary Nvidia driver

Yandex Browser 23.7.1.1148 beta can be a little bit faster than Chrome version 117.0.5911.2 in this benchmark.

Maybe the differences have to do with how hot my CPU is. It is an older CPU so maybe it performs better in benchmarks if it is completely cool before the benchmark starts.

 

[SchumannFrequency]

AMD R5 PRO 3400G + dual-channel DDR4 @2666MHz + integrated graphics
windows11 + Edge version 115.0.1901.188

 

[cvaldes]

this is what I get

View attachment 306636

reran it 3 times get the same message everytime at the exact same spot if I omit this test it gets the same message with Particles aswell

My Mac mini M2 Pro also stopped at an earlier point of this benchmark.

One of my Windows PCs did the same with a browser that I don't remember. I'm not a benchmark score chaser so I didn't spend any time trying to figure out why.

 

[Marcus Vinicus]

I tried this on Google Chrome Canary, Brave and Edge (Stable Channel) and it froze at the same place each time.

 

[SchumannFrequency]

I tried this on Google Chrome Canary, Brave and Edge (Stable Channel) and it froze at the same place each time.

Both you and Athlonite are using an AMD Navi GPU. My guess is that you are not going to see the problem with all GPUs prior to the Navi architecture. This may explain why it works for me on windows11 + Radeon RX Vega 11 integrated graphics.

My Mac mini M2 Pro also stopped at an earlier point of this benchmark.

M2 chips may have the same problem as AMD Navi. Namely that they have poorer compatibility with certain older software and games. Like Navi, it is a very modern tech that still has bugs. My guess here is that you are not going to see the problem on MacBooks prior to the M1 and M2 chips.

In any case, I think you can report these bugs to AMD and Apple. I do think it is their intention that this test works perfectly on new and very expensive hardware.

 

[Athlonite]

Namely that they have poorer compatibility with certain older software and games

Funny I seem to have no problems running Quake III Arena, Diablo II LOD or Unreal 2k2 I think it's just because this benchmark is old and was not very stable to start with and probably hasn't been kept upto date by the developer

 

[SchumannFrequency]

Funny I seem to have no problems running Quake III Arena, Diablo II LOD or Unreal 2k2 I think it's just because this benchmark is old and was not very stable to start with and probably hasn't been kept upto date by the developer

I quite believe that in most situations you will have no problems. However, it is important that you report such problems to AMD and Apple. Otherwise, these are minor problems that are becoming more and more numerous. The M1 and M2 processors were a very big step backwards in terms of compatibility. Countless basic apps no longer worked on these CPUs. This was necessary but you still notice that they are not equally compatible, even though things have greatly improved.

Navi too was not a step forward in terms of compatibility: https://www.reddit.com/r/Amd/comments/exa7cb
I do find it strange that new/expensive hardware has multiple problems with this test.
I didn't have any issue in the test on FreeBSD/multiple Linux systems/windows11 and AMD/Intel cpu and Nvidia/AMD graphics and three different browser engines (Blink/Gecko/WebKitGTK)

Here's a new test result.

Hardware: AMD R5 PRO 3400G + dual-channel DDR4 @2666MHz + integrated graphics
Software: Clear Linux + i3wm + Chrome version 117.0.5911.2 + Mesa open-source GPU driver

 

[Marcus Vinicus]

Looks like this benchmark is probably old hence it doesn't work on some modern systems. Cannot blame modern GPU's for that.

 

[SchumannFrequency]

Looks like this benchmark is probably old hence it doesn't work on some modern systems. Cannot blame modern GPU's for that.

As far as I know, JavaScript tries to stay 100% compatible with old JS code. At least certainly with perfect working code from 8 years ago.

It seems to me that this bug can also be caused by modern GPU drivers making errors with standard code.

"Out of memory" is a typical error message that you get when the GPU cannot enlarge memory arrays because it assigned too little memory.

Maybe it would help to find whether there is any leaks of the JS objects with the profiler in devtools.

