Gzdoom 2.2.0 speed regression

[blueinferno776]

For starters gl_vid_multisample 8 ran perfectly in 2.1.1 on my dgpu with large detailed maps. Now just throwing the multisample to 2x it starts to studder already.

I can run 2.1.1 on my integrated gpu smoothly with no multisample and made it through genesis with no slowdowns anywere. I need my dedicated card to run 2.2.0 reasonably.

The dev builds are unplayable on both my gpus, the engine micro studders like crazy on both x32 x64 in doom2s first map. Am worried this is going to be present with the next release.

Edit. gl_renderbuffers false did next to nothing.

[Rachael]

Have you tried setting gl_renderbuffers to false?

There's really not much that can be done, otherwise.

[blueinferno776]

Holy crap that was quick. Tried that and nothing sadly. Coming from a laptop that can run doom 2016 between 32 to 46 fps on low 1366x768 with 100 percent resolution scaling.

Edit: forgot to mention the vulcan api I used. OpenGL is a slide show. I away have had bad luck with any OpenGL based engines.

[Rachael]

What is your GPU and what is its OpenGL version?

[blueinferno776]

1- Intigrated Radeon R5 graphics built into the Amd a8 7410 apu.

2- Radeon R7 M360

Both use OpenGL 4.5 at startup. My laptop has dynamic switchable graphics and I always look in gzdooms console to see what gpu is used.

[Graf Zahl]

Are you using the latest drivers (and are they official or OEM?) This is a bit odd, I never heard of something like that on modern AMD hardware.

[Graf Zahl]

One thing you should do is:

Start the same scene on a somewhat larger map with a bit of scenery, bind the 'bench' console command to a key and run this on both GZDoom versions. That will give some timing values which may help to narrow down the cause.

Please do this with both vsync on and off.

[blueinferno776]

I stay far away from oem drivers.

My driver is Radeon Software version 16.10.2

I will bench both of my gpus with vsync on and off useing wispers of satan map28. That is my goto to test resource heavy gameplay mods.

[blueinferno776]

Okay. The weirdest thing I ever seen, my integrated graphics is running gzdoom slightly faster than my dedicated card!

R5 = 128 stream possessors across 2 compute units at 850mhz. 1gb ddr3 667mhz memory.

R7 = 384 stream possessors across 6 compute units at 1125mhz. 2gb ddr3 900mhz memory.

This makes no sense what so ever. None of my other games are effected like this.

[blueinferno776]

forgot the benchmarks

[blueinferno776]

Sorry for spamming. For some reason the zip file did no load.

[blueinferno776]

The zip had a screenshot inside which made it too large to load. Did not know about file size restrictions my bad

Edit: I will bench the dev builds later.

[Graf Zahl]

Those benchmarks show nothing extraordinary. Drawcall time is too high, but that has been plaguing AMD since all eternity. Aside from that your CPU is 3x slower than mine. With this setup it's not that surprising that the faster card doesn't make a difference. The combination of relatively slow CPU and AMD drawcall overhead pretty much makes actual hardware performance irrelevant.
Or was this the old version that doesn't have performance issues?

[blueinferno776]

2.1.1 x32 ran much better overall, It never stuttered with the R7 having gl_vid_multisample on 8 with the exception with maps like the one I benched. Now it lags on doom 2 map15 which it never did before and the most recent dev builds lag in the most basic vanilla doom maps. There unusable so far.

I

[Graf Zahl]

Ok. Can you do a bench of MAP15 then with both 2.1.1 and 2.2.0?