Testing a new rendering backend (New tests needed)

[Rachael]

Thank you Nash - that is actually very interesting.

It looks like GZDoom performed only slightly worse on that hardware with the exception of Frozen Time where it was half the FPS of LZDoom. This is the kind of data that is really useful. Still though, seeing some results from some genuine GL2 and GL3 hardware will still tell us a lot, so keep them coming!

Unfortunately the wall render time for GZDoom in Frozen Time on that system was nearly 4 times the amount of time LZDoom took to render them.

[Graf Zahl]

These numbers look like Frozen Time still uses distance clipping - the number of linedefs in LZDoom is only half of what should be seen.

[Rachael]

Hmmm...

Code: Select all

start /b lzdoom.exe +gl_lightmode 2 +win_x 0 +win_y 0 +vid_defwidth 1024 +vid_defheight 600 +fullscreen 0 +vid_renderer 1 +r_sprite_distance_cull 0 +r_line_distance_cull 0 -loadgame %*

Am I missing any CVARs? AFAIK those are the distance clipping CVARs for LZDoom, and those are the ones that were included with the script.

[Graf Zahl]

No idea, I only looked at the line count in the printout. Can't do much more right now.

[drfrag]

No, for the GL renderer they are gl_line_distance_cull and gl_sprite_distance_cull.
But AFAIK this thing still doesn't run on GL2 since Beloko didn't change the GL load code. He said he was going to test it on real hardware but he hasn't done it yet. I think the MESA software driver could be useful too. There LZDoom should be faster since it had no shaders.

[Rachael]

Alright, then, I will have to remake the test kit again then. There's been some changes I wanted to make to it, anyway.

[Rachael]

Okay so new tests are needed for this. I disabled the wrong distance culling CVARs in LZDoom which invalidates the previous test results.

The new archive can be found here:

https://drive.google.com/file/d/1BgpjR4 ... sp=sharing

Instead of a bunch of batch files this one has only one called "Testing - Open Me!.cmd" - this allows you to select the mod and the source port and automatically loads the settings for you. This batch file also auto-selects either the 64-bit version or 32-bit version based on your system, so you don't have to worry about that.

As before, please test one source port first, running through each mod - then test the other.

[vaa44]

CPU: AMD Athlon X2 215 (2.7GHz)
GPU: GTS 450
RAM: 4 GB DDR2 (800MHz)
Win7x64

[Graf Zahl]

Well, even that one hardly makes a difference. BTW, the distance clipping is still active in your tests, Nash, and it is all that makes Frozen Time run better.

[Rachael]

That is a fairly modern card (DirectX 11) so the results are hardly surprising. LZDoom ran way better on that one, like all the other more modern ones.

The only way this makes any difference is like with Nash's test - the graphics chip was modern but the GPU was integrated not dedicated - so the processor had far fewer cores and that made the pixel fill rate much slower, and that gave the CPU more breathing room between frames, but GZDoom was still slightly slower.

We really need a lot more pre-2010 chips shown here. The integrated Intel chips are showing some promise though so further results from those, modern or not, would be good too.

(Obviously it is desirable if Nash you could rerun the test with the new kit, if possible, as well)

[Graf Zahl]

I have a suspicion that we'll have a hard time finding enough users still using such old hardware. Looks to me like the need for this may not even be there anymore.
I actually do not expect this to make any significant difference on a Geforce 8xxx or 9xxx - these cards generally had very good branching performance with shaders.
All that really leaves is Intel HD3000 and older integrated chipsets. Yes, essentially 2010 and older. Anything from 2012 upward is likely not to benefit from this.

[Redneckerz]

That is a fairly modern card (DirectX 11) so the results are hardly surprising. LZDoom ran way better on that one, like all the other more modern ones.

The only way this makes any difference is like with Nash's test - the graphics chip was modern but the GPU was integrated not dedicated - so the processor had far fewer cores and that made the pixel fill rate much slower, and that gave the CPU more breathing room between frames, but GZDoom was still slightly slower.

We really need a lot more pre-2010 chips shown here. The integrated Intel chips are showing some promise though so further results from those, modern or not, would be good too.

(Obviously it is desirable if Nash you could rerun the test with the new kit, if possible, as well)

Ill post results of my junk hardware as soon as possible.

