First results from the 4.11.0 survey

[Graf Zahl]

We now have 2000 survey reports so it is time for a first summary.

Here's the most interesting numbers:

25.4% use a Geforce RTX card
20.9% use a Geforce GTX960 or better
5.6% use an AMD Radeon RX high end card

8.1% use an Intel UHD with Vulkan compatibility, a further 6-7% use other low end hardware
7.6% use hardware that requires OpenGL and has no Vulkan support, of this 0.4% are performant enough to run the full GL backend

2.5% use Macs, 2.2% on ARM, 0.3% on x64. (The Mac version was the last one released so this will probably increase.)
12% use Linux
1.5% use the Steam Deck


The trend that has shown in the last two surveys has continued. OpenGL usage in general has shrunk but not much - but what did happen is that the high end of OpenGL has nearly been entirely vacated (the numbers are so small that a single report results in wild fluctuations), what is left here is ultra low performance integrated chipsets. At the same time, uptake for high end graphics hardware has stopped. We can see that within the segment the specs increase (RTX went from 20% to 25%, this was the only segment with actual growth but it all came at the cost of the high end GTX segment) but we see virtually no migration from the mid range to here. On the other hand we see the oldest still supported GPUs slowly decline but they seem to be replaced with equally weak ones only two or three years more recent.

The only conclusion that can be drawn by now is that we got two mutually exclusive groups at both ends - those with good computers will continuously upgrade them and those with low spec systems will tend to ride them out until they break and then get another low end system again, so I do not see the market shift in any meaningful way for the foreseeable future. What Intel puts into their low end CPUs is still pathetic.

The only thing I can say with certainty is that the current market situation makes it impossible to let OpenGL continue to dictate the engine's roadmap. With all the work being done on vkDoom a split is inevitable, so that the modern Vulkan backend can continue unencumbered and the remaining OpenGL hardware can be serviced by a legacy fork only receiving the game-side changes but have the render backend feature frozen, also using GLES as default.

[Professor Hastig]

The only thing I can say with certainty is that the current market situation makes it impossible to let OpenGL continue to dictate the engine's roadmap. With all the work being done on vkDoom a split is inevitable, so that the modern Vulkan backend can continue unencumbered and the remaining OpenGL hardware can be serviced by a legacy fork only receiving the game-side changes but have the render backend feature frozen, also using GLES as default.

To be honest, it is about time.
So what's the future for Vulkan then? Will it just be new features like vkDoom or are there also plans to improve performance? With a port like Helion making strong inroads I think this may become more important as time goes by. Not that i'd need 1000 fps but being able to play some of these maps with reliable 144 fps would be nice, but some still miss that mark.

[Graf Zahl]

The issue here is time. All this stuff takes a lot to find the right solutions and that's the one thing we never have enough of...

[dpJudas]

In VKDoom we have this thing called the level mesh. It is used for the ray tracing acceleration structure and the light mapper. It is basically the Doom level as a mesh. One way of improving performance would be to slowly add more and more to it so that it actually contains all the information required to render the level. It isn't really the focus of the current work in VKDoom though, so don't expect it to happen anytime soon.

[Graf Zahl]

Just a quick note about the OpenGL numbers.
The almost non-existent decline of OpenGL always seemed a bit low to me so I went over the 4.9 numbers again with exactly the same formulas as I did for 4.11 and which is proven to give correct numbers, and it turned out I made a dumb error when compiling the final results. For the most part of the survey GL sat at 10.5% but then at the end I arrived at 8.7%.
Using the exact same formulas, the 4.9 survey ended with a user share of 10.7% of OpenGL users. For 4.9 I did the final counting by hand and not by using SQL queries on the data base and it looks I accidentally deducted the macOS users twice from the result which explains the dropoff of ca. 1.8%. At the time macOS x64 was at 1.6% and the discrepancy between 1.6% and 1.8% comes just from more users reporting in afterward.

With that the current GL numbers make sense again. Right now with 4100 users reporting OpenGL is at ca. 7.9% which means a decline of 2.8 percentage points (equivalent to an overall decline of 25% of its market share) which is well within reasonable expectations for a year.

[Professor Hastig]

That definitely makes a lot more sense now. That sharp dropoff between the running results and the final rundown for last year had always seemed odd, as has been the low dropoff to now. With that being corrected, when looking at past surveys we had:

2018 -> 2019: 29% -> 22% : -25%
2019 -> 2021: 22% -> 14.7% : -33%, split across both years approx. -20% each
2021 -> 2022: 14.7 -> 10.7% : -27%
2022 -> 2023: 10.7 -> 7.9% : -26%

So the numbers are quite consistent and projecting into the future we get:

2023 -> 2024 -25% => 5.9%
2024 -> 2025 -25% => 4.4%
2025 -> 2026 -25% => 3.3%
2026 -> 2027 -25% => 2.5%

This doesn't consider the increasing age of the affected systems and it may well be that they have to be retired for other reasons, like low RAM or insufficient cores that make it ever harder to run modern software on them. If I had to take a guess, 4 years of token support is the best that can be gotten out of it from now on.

[BFeely]

What's the advantage by the way of GLES over desktop GL? Requires a trampoline initialization method either way?

[Graf Zahl]

Its primary advantage is that it runs faster on low end hardware at the cost of some secondary features.

[Graf Zahl]

Here's the final numbers after crunching more than 9000 reports

Here's the most interesting numbers:

26% use a Geforce RTX card
21% use a Geforce GTX980 or better
10.4% use other modern hardware with good performance.
---------------
57.4% are on the very good to excellent side of performance.