A related topic started a month ago:

Edge default ANGLE graphics backend is glitchy - Microsoft Q&A

I am running the latest version of Windows 11 and using the latest version of Microsoft Edge (although I have confirmed that the bug also happens on both older versions of Windows 11 and Edge as well). Here is my GPU information: GPU 0 AMD Radeon(TM)…

learn.microsoft.com

The fact that it happens on Firefox makes me think that it's not a Chromium specific issue. Interestingly the RX 6600 he was using is an AMD GPU. This leads me to believe that it might be an AMD driver bug rather than a browser problem.

Fix for WebGL Context Lost on Windows Devices with Some Integrated Intel GPUs

Fix for RAM usage & WebGL Context Lost on Windows Devices with Some Integrated Intel GPUs · Issue #3207 · OHIF/Viewers

Seems like there is a bug in Chrome for Windows users with some models of the integrated Intel GPUs. As a result, OHIF and/or Cornerstone3D demos may experience rendering issues, with the GPU proce...

github.com

Seems like there is a bug in Chrome for Windows users with some models of the integrated Intel GPUs. As a result, OHIF and/or Cornerstone3D demos may experience rendering issues, with the GPU process consuming an excessive amount of memory for volume viewports. As a result, the webGL context may be lost and the browser crashes.

Browser Memory Issues

Roll20 - Virtual Tabletop for Playing D&D, Pathfinder, Call of Cthulhu, TTRPGs

Easily create characters, organize games, purchase content, find players, and customize your story for your group - all in your browser.

app.roll20.net

It happens in both Chrome and Edge, so both Chromium.
It seems like this is related to Dynamic Lighting for as I move around I can see, with every update which includes other tokens which emit light, that my memory usage continues to increase and is never freed.
After moving around various numbers of squares:
Pre-module load: 132MB
Module Loaded: 569MB
Moved 10 squares: 2.2GB
Moved 30 squares: 4.8 GB
Moved 60 squares: 8.7 GB

Is threejs / WebGL so heavy or is it just me (and my old MacBook)?
https://www.reddit.com/r/threejs/comments/ptv5fe I'm using an old MacBook Pro 13" 2010 and while I can play (older) OpenGL games in full screen, every time I try to open a threejs website the window freezes and when it resumes the frame rate is like 1 frame per 5 seconds. I know my hardware is old but I find it surprising that a Core 2 Duo and a GeForce 320M cannot even handle 90s-level 3D graphics.

Memory leaks are crippling my M1 MacBook Pro–and I’m not alone

Memory leaks are crippling my M1 MacBook Prou2013and I'm not alone

With the release of Monterey and the new MacBook Pro, reports are swirling about users experiencing issues with memory leaks.

www.macworld.com

In the end, the test runs perfectly on AMD Vega GPUs and it does not work on AMD Navi. Can't we blame modern GPUs if they don't work properly?

Everyone loses if AMD doesn’t fix its driver issues before Big Navi’s release date

Everyone loses if AMD doesn’t fix its driver issues before Big Navi’s release date

It's not only AMD fans who should want Navi's driver bugs squished...

www.pcgamesn.com

But those of you who follow such things already know: AMD has been having some serious issues with its Navi drivers. This isn’t just a few people screeching over on Reddit, either: AMD itself has recognised the issues and released a new driver update that is supposed to fix most of them… but not all of them.

 

[cvaldes]

M2 chips may have the same problem as AMD Navi. Namely that they have poorer compatibility with certain older software and games. Like Navi, it is a very modern tech that still has bugs. My guess here is that you are not going to see the problem on MacBooks prior to the M1 and M2 chips.

In any case, I think you can report these bugs to AMD and Apple. I do think it is their intention that this test works perfectly on new and very expensive hardware.

I won't be filing any bugs with Apple since this benchmark has failed on x64 Windows PCs in my house as well. My view is that this benchmark is a poor tool for today's systems if it won't even complete one iteration. It's not an operating system issue, it's a problem by whoever wrote this benchmark code.

It doesn't appear to be an officially sanctioned Unity tool and there is no support information pro-offered. And people who write operating systems don't need to make sure every single weird little tool works with their OS. It's up to those application developers to take advantage of developer betas and early access to ensure their code runs on the latest and greatest operating systems. This benchmark's author did not do so.

