First results from the 4.7.0 survey

Here, developers communicate stuff that does not go onto the main News section or the front page of the site.
[Dev Blog] [Development Builds] [Git Change Log] [GZDoom Github Repo]

Moderator: GZDoom Developers

User avatar
drfrag
Vintage GZDoom Developer
Posts: 3141
Joined: Fri Apr 23, 2004 3:51 am
Location: Spain
Contact:

Re: First results from the 4.7.0 survey

Post by drfrag »

What are you talking about? To compile for WinXP you have to use the old 7.1A Windows SDK. Have you ever done it?
Also the toolset was dropped after 2017 anyway like Graf said.
User avatar
Rachael
Posts: 13527
Joined: Tue Jan 13, 2004 1:31 pm
Preferred Pronouns: She/Her
Contact:

Re: First results from the 4.7.0 survey

Post by Rachael »

Apeirogon wrote:To clarify, CRT in this context is a "C RunTime", not "Cathode Ray Tube". Visual Studio still support it, I hope/think.
He knows that...
User avatar
Apeirogon
Posts: 1605
Joined: Mon Jun 12, 2017 12:57 am

Re: First results from the 4.7.0 survey

Post by Apeirogon »

Rachael wrote:He knows that...
I didnt know that.
I spent around a minute wondering "how MVS is detects that computer uses CRT monitor instead of LCD monitor and why is it matter".

Maybe for programmer with big experience its obvious what does it mean here, but for electrician its not.
User avatar
Graf Zahl
Lead GZDoom+Raze Developer
Lead GZDoom+Raze Developer
Posts: 49053
Joined: Sat Jul 19, 2003 10:19 am
Location: Germany

Re: First results from the 4.7.0 survey

Post by Graf Zahl »

Here's some more comprehensive numbers, from roughly 6700 distinct users, in parentheses the numbers from 4.2.0 with roughly 2000 users:

2.5 % of users report the Softpoly renderer, 0.5% on 32 bit, 2% on 64 bit. But I think it is a fair assessment to assume that like with all other such legacy features, the majority has capable modern hardware, but unlike the other categories there is no way to filter it out here due to lack of information. For the remaining numbers these are being excluded.

1.3% used the GLES renderer on a system it was made for. The majority of users of such systems still used the full GL renderer instead - and the majority of GLES users used it on Vulkan capable hardware... :?
85% (79%) use Vulkan compatible hardware.
6.4% (11%) use hardware which can run OpenGL with all features enabled but cannot run Vulkan. Note: The actual number may be up to 1% lower, but many Intel GPUs cannot be conclusively identified here.
7.6% (10.6%) use legacy hardware which requires fallback solutions in the renderer and only has limited support for some features. Note: I had to take out Mac users here because their broken GL support falsifies the information to a significant degree by now.

1% (1.3%) use a real 32 bit system.

76% (69%) use a system with 4 CPU cores and more - among the Vulkan compatible systems this is 82% (no change).

7.9% (6.0%) use Linux.
2.5% (2.5%) use macOS, 2% with Intel CPUs, 0.5 with Apple M1 (yes, that's 1/5th after less than a year of its existence!)

So a few of the weird numbers corrected themselves again, but the anomaly with legacy OpenGL systems was at least partially explained by macOS having deprecated OpenGL at 4.1, this meant that the entirety of non-M1 Mac users was falling into this segment, but with the shrinking number of legacy hardware was becoming far more significant than two years ago. Since I had no desire to sort these manually I calculated the average without Mac numbers, and now it makes a lot more sense than before. I remember having done this manually last time for 4.2.0 but it's clearly not worth the effort.
User avatar
Graf Zahl
Lead GZDoom+Raze Developer
Lead GZDoom+Raze Developer
Posts: 49053
Joined: Sat Jul 19, 2003 10:19 am
Location: Germany

Re: First results from the 4.7.0 survey

Post by Graf Zahl »

With the upcoming 4.7.1 release the survey will close now. We got reports from ~16000 users, so here's the final rundown - for comparison the results for the 4.2 survey with 6500 users is listed:

2% of users report the Softpoly renderer, 0.5% on 32 bit, 1.5% on 64 bit.

1.5% used the GLES renderer on a system it was made for. Strangely enough the number of users of Vulkan capable hardware using the GLES renderer is even higher.
85% (77%) use Vulkan compatible hardware.
6.4% (13.6%) use hardware which can run OpenGL with all features enabled but cannot run Vulkan. Note: The actual number may be up to 1% lower, but many Intel GPUs cannot be conclusively identified here.
8.1% (10%) use legacy hardware which requires fallback solutions in the renderer and only has limited support for some features. Note: I had to take out Mac users here because their broken GL support falsifies the information to a significant degree by now.

1% (1.5%) use a real 32 bit system.

76% (69%) use a system with 4 CPU cores and more - among the Vulkan compatible systems this is 82% (no change).

7.9% (6.0%) use Linux.
2.6% (2.5%) use macOS, 2% with Intel CPUs, 0.6 with Apple M1


So overall not much change since the first result.

So, now to the important question: What do these numbers mean?

First, it should be clear that 32 bit was a one-off for the 4.7 release. As expected, the numbers have declined even further since the last survey two years ago and it takes no rocket science to see that with Windows 11's increased requirements and the inevitable wave of resulting hardware upgrades it will shrink even further. So for 4.8 32 bit will be retired for good.

The second important data point is use of the softpoly renderer. If we factor out the 32 bit users we are left with just 1.5% of our users actually using this backend.
If we extrapolate from other data points one can safely assume that a large portion of those is actually using hardware capable of hardware acceleration and has no need for this backend.
My stance here has always been that we can tag this along as long as it won't become an obstacle. Right now this isn't the case, but future developments are hard to predict.

