Benchmark Tests - August 2021
Moderator: GZDoom Developers
- Graf Zahl
- Lead GZDoom+Raze Developer
- Posts: 49089
- Joined: Sat Jul 19, 2003 10:19 am
- Location: Germany
Re: Benchmark Tests - August 2021
That card shows some very interesting behavior. Draw call throughput with GL2 is a lot better than with modern OpenGL. Aside from that there's no surprises and like all other AMD cards with Vulkan support clearly suggests to use that API.
That said, we really need some numbers of older GL3 cards. But from the looks of it these have become so rare by now that it's going to be difficult.
That said, we really need some numbers of older GL3 cards. But from the looks of it these have become so rare by now that it's going to be difficult.
- ChopBlock223
- Posts: 80
- Joined: Tue Apr 02, 2019 3:19 pm
- Operating System Version (Optional): Windows 10 with extensive trimming and regedits.
- Graphics Processor: nVidia (Modern GZDoom)
Re: Benchmark Tests - August 2021
Would Nvidia GPUs typically give the best performance with Vulkan then? I'm looking at a new computer sometime soon.
- Caligari87
- Admin
- Posts: 6179
- Joined: Thu Feb 26, 2004 3:02 pm
- Preferred Pronouns: He/Him
- Contact:
Re: Benchmark Tests - August 2021
Received test results from a user without a forum account. They did options 1, 2, 5, and 7. Reported a crash on Option 5, test 2 (GLES)
Update: User re-ran option 5 on a ramdisk and did not get a crash. Uploaded secondary results for this.
Update: User re-ran option 5 on a ramdisk and did not get a crash. Uploaded secondary results for this.
- Attachments
-
- Galaeo_benchmarks-2.zip
- option 5 on ramdisk, no crashes
- (3.5 KiB) Downloaded 118 times
-
- Galaeo_CrashReport.zip
- Crash report for Option 5, test 2 with GLES
- (20.41 KiB) Downloaded 121 times
-
- Galaeo_benchmarks.zip
- Crash on Option 5, test 2 with GLES
- (14.45 KiB) Downloaded 131 times
- Graf Zahl
- Lead GZDoom+Raze Developer
- Posts: 49089
- Joined: Sat Jul 19, 2003 10:19 am
- Location: Germany
Re: Benchmark Tests - August 2021
So we can forget about that. Not that it worries me one bit - this was the most recent GPU I already had declared unsupported.
One thing to know here is that in this case LZDoom full and LZDoom GL2 are the same render path - the modern renderer never gets activated on this chipset due to the poor shader performance. So the only way to get a little bit of juice out of it is by running a naked GL2 fixed function renderer on it. Which excludes it from the support roster by default as there is no shader-less render path in GZDoom.
So even with the GLES renderer there's simply no point whatsoever to even think about supprting real GL2 hardware because if the HD3000 can't shoulder it, none of the rest will either.
One thing to know here is that in this case LZDoom full and LZDoom GL2 are the same render path - the modern renderer never gets activated on this chipset due to the poor shader performance. So the only way to get a little bit of juice out of it is by running a naked GL2 fixed function renderer on it. Which excludes it from the support roster by default as there is no shader-less render path in GZDoom.
So even with the GLES renderer there's simply no point whatsoever to even think about supprting real GL2 hardware because if the HD3000 can't shoulder it, none of the rest will either.
- drfrag
- Vintage GZDoom Developer
- Posts: 3141
- Joined: Fri Apr 23, 2004 3:51 am
- Location: Spain
- Contact:
Re: Benchmark Tests - August 2021
I think that card could run the modern path with gl_riskymodernpath.
More benchmarks from another user (Gaia74) with a Geforce GT210.
More benchmarks from another user (Gaia74) with a Geforce GT210.
- Attachments
-
- Gaia_benchmarks.zip
- (22.34 KiB) Downloaded 121 times
- Graf Zahl
- Lead GZDoom+Raze Developer
- Posts: 49089
- Joined: Sat Jul 19, 2003 10:19 am
- Location: Germany
Re: Benchmark Tests - August 2021
That CVAR no longer exists. But it doesn't matter. You can see from the GLES numbers that it is far, far too slow. This mirrors my experience with a HD3000 equipped Mac some 6 years ago. It was slow as shit with the fixed function path and totally unusable with shaders. It's a 10 year old, totally obsolete integrated chipset so there's nothing to expect.