The best place for this benchmark is the circular file.

 

[SchumannFrequency]

I won't be filing any bugs with Apple since this benchmark has failed on x64 Windows PCs in my house as well. My view is that this benchmark is a poor tool for today's systems if it won't even complete one iteration. It's not an operating system issue, it's a problem by whoever wrote this benchmark code.

It doesn't appear to be an officially sanctioned Unity tool and there is no support information pro-offered. And people who write operating systems don't need to make sure every single weird little tool works with their OS. It's up to those application developers to take advantage of developer betas and early access to ensure their code runs on the latest and greatest operating systems. This benchmark's author did not do so.

The best place for this benchmark is the circular file.

AMD 3400G PRO I don't think is that old and on that it works perfectly. I just tested Void Linux and Opera and although the benchmark was not developed for Opera the full test worked fine.
Linux gamers using Navi have the same problem as Athlonite. I asked them to check.. It might be a browser bug or a GPU driver bug.

Speedometer is the most popular browser benchmark and an official benchmark. But according to many specialists, it is one of the most unrealistic benchmarks.

I personally would not be surprised if this Unity WebGL test says more about real-life browser usage than Speedometer. And also remember that several years ago there were many people reporting that the Speedometer test crashed on their MacBook. It is because many persons reported those problems that the test now runs so smoothly on all browsers.

 

[cvaldes]

You keep dodging several facts.

First, this appears to be an old, unsupported benchmark. It's not supported by Unity, it might as well just be some hack by a Unity engineer who may no longer be with the company.

Second, Speedometer is an official benchmark and is actively supported by developers who periodically change the code (hence, Speedometer 3.0).

I agree with the critics that Speedometer is a pretty useless benchmark but I flatly believe that web browser benchmarks are useless in general because no one only uses a web browser for visiting sites that use one and only one specific web technology. And I've been surfing the World Wide Web since late 1993/early 1994 (NCSA Mosaic running on UNIX workstations).

Third, you are repeatedly and blatantly ignoring the fact that this benchmark is crashing on x86-x64 powered Windows PCs (and not just mine).

For what it's worth, this "benchmark" is crashing on a Ryzen 7 3700X CPU, B550 motherboard, RTX 3060 GPU, running Microsoft Edge on vanilla Windows 10 Pro 22H2, with the latest drivers, updated firmware, yadda yadda. Not exactly a bleeding edge system.

There are other Windoze system in the house. I have a Radeon RX 580 graphics card. I'm most certainly not planning on moving hardware components around trying to get this stupid benchmark to run.

As to the issue with M1 Macs crashing earlier with Speedometer I don't remember that because I'm no benchmark score chaser. I don't keep track of that sort of stuff. In fact, I anticipated that there would be software incompatibilities when Apple Silicon was new. I waited 2.5 years for the M2 SoC to upgrade my Intel-powered Mac mini 2018 to Apple Silicon. For sure there would be some software compatibility early on.

Some dude on TPU asked me to run a benchmark. I did and said it crashed. That's the extent of my participation in this type of thing because web browser benchmarks are ultimately a failure at measuring usability.

I just need a web browser that will behave politely on the top 50 or so sites that I use. It's more important that my web browser doesn't crash when accessing my brokerage account's site, Amazon.com or SeriousEats.com, not some random lame benchmark. For someone else, it might be Facebook, TMZ, and Etsy.

Again, the burden of this benchmark's crashy behavior is on the writer of the benchmark code. I have no idea who that person is nor any idea how to get a hold of them. If you find out, point that person to this discussion thread. Then ask them why their benchmark is crashing MS Edge on a vanilla Windoze 10 box. Their response will pretty much tell whether or not they actually know anything about publishing quality code.

Just because some can post something they label a "benchmark" doesn't mean they're actually competent at writing such tools. In the same way, someone could post some shït recipe that produces a wheat flour hockey puck that they label "bread roll."