GLES uptake is surprisingly low, and more weirdly, to a large part on hardware that does not need it and runs worse on it than on full GL or Vulkan! This makes me wonder if we should restrict backend choices in the startup depending on what the hardware is capable of. For example, there's really no point offering GLES and Softpoly to users of Vulkan capable hardware as inexperienced users on such systems may accidentally cripple their experience.

Most importantly, Vulkan capable hardware has reached a level by now where it makes sense to think about optimizing the renderer for this. Currently the entire render engine is geared toward how OpenGL works and cannot really make any use of Vulkan's strengths. Whether that necessitates another engine split remains to be seen, but it is likely because what makes Vulkan run well won't do much good on OpenGL.
User avatar
Apeirogon
Posts: 1605
Joined: Mon Jun 12, 2017 12:57 am

Re: First results from the 4.7.0 survey

Post by Apeirogon »

by Graf Zahl » 20 Oct 2021 09:19
Graf Zahl wrote:We got reports from ~16000 users
GZDoom 4.7.0 released by Graf Zahl » 22 Sep 2021 13:00
16000 / 31 ~ 516
Thats more than a single Boeing of people each single day.
User avatar
mjr4077au
Posts: 829
Joined: Sun Jun 16, 2019 9:17 pm
Graphics Processor: nVidia with Vulkan support
Location: Gosford NSW, Australia
Contact:

Re: First results from the 4.7.0 survey

Post by mjr4077au »

Aside from the various backends used, does 4.7 report on how many people use software vs. hardware rendering?
User avatar
Graf Zahl
Lead GZDoom+Raze Developer
Lead GZDoom+Raze Developer
Posts: 49053
Joined: Sat Jul 19, 2003 10:19 am
Location: Germany

Re: First results from the 4.7.0 survey

Post by Graf Zahl »

No, it doesn't. Last time we had such a metric was in 2018 with the old software rendering backend, back then it was 8%.
Blzut3
 
 
Posts: 3144
Joined: Wed Nov 24, 2004 12:59 pm
Graphics Processor: ATI/AMD with Vulkan/Metal Support
Contact:

Re: First results from the 4.7.0 survey

Post by Blzut3 »

Graf Zahl wrote:GLES uptake is surprisingly low, and more weirdly, to a large part on hardware that does not need it and runs worse on it than on full GL or Vulkan! This makes me wonder if we should restrict backend choices in the startup depending on what the hardware is capable of. For example, there's really no point offering GLES and Softpoly to users of Vulkan capable hardware as inexperienced users on such systems may accidentally cripple their experience.
Given that the survey is taken once at first startup, wouldn't it be possible that the number in question are inflated as people were possibly just curious to check out the new code path? (Not that it would be a large enough number of people to swing the results any particular direction, also worth noting that renderer usage on Linux might be inaccurate since on Linux the backend setting would be changed after survey sample is sent.)
User avatar
Graf Zahl
Lead GZDoom+Raze Developer
Lead GZDoom+Raze Developer
Posts: 49053
Joined: Sat Jul 19, 2003 10:19 am
Location: Germany

Re: First results from the 4.7.0 survey

Post by Graf Zahl »

Yes, that's surely a possibility, but this has been something I have seen with other metrics as well. 32 bit is also used predominantly by owners of a 64 bit OS. And here the first start issue does not apply.
My suspicion is more that there's people out there that think if it is made for old hardware it also must run better on more recent one.
User avatar
Rachael
Posts: 13527
Joined: Tue Jan 13, 2004 1:31 pm
Preferred Pronouns: She/Her
Contact:

Re: First results from the 4.7.0 survey

Post by Rachael »

Graf Zahl wrote:My suspicion is more that there's people out there that think if it is made for old hardware it also must run better on more recent one.
The thing is - on most software that is not a fallacy at all. A lot of software that is written to work properly on a wide range of hardware setups usually aren't completely aware of how the software works in all of the conditions that it will be run in.

One painful truth I've discovered is just because my GPU's can run Vulkan doesn't mean Vulkan software is guaranteed to run faster on them. It's far more complicated than what hardware is capable of running to determine how best to make use of it, tailored obviously to a single individual's own needs.

I know in the long run you want to eliminate all non-Vulkan backends, and that is the reason for a lot of this commentary, here. To be quite honest - don't let that stop you. That's forward-thinking and it's something you'll inevitably have to do, anyway.

But don't go thinking, either, that the lesser backends aren't useful in one form or another on Vulkan-capable hardware. And it's not just for tinkering, either. Everyone has their own reasons for running GZDoom the way they do, and that's something that simply doesn't show in a survey like this. It could have been for testing, for instance - "how does GLES break my mod?" - or - "can I run X mod faster using GLES instead of Vulkan?" - and unfortunately, sometimes the answer is yes.

I'm fairly sure it will eventually be possible to emulate Vulkan on all machines, anyhow. There already exist translation layers for this (MoltenVK is one of them), and while they are not always in a very mature state, the projects are still active and maintained.
User avatar
MartinHowe
Posts: 2021
Joined: Mon Aug 11, 2003 1:50 pm
Location: Waveney, United Kingdom
Contact:

Re: First results from the 4.7.0 survey

Post by MartinHowe »

Also, there must surely be people who are putting off upgrading old but perfectly working kit until chip & GPU prices fall back to pre-insane levels - me for one :mrgreen:
Post Reply

Return to “Developer Blog”