I also have some Thin Clients:
One with a Atom N280, 2 GB DDR3 and Intel GMA4500MHD
The other with a VIA VX900 dual core, also 2 GB DDR3 and Chromotion HD graphics (Think S3 Deltachrome 2004-2005 stuff, supporting Pixel Shader 2.0 at best).

They are completely blank though, so ill have to haul them to work and get them running there. Problem is, ill be off till Wednesday starting from Friday.
Work has some great junk laptops though, including a hella old Pentium M with 512 MB DDR1. I may ask my colleague on that if i can throw your test on it.. (Windows 2000 based, even. Its a storage laptop.)

Work thin clients are AMD GX-415 based, so quad core Jaguar's with Radeon R3E GPU with 128 cores. Basically a Athlon 5350. I'd consider that the upper low end ''limit'' of what you want to know since they have modern API's, but carry low low cores. I might be able to get a test running on one of these when people have breaks. Likely friday though, tomorrow there is an emergency to solve first

I have a suspicion that we'll have a hard time finding enough users still using such old hardware. Looks to me like the need for this may not even be there anymore.
I actually do not expect this to make any significant difference on a Geforce 8xxx or 9xxx - these cards generally had very good branching performance with shaders.
All that really leaves is Intel HD3000 and older integrated chipsets. Yes, essentially 2010 and older. Anything from 2012 upward is likely not to benefit from this.

Trust me this will do wonders for anyone with an older rig that still wants to retain full on GZ in everything but the rendering part. And who knows, if this works well and you market that rightly, could bring in some renewed interest.

[Graf Zahl]

Trust me this will do wonders for anyone with an older rig that still wants to retain full on GZ in everything but the rendering part. And who knows, if this works well and you market that rightly, could bring in some renewed interest.

Depends on how many "anyone"s there still are. Remember: This requires a 10+ year old computer with low end graphics hardware! When we had the last survey with GL 2 support it was somewhat around 3%. But that was over 2 years ago! How many such systems do you think are there now? We also ran some performance tests back then - Intel's GL 2 chipsets were slow enough that anything but vanilla levels at 640x480 turned out too slow.

Don't forget that such a low end port also needs to be maintained. If there is no or merely marginal interest it may simply turn out uneconomical.

[Redneckerz]

Depends on how many "anyone"s there still are. Remember: This requires a 10+ year old computer with low end graphics hardware! When we had the last survey with GL 2 support it was somewhat around 3%. But that was over 2 years ago! How many such systems do you think are there now? We also ran some performance tests back then - Intel's GL 2 chipsets were slow enough that anything but vanilla levels at 640x480 turned out too slow.

Don't forget that such a low end port also needs to be maintained. If there is no or merely marginal interest it may simply turn out uneconomical.

Throwing a controversial take here that i can't substantiate but: What do you think people associate with the term GZDoom? Moddable, flexible, but also requiring semi-modern hardware. People who aren't in the know (or simply who don't care) will ask why that is.

Ofcourse, that isn't your concern. Your work is in GZDoom and supporting a certain feature set of hardware that by most measurements should be enough to please most people. (An 8800 series is from 2006.) But GLESZDoom targets that group of people who isn't aware of a hardware survey, who just can figure out that GZDoom is the go-to port and want to try it on their hardware. And many of those have better rigs than the bottom barrel. But many will not.

And this may also be a good choice for retrorigs. Vogons in particular.

[Graf Zahl]

But GLESZDoom targets that group of people who isn't aware of a hardware survey,

Can we please leave this nonsense behind? Every time the user share of old hardware gets mentioned some retro-fanatic starts with this vast miraculous group of users living beyond the internet, using ancient toaster systems and not caring for any advancements.

If you ask me, this group does not exist. I think our surveys were representative of the true user share.
This has been shrinking for many years and by now is close to going away - just like XP went away (last survey was 0.1%!) and 32 bit has probably gone away by now as well (last survey was 1.5$%)
At best you'll find a small number of users that are forced to make do with an underpowered machine due to money constraints.

And regarding Vogons, that's really a group of users we do not need to cater for - at all! It's a major difference to give some people who cannot afford to upgrade an option or such tinkerers who normally have better hardware at their disposal.