 

[crlogic]

Crashes on my base M2 Mac Mini as well on the astroid field and iPhone 14 Pro Max on the bears. But finished with a score of 87,473 on an i5-10210U laptop. Also finished on a Ryzen 7 4700U laptop with a score of 96,830

 

[Dahita]

Crashing for the same reason as everyone else (Out of memory. If you are the developer of this content, try allocating more memory to your WebGL build in the WebGL player settings.) during the particle scene.

AMD5950x / 2070SUPER / 16Gb RAM (23% used) / Win 11 pro latest update using Firefox laster

 

[Marcus Vinicus]

I noticed there is a number in the benchmark url at the end which is 2015. Probably when this was developed\released? The graphics in the app look about 8 years old to me.

 

[claes]

I flatly believe that web browser benchmarks are useless in general because no one only uses a web browser for visiting sites that use one and only one specific web technology

This is the real problem with all of these threads Schumann keeps making. No web developer is designing around maximum performance, they’re developing around maximum compatibility/q’s and then only benchmarking/speed testing for asset management. We don’t care how fast it is, we care if it works. Benchmarking makes sense for games, but for the web it’s a total waste of time. A good website should perform well on a device that’s 15 years old. Even mentioning bleeding edge tech doesn’t make sense because, surprise, bleeding edge web technologies don’t work on most web browsers (and fail to capture a market more often than not).

SeriousEats.com

I kind of always thought you might be a fan

 

[QuietBob]

I noticed there is a number in the benchmark url at the end which is 2015. Probably when this was developed\released? The graphics in the app look about 8 years old to me.

The benchmark itself was created with Unity 4.x in October 2014. It seems that the final update (the one in OP) -- using Unity 5.3.0 -- took place in December 2015.

 

[cvaldes]

This is the real problem with all of these threads Schumann keeps making. No web developer is designing around maximum performance, they’re developing around maximum compatibility/q’s and then only benchmarking/speed testing for asset management. We don’t care how fast it is, we care if it works. Benchmarking makes sense for games, but for the web it’s a total waste of time. A good website should perform well on a device that’s 15 years old. Even mentioning bleeding edge tech doesn’t make sense because, surprise, bleeding edge web technologies don’t work on most web browsers (and fail to capture a market more often than not).

At least you get it. Some people in this subforum do not.

Web developers do have to design for maximum compatibility. I haven't seen the "this site runs best on ____ browser" footer graphic for over twenty years.

This is even more critical now that many people are accessing the World Wide Web on mobile devices of varying age and capabilities. 100 fps at 4K on a particular PC game? Yeah, you can say it needs an RTX 4080 or higher.

But you can't say that ___ website needs to viewed on the iPhone 19 Pro or Samsung Galaxy 29 or the Dodgy Bros. Super XXL Gigachad Edition. It just needs to work.

I haven't bothered to try this particular benchmark on my older iPhone XS or my even older iPod touch 6th generation (running iOS 12). Why should I? I know it crashes on two other devices under the same roof. There's a Raspberry Pi 4 in the house. I certainly am not going to waste my time booting Raspian to try that.

I'm sure OP has some lovingly crafted LibreOffice spreadsheet showing each web benchmark and how well it performs on each of his devices under whatever version of whatever OS poison+web browser poison combo he has tried.

Great, what does that prove? NOTHING. Absolutely nothing. Do benchmarks help him decide which web browser to use for renewing his vehicle registration? Buying concert tickets? A bar of soap? What does it matter if Web Store A loads 0.32 seconds faster on Browser Y versus Browser X? The time spent referring to his benchmark spreadsheet would negate any time benefit anyhow. And what happens when Web Store B devs change the site code so Browser X runs 0.19 seconds faster than Web Store A on Browser Y?

This is pure hilarity.

 

[tugrul_SIMD]

Says out of memory error but system has 64GB memory.

 

[Athlonite]

Says out of memory error but system has 64GB memory.

What hardware and what OS/browser are you using

We'd not need to ask this if you had of filled in the system specs part of your profile

 

[tugrul_SIMD]

What hardware and what OS/browser are you using

We'd not need to ask this if you had of filled in the system specs part of your profile

Win 11, ryzen 7900, rtx 4070, gskill 4800mhz 2x32GB, msi mag x670e. Edge browser, chrome browser.