- Graf Zahl
- Lead GZDoom+Raze Developer
- Posts: 49089
- Joined: Sat Jul 19, 2003 10:19 am
- Location: Germany
Re: Benchmark Tests - August 2021
drfrag wrote:I think that card could run the modern path with gl_riskymodernpath.
More benchmarks from another user (Gaia74) with a Geforce GT210.
That's more like it. This hardware generation benefits greatly from the GLES backend.
- kadu522
- Posts: 219
- Joined: Fri Mar 03, 2017 12:32 pm
- Preferred Pronouns: He/Him
- Operating System Version (Optional): Windows 10
- Location: Brasil,Rio de janeiro
- Contact:
Re: Benchmark Tests - August 2021
Is my laptop specs usefull to pass via the benchmark considering i can't even run LZdoom in openGL mode?
Re: Benchmark Tests - August 2021
The benchmarks are not set up to run anything other than OpenGL or Vulkan, so if you can't run OpenGL then no the benchmarks won't be too useful.
- Redead-ITA
- Posts: 439
- Joined: Sun Dec 20, 2015 8:06 am
- Location: At the Pizza Tower
- Contact:
Re: Benchmark Tests - August 2021
Here are my results, i changed my laptop hdd to an ssd but outside of that i have been wondering why the darkest hour just showed the menu and thats it so i would usually start a new game in it, if it is necessary il redo the tests with darkest hour, outside all instances of gzdoom and lzdoom run fine with mostly lag spikes, there was a crash when i tabbed out of the first test in option 5 but it happend only once so i redid the test on that one.
here are the crash reports if you need them
Re: Benchmark Tests - August 2021
Here are my results:
- Attachments
-
- benchmarks.zip
- (27.41 KiB) Downloaded 120 times
- Redneckerz
- Spotlight Team
- Posts: 1061
- Joined: Mon Nov 25, 2019 8:54 am
- Graphics Processor: Intel (Modern GZDoom)
Re: Benchmark Tests - August 2021
Here is mines, on Intel HD4600. Didn't run the Vulkan bench since i don't support that. The pipelining ones did kick in the processor a little, and the GLES version has some rather significant performance boosts (+20 FPS) on a few of the maps. Hope it is of use.
- Attachments
-
- benchmarks.zip
- (22.63 KiB) Downloaded 123 times
Re: Benchmark Tests - August 2021
Modern desktop hardware, discrete GPU.
Amusing how slow the bridge in Frozen Time is compared to...everything.
Amusing how slow the bridge in Frozen Time is compared to...everything.
- Attachments
-
- benchmarks.zip
- (27.26 KiB) Downloaded 112 times
- Graf Zahl
- Lead GZDoom+Raze Developer
- Posts: 49089
- Joined: Sat Jul 19, 2003 10:19 am
- Location: Germany
Re: Benchmark Tests - August 2021
It's one of the scenes in all existing Doom maps with the most rendered number of linedefs around - only recently surpassed by MAP02 of Archi-Tek. Were all those crossbeam bridges made from 3D floors the map would render 1.5x faster. I once made such a map for testing, but can't release it due to missing permissions.
Re: Benchmark Tests - August 2021
The crossbeam bridges aren't even the biggest problem in the scene as I understand it. Just - literally everything in the scene has so many linedefs, especially the cliffs on the sides. It's a prime example of everything to do in how *NOT* to make a map, at least if your intention is to make it playable and have a decent frame rate.
Yes - the result is beautiful (no question about that) - but is it really worth it in the end? It still beats 60 FPS in most renderers on modern systems though (with the monsters frozen in place, though) so I would say that is pretty good, all things considered. So that isn't an issue - at least until the monsters start processing ... then it's more of an issue again.
Yes - the result is beautiful (no question about that) - but is it really worth it in the end? It still beats 60 FPS in most renderers on modern systems though (with the monsters frozen in place, though) so I would say that is pretty good, all things considered. So that isn't an issue - at least until the monsters start processing ... then it's more of an issue again.