[GZDoom 3.7.1] Performance measurements

Post a reply

Smilies
:D :) :( :o :shock: :? 8-) :lol: :x :P :oops: :cry: :evil: :twisted: :roll: :wink: :geek: :ugeek: :!: :?: :idea: :arrow: :| :mrgreen: :3: :wub: >:( :blergh:
View more smilies

BBCode is OFF
Smilies are ON

Topic review
   

Expand view Topic review: [GZDoom 3.7.1] Performance measurements

Re: [GZDoom 3.7.1] Performance measurements

by TDRR » Fri Jan 11, 2019 12:21 pm

Maybe it could be useful for the ocassional person who comes in with low specs and gets a crash or doesn't get OpenGL-accelerated graphics or super-low framerate. It does happen, however the sad thing is that oftentimes those are the ones who often ignore the existence of this forum (like people who just come to play Brutal Doom) so it would be quite useless in that aspect. I'm just going to make the guide and see how it goes.

Re: [GZDoom 3.7.1] Performance measurements

by Graf Zahl » Fri Jan 11, 2019 1:29 am

Rachael wrote: I'm not trying to say they'd be useless - all I'm saying is the greater majority of people in the community will be running systems capable of handling SSAO and shadowmaps smoothly.

With these two features I have my doubts. They really require modern high end hardware which currently makes up roughly 30% of our user share.
On the other hand, once you ignore these two features I'd say that the amount of problematic low end hardware is roughly 15%.

Re: [GZDoom 3.7.1] Performance measurements

by Rachael » Fri Jan 11, 2019 12:11 am

It should be fine if you post the guide.

But I'll be honest - I think the guides will be fairly niche. One statement I hear a lot - it's extremely difficult to escape your own perspective. You might have legacy hardware that would benefit from using such a guide, yourself, but it's easy to inflate your perspective of its popularity and usefulness based on your own personal experiences.

I'm not trying to say they'd be useless - all I'm saying is the greater majority of people in the community will be running systems capable of handling SSAO and shadowmaps smoothly.

Re: [GZDoom 3.7.1] Performance measurements

by TDRR » Thu Jan 10, 2019 11:36 pm

Rachael wrote:Even if there were, they'd run terribly on old cards. There's a reason why GPU vendors stop developing OpenGL features after a certain version - it's the only version that runs well.
That reminds me- I was going to make a sort of guide on running GZDoom on older systems, giving recommendations for performance settings, trying out forks which are meant for older computers, that sort of stuff. Do you think it would do any harm or is it fine if i post it?

EDIT: Also when talking about "legacy drivers" i wasn't meaning stuff that augmented the featureset of old GPUs, that's ridiculous. I meant drivers that had modifications for performance, i could only find one for my Intel G41 but not for this Intel HD 4000. The performance increase in stuff like Doom 3 was very noticeable and actually made the game run at 30fps at worst which is perfectly playable.

Re: [GZDoom 3.7.1] Performance measurements

by Rachael » Thu Jan 10, 2019 8:36 pm

Even if there were, they'd run terribly on old cards. There's a reason why GPU vendors stop developing OpenGL features after a certain version - it's the only version that runs well.

Even GZDoom's basic OpenGL 3.3 implementation (in the modern branch) would run terribly on such systems - on the sheer fact that at that point it's already offloading a lot from the CPU to these old and slow devices.

Re: [GZDoom 3.7.1] Performance measurements

by Graf Zahl » Thu Jan 10, 2019 4:57 pm

I have no idea. It's something I don't need so I never thought about such a thing.

Re: [GZDoom 3.7.1] Performance measurements

by TDRR » Thu Jan 10, 2019 3:54 pm

drfrag wrote:
TDRR wrote:Hey drfrag, do something!
Already did, LZDoom has D3D. Like i said in the other thread ZDoom32 should be fast (VS 2015 build)
I know, hence the (just kidding). Anyways, your efforts to continue support of GZDoom in older machines is absolutely fantastic and LZDoom runs a lot better than GZDoom. I also often use ZDoomLE but i haven't used ZD32 yet. What are the some of the differences between both?

Graf, are there any sort of legacy drivers for these iGPUs? I heard of them once, but i don't know if they are simple rumors.

Re: [GZDoom 3.7.1] Performance measurements

by drfrag » Thu Jan 10, 2019 1:05 pm

Here (GMA 4500M) the vintage build crashes in software, right from the start as i've found recently. On my old ati GL mode crashed instead.
TDRR wrote:Hey drfrag, do something!
Already did, LZDoom has D3D. Like i said in the other thread ZDoom32 should be fast (VS 2015 build), about LE now i'm not sure since it's a MinGW build but can be compiled with VS as well of course if you don't mind about 98 compatibility. I thought MinGW was good (and it's free) but turned out it isn't, worked well for that old code since it wasn't even C++ 11 but more recent versions have serious problems. That doesn't mean it's not useful but not for releases.

Re: [GZDoom 3.7.1] Performance measurements

by Graf Zahl » Thu Jan 10, 2019 12:19 pm

The problem in all those cases is that Intel's GL drivers of that vintage are pure garbage.

