Benchmark Tests - August 2021

[Rachael]

Hello everyone! We are asking for a new round of benchmarks.

The maps that need tested are the same as last time, however, the things we are testing are not.

We're mostly interested in people who have early OpenGL 3/4 hardware, but anyone is welcome to participate, including those who have modern hardware.

Right now this test is Windows only.

Here is a link to the current version of the benchmark suite:

https://drive.google.com/file/d/1CMiuLX ... sp=sharing

When you run the program you will be presented with several options:

Code: Select all

[ ] 1) Test with LZDoom (GL Version 3.3)
[ ] 2) Test with LZDoom (GL Version 2.0)
[ ] 3) Test with GZDoom (Full OpenGL 3.3+ w/out Pipelining)
[ ] 4) Test with GZDoom (Full OpenGL 3.3+ w/ Pipelining)
[ ] 5) Test with GZDoom (OpenGLES 2.0 w/ Pipelining)
[ ] 6) Test with GZDoom (Vulkan)

[ ] 7) Get current hardware profile

[ ] 8) Zip up all your current info for forum submission
    9) Exit

Those of you who cannot run OpenGL 3.3+ apps, you will only need to test options 2 and 5. (Please use 7 too to get a hardware profile). Then use 8 to create an archive with your results that can be posted directly to the forum.

Those of you with early OpenGL 3.3 or OpenGL 4.0 cards - All options except 6 will be relevant to you. We're interested in all 5 of the other tests.

If your card is Vulkan capable, then we'd like you to test for that, as well.

Please, if possible, get your hardware profile included in the results, as well.

The batch file is now designed that the tests occur automatically for each port you pick. So if you select test "1" for example - it will load each mod for LZDoom and run benchmarks and save them.

We would be grateful for any assistance we can get for this! Thank you.

[Redneckerz]

Will try as soon as i got all my Doom stuff on the new rig.

One thing is for certain, i can now run them all (Well, except Vulkan).

[Rachael]

Here are my tests. I tested 6 GPU's.

Hurt crashed on my AMD with the pipelining enabled, but all other tests successfully ran. So that particular result is missing.

[PlayerLin]

My GTX 1060 is MSI GTX 1060 6GB version, with 452.06 driver.

All test running fine, but some of tests seems the LZDooM/GZDooM startup will take a little while and Windows complains it "stops responses"...still then loads and runs fine, maybe not big deal I guess.



I do have a GTS450 but it installed on a old low end AMD CPU(Athlon II X2 250) with WinXP 32bits machine and the beachmark setup could not working on XP system so I can't test on it, and I just too lazy to setup a system that would works, sadly.

[Graf Zahl]

Looking at Rachael's numbers was quite interesting. Even on the slower cards LZDoom came in last - the pipelining is clearly the biggest boost on problem hardware as it allows better parallelization of work when the engine is GPU bound.

Also, OpenGL on AMD is just pathetic. Some of the numbers on the 550 even suggest that the driver is occasionally recompiling the shaders on the fly.

@Playerlin: No, we no longer support XP because Microsoft has dropped compiler support for it. This means that we'd have to compromise our entire setup for a vanishingly small group of potential users.

[PlayerLin]

@Playerlin: No, we no longer support XP because Microsoft has dropped compiler support for it. This means that we'd have to compromise our entire setup for a vanishingly small group of potential users.

I think that's fine, it's just too much problem to run newer shit on XP machine these days. I can't use some newer programs on that XP machine because they didn't worked at all! So yeah, I'm perfectly fine with that.

It just too bad I can't run test on that GTS450 card, as I don't have valid extra OS key for install newer system(like Win7) on it and my old GTX660Ti machine is not mine and not on my house anymore so I could not test on that one too.

[Gez]

Here's mine. I have a GeForce 610M, it's not Vulkan-capable.

[Rachael]

Windows 7 can be installed without a key. You will want to put it in its own partition, however, not install it permanently, because eventually it will lock you out after the trial period ends. Windows 8 and above can be installed to and booted directly off of a USB drive without too much trouble, using a utility like Rufus. Not sure if Windows 7 can be booted that way but I think it can.

[phantombeta]

Here's mine. GTX 1060 on latest drivers with an i7 7700K, on latest Win 10. I'd try on my integrated GPU too, but that'd be a pain in the ass.

[Graf Zahl]

We don't really need any more numbers of Geforce 1060 or better. The results for these modern NVidia cards are very clear already. What you got here looks very close to my own, just with a CPU that's a little faster.

[PlayerLin]

Thanks Rachael's suggestion, I found my USB bootable drive with Win7 installations that I did for rescue(did used that drive to fix a system dll missing error months ago, stupid Windows deleted a system dll after shutdown normally...still no idea why that happened). I decided install Win7 64bits HP on that XP machine(install on other one partition that is bigger, since it just a temp. system for doing the benchmark test, after installed, use EasyBCD for reconfiguration to let me boot on both XP and Win7), I only installed nVidia Geforce Gameready driver 381.89 and then running the benchmark after reboot.

Yeah, no Valkan support for that ASUS GTS450. (After I did the Valkan test and check the result file about system spec. I manually removed those Valkan results because they're not valid... lol)

(The GPU-Z shot was taken on XP system if you wonder why the Driver Version said different version)

[Rachael]

Those results seem to show that performance tends to be way better with OpenGL pipelined on this card. Surprising though - I would have thought OpenGL ES would've done a little bit better. And to some extent it does do better than the full-featured OpenGL without pipelining.

[Graf Zahl]

That card is heavily GPU bound, on those switching pipelining on will definitely help a lot because it won't let the GPU sit idle while the engine does pure CPU work. I am not really surprised about GLES performing worse. The shader units are well capable of doing branching on it, so the added time from constantly changing shaders will be the dominating factor.

But yet again LZDoom loses. I think for GL 4.x hardware we can already see a pattern. On extremely weak hardware the lighter shaders of the GLES version will win and on anything a little bit more powerful the full shader will win.
About the pipelining, though, it should be said that with values above 150 or so it's mere cosmetics. The numbers look better but I doubt it will make the game feel smoother.

[vaa44]

About the pipelining, though, it should be said that with values above 150 or so it's mere cosmetics. The numbers look better but I doubt it will make the game feel smoother.

I think that on a 144 Hz monitor, the difference will be noticeable. And in heavy mods, an increase of 10-20 fps will not be superfluous.

[drfrag]

Out of curiosity i ran the benchmark on my AMD Radeon R2, performance in both GZDoom and LZDoom is pretty much identical here minus Vulkan which doubles it.
Between GLES and GL2 LZDoom GLES is only faster on KDIZD, usually full LZDoom is slower and GL2 LZDoom is faster as expected but there are a few surprises.
I still need to investigate the crash on Intel GL2 but most likely will be slow.
And i even ran an alternative benchmark with lights, enabling autoload for lights and brightmaps and removing the -glversion 3.3 parameter for LZDoom. But again the results are very similar across the board here.