17.2% use mid range hardware, this very evenly distributes between lower and upper mid range
8.3% use low end Vulkan hardware
2.7% use ultra low end Vulkan hardware (i.e. these are so slow that running at full HD resolution with 60 fps should be next to impossible)
7.3% used Vulkan hardware that could not be classified because the model string was too generic.

8.5% used OpenGL hardware (see my note below)


2.4% use Macs, 1.76% on ARM, 0.64% on x64.
8.3% use Linux, among these 1.3% use the Steam Deck

Regarding the OpenGL numbers I have a strong suspicion that some manipulation had been going on here.
The distribution of reported systems changed quite dramatically in the second half of the survey, resulting in Linux and Mac numbers going down and OpenGL numbers disproportionately going up - it had been holding steady at 7.9% until this started. This is not how statistics normally work so I have reason to believe that some 'interested people' were trying to increase these numbers.
All I say is 'Sorry, did not work' If one particular market segment shows such behavior while all others continue like before it immediately raises some suspicions.

Nevertheless, the numbers speak for themselves. OpenGL market share is definitely shrinking, but like I mentioned before, the discrete cards of this era have almost vanished, it gravitates even stronger to low end laptops than before.
Other than that, no surprises. RTX has gone up, but mainly at the cost of older high end cards, the rest of the low end sees a similar slow erosion, being replaced with modern laptops that are a bit better. And there is virtually no crossover between low end and high end.
With the upcoming split between Vulkan and OpenGL I think the survey has outlived its usefulness and I do not expect another to be run until some dramatic change in the computing hardware occurs.

[yum13241]

How long does hardware like https://www.intel.com/content/www/us/en ... tions.html have?

I'd guess 2 years. (Under Linux, this shitty iGPU does support Vulkan, so until New-Graphics-API-Ultra is 10 years old)

[Rachael]

As long as it can do Vulkan it will still go for a while yet. But having Vulkan and running it well are two different things - old iGPU's might be problematic with the transition.

With a big however - The Vulkan base that GZDoom is set to transition to currently has a shader split similar to the OpenGL ES backend. So - being able to run Vulkan might actually be all you need and it may still run acceptably, at least for a while, yet. However, over the long term you will need support for bindless textures, and not all Vulkan implementations can do that.

[Graf Zahl]

The lowest end crap that Intel produced while still supporting Vulkan isn't any better than an ancient HD 3000 which is WAYYYYYY too weak to run GZDoom even at semi-acceptable frame rates. This hardware is so bad that I even think that Helion's 'render the map in one go' approach will actually be slower on it.

I do not know where precisely this one fits in but it cannot be fun playing anything demanding on it.

The real irony with OpenGL is quite similar to the GL 2 to GL 3 transition: If it wasn't for Inte'ls apathy to give older iGPUs proper driver support, we probably could declare OpenGL dead already. Half of the non-Vulkan hardware right now is only there because Intel does not provide a working driver, not because the hardware cannot do it.

[Rachael]

That's been the Intel standard business practice since the day they started producing iGPU's. There's a really good reason why Linux drivers for Intel always have higher OpenGL versions and much more frequent Vulkan support than with Windows. It's the same reason why Intel iGPU's get better support on the Mac OS platform too.

[Blzut3]

How long does hardware like https://www.intel.com/content/www/us/en ... tions.html have?

I'd guess 2 years. (Under Linux, this shitty iGPU does support Vulkan, so until New-Graphics-API-Ultra is 10 years old)

I can't easily find a strong confirmation, but I believe that uses the same GPU architecture as Ivy Bridge (Gen 7). While that does nominally have a Vulkan driver on Linux, my understanding is it's at best a "best effort" and it lacks some features needed for a full Vulkan implementation. The Gen 7.5 graphics in Haswell is from what I hear the first generation to be Vulkan capable, but I believe I've heard Broadwell (Gen 8) is the first to have a reasonable quality implementation.

Ivy Bridge, Haswell, and Broadwell recently had their Vulkan driver split off from ANV, and the reason for that split doesn't give a whole lot of hope that the Vulkan implementation will be high quality (effectively hobbyist support now): https://www.phoronix.com/news/Intel-HAS ... kan-Gen7-8

So while it's true that Mesa has a Vulkan driver for these GPUs, it sounds like it's a case of "something is better than nothing" and you should probably be looking to upgrade to Gen 9 (Skylake) or newer if you want to seriously run Vulkan software even on Linux. That's even ignoring what Graf said about performance.

[inkoalawetrust]

Regarding the OpenGL numbers I have a strong suspicion that some manipulation had been going on here.
The distribution of reported systems changed quite dramatically in the second half of the survey, resulting in Linux and Mac numbers going down and OpenGL numbers disproportionately going up - it had been holding steady at 7.9% until this started. This is not how statistics normally work so I have reason to believe that some 'interested people' were trying to increase these numbers.
All I say is 'Sorry, did not work' If one particular market segment shows such behavior while all others continue like before it immediately raises some suspicions.

Why is it so hard to believe that someone doesn't have a 500+ euro computer ? Not everyone has a gaming setup least of all with a game as old as Doom, regardless of how different GZDoom is from stock id Tech 1, most people that aren't within this inner circle of the community are not even actually aware of the differences. Besides, like you said, this is probably the last survey, so is it surprising that a few more people than normal chimed in because they don't want support dropped even more than before ?