Re: [GZDoom 3.7.1] Performance measurements

by TDRR » Thu Jan 10, 2019 11:42 am

Graf Zahl wrote:That said, I have very similar hardware as phantombeta and 3.7.1 runs a lot better than older versions on my system, especially on maps like Frozen Time.
I know the Intels are pure trash, there's no need to tell me :D

But yeah, right now there isn't much point to supporting GL 2.0 because the vast majority of those using GL 2.0 are likely stuck with a Intel iGPU and because it's pretty much unplayable right now in those cards there's not much point in keeping the legacy renderer.

Except because, well, Software renderer performance isn't very good on those same computers. Hey drfrag, do something! :lol:
(just kidding)

Re: [GZDoom 3.7.1] Performance measurements

by Graf Zahl » Thu Jan 10, 2019 3:02 am

The problem at hand is clearly the weak graphics chip. These older Intel GPUs have very different bottlenecks than modern systems and I have a good idea which optimizations for modern discrete GPUs make the frame rate go down: First there is the increased vertex count by rendering sectors as a whole instead of by visible subsectors only. This already showed similar effects on older Intel hardware and led to the vintage build. Second, in 3.7.1 the postprocessing stage is always on, so there's an additional copy of the screen being performed. On modern graphics hardware this normally runs faster than the parallelly executed CPU tasks so it never registers much in terms of performance. However, with such low end hardware this can seriously stall the pioepeline.
That said, I have very similar hardware as phantombeta and 3.7.1 runs a lot better than older versions on my system, especially on maps like Frozen Time.

Re: [GZDoom 3.7.1] Performance measurements

by phantombeta » Wed Jan 09, 2019 12:06 pm

Hardware is an i7 7700K CPU (running at stock clock), an EVGA GTX 1060 SSC and 16GB of Corsair Vengeance LPX DDR4 RAM

1.8.6 Software: 880~894 FPS
1.8.6 OpenGL: ~3160 FPS

3.7.1 Carmack Software: ~1000 FPS
3.7.1 Truecolour Carmack Software: ~726 FPS
3.7.1 Softpoly: ~920 FPS
3.7.1 Truecolour Softpoly: ~880 FPS
3.7.1 OpenGL: ~2220 FPS

All on the start of Doom 2 MAP01, at 640x480

This is what I get on Frozen Time's bridge scene, though: (X: 4684 Y: 6144 Z: 328, specifically, since exact XYZ coords matter)
1.8.6 OpenGL, dynamic lights enabled: ~56 FPS
1.8.6 OpenGL, dynamic lights disabled: ~60 FPS

3.7.1 OpenGL, dynamic lights enabled: ~65 FPS
3.7.1 OpenGL, dynamic lights disabled: ~70 FPS

So, as you can see, on maps made for GZDoom, the modern 3.x OpenGL renderer will most certainly beat the old GL 2.x renderer.

Re: [GZDoom 3.7.1] Performance measurements

by TDRR » Wed Jan 09, 2019 10:57 am

dpJudas wrote: You didn't specify which resolution you ran it at, but I'm assuming given your stats that it is probably 1980x1080 or less. So for comparison on what a new 2018 high end system would give you should probably compare with the 1920x1080 numbers.

The thing I'm the most unhappy about currently is that softpoly is beating the software renderer on my system.
...I ran it at 640x480 :cry:

Also, yeah i don't know why Softpoly is ever so slightly faster than the Software renderer, i noticed this on a 3GHz Dual Core PC and it was quite noticeable, maybe 40fps of difference (for around 160fps or so) but on most other computers i have tried it's barely there, like only 5-10fps

Re: [GZDoom 3.7.1] Performance measurements

by dpJudas » Wed Jan 09, 2019 10:50 am

Here's my numbers for 3.7.1:

default settings, Doom 2 MAP01

vid_vsync 0
vid_maxfps 0

3840x2160 (native monitor 4K resolution)

OpenGL: 958 fps
Software: 152 fps
Software truecolor: 44 fps
Softpoly: 63 fps
Softpoly truecolor: 35 fps

1920x1080 (vid_scalefactor 0.5)

OpenGL: 1165 fps
Software: 305 fps
Software truecolor: 178 fps
Softpoly: 417 fps
Softpoly truecolor: 243 fps

CPU: AMD Threadripper 2990WX
GPU: Nvidia Geforce RTX 2080

You didn't specify which resolution you ran it at, but I'm assuming given your stats that it is probably 1980x1080 or less. So for comparison on what a new 2018 high end system would give you should probably compare with the 1920x1080 numbers.

The thing I'm the most unhappy about currently is that softpoly is beating the software renderer on my system.

[GZDoom 3.7.1] Performance measurements

by TDRR » Wed Jan 09, 2019 10:19 am

I just had the curiosity of seeing exactly how differently did GZDoom perform after some updates and here's my results:

[GZD 3.7.1]
Classic Software: 55fps
Truecolor Software: 35fps
Softpoly: 35fps
Truecolor Softpoly: 34fps
OpenGL: 60fps

[GZD 1.8.6]
OpenGL: 160fps
Software: 154fps

[ZD 2.1.4] ye olde framerate
Software: 185fps

The conditions for these tests are: Default settings, on Doom 2 MAP01's starting position in HMP. Average FPS is written down.
The hardware in which they were tested: Intel Celeron N2805 1.5GHz, Intel HD4000 and 2GB of RAM.

Feel free to post your own.

Top