Zink OpenGL-Over-Vulkan Driver - Performance Is Turning Out Better Than Expected

[schmidtbag]

It'd be nice if it could be used to run entire Xorg/Wayland sessions without GL. I tested Xorg and it starts, but shows a lot of corruption.

Isn't that what xrender is for?

[marek]

Note that the article doesn't say anything about performance in benchmarks and games. Only piglit tests, which I don't consider relevant. I'd like to see Zink vs RadeonSI.

[zxy_thf]

I get where you're coming from, but that's not the same because even in 2020 we can still buy non-AVX CPUs brand new from Intel. Until Intel quits making these shitty ass Atom CPUs we'll never get to a point where mainstream distributions can consider depreciating non-AVX CPUs period.

With GPUs and graphics technology, neither AMD nor NVIDIA release new GPUs with old technology outside of various product refreshes between generations like GTX 1600/1650 or RX 480/580 as well as they almost always support the older graphics technologies outside of multimedia CODECs (whistling in Mantle). Just imagine if AMD released the RX 5300 that was announced the other day as nothing more than a PITCAIRN XT refresh instead as a low end Navi. Because that's what an Atom is from a consumer perspective and, IMHO, is the biggest way that Intel is holding the community back.

If AMD and NVIDIA (and Intel for that matter) still released non-Vulkan GPUs it would be a more apples to apples comparison.

I agree with you, and that's exactly why I said "easier".
The challenge is we don't know how many old GPUs are still in service (for example I still have an R250* in use), and how much the impact would be.
(Gnome4 does have a vulkan based Gsk renderer, maybe the foundation can have some brief data through surveys).

* R250 does support vulkan in hardware, but not through current OSS stack without hacking.

[artivision]

Note that the article doesn't say anything about performance in benchmarks and games. Only piglit tests, which I don't consider relevant. I'd like to see Zink vs RadeonSI.

Be sure that if async_present like futures will be used to uplift the future level, your work will seem like decades old.

[smitty3268]

Note that the article doesn't say anything about performance in benchmarks and games. Only piglit tests, which I don't consider relevant. I'd like to see Zink vs RadeonSI.

Yeah, I imagine this is mostly testing the speed of the GLSL compiler in Mesa vs how fast it can translate GLSL to SPIR-V and then compile the resulting SPIR-V, rather than the actual runtime performance that people care about.

[illwieckz]

With GPUs and graphics technology, neither AMD nor NVIDIA release new GPUs with old technology outside of various product refreshes between generations like GTX 1600/1650 or RX 480/580.

The problem is not really that they don't sold old tech, is that old tech may still be very usable today.
The AGP (yes, AGP) ATI Radeon HD 4670 (TeraScale) is an old card from year 2008 but it still relevant today, it has 1GB of VRAM and can run Unvanquished on medium profile on FullHD screen at 60fps:



In fact the Radeon X1950 PRO AGP from 2006 (R500, pre-TeraScale, 512MB of VRAM) is almost as good: it starts experiencing problems with in-game models with many bones but otherwise I was surprised it driven my screen at 4K before I switched to FullHD to make it more stable.

The thing is that we only have one planet. Those two cards both have PCIe counterpart you can still plug on a modern computer, something you can't do with a CPU.

Also, you can still buy brand new GT218-based GPU, like the Asus EN210 which is still on sale? For reference, the Geforce 210M, maybe the first product to ship with the GT218 chip, was launched on June 15, 2009. This means we can still buy brand new Nvidia GPU based on 11 years old tech.

NVIDIA releases new GPUs with old technology outside of various product refreshes, they does it to release revisions of their products. For example the Geforce 8400GS, one of the very first Nvidia GPUs targetting modern OpenGL (3.0 and beyond) in 2007, got three revisions and the the third one, launched on July 12, 2010 was using the same GT218 chip.

Note: this chip is a pure crap the Nvidia official driver is bugged with it and games may have to implement specific workarounds for it… The sad thing is that many people with few money has it: some years later the EN210 was sold around 25$, it has 1GB of VRAM etc. And as we seen, Nvidia sold this GPU hidden under revision of some well-known other cards. So many people own this card. It looks like we will still have to ship quirks for it for the next decade to come.

[arQon]

The problem is not really that they don't sold old tech, is that old tech may still be very usable today.
The AGP (yes, AGP) ATI Radeon HD 4670 (TeraScale) is an old card from year 2008 but it still relevant today, it has 1GB of VRAM and can run Unvanquished on medium profile on FullHD screen at 60fps:

I dug out the PCIE version of one those just a couple of months ago, to run in my old machine (now a server) while I dismantled and re-pasted the 7970 in it before donating that card to a friend. So I ran some benchmarks and even some older games during breaks, and I totally agree: it's a very functional card, even today! (and the power draw at idle is shockingly good). In terms of actually being able to keep up, it ran out of steam around 2011, more as the result of VRAM bandwidth issues once games started running multiple fullscreen post-processing effects per frame than from shader exhaustion, but yeah - it's a great little card, and has the added bonus of not needing external power.

While you can (and should) generally run a server headless and just ssh or X into it, if it's having a hardware problem or systemd has effed up the fstab, you NEED some kind of onboard video out, and cards like that are ideal for the job. A GTX720 or whatever may be passively cooled and use fractionally less power, but if you don't already have one it's a pretty big increase in e-waste to get one just for that rather than repurpose old gear for the task, and downright stupid to do so when that old gear is already extremely well suited to it.

[onlyLinuxLuvUBack]

I dug out the PCIE version of one those just a couple of months ago, to run in my old machine (now a server) while I dismantled and re-pasted the 7970 in it before donating that card to a friend. So I ran some benchmarks and even some older games during breaks, and I totally agree: it's a very functional card, even today! (and the power draw at idle is shockingly good). In terms of actually being able to keep up, it ran out of steam around 2011, more as the result of VRAM bandwidth issues once games started running multiple fullscreen post-processing effects per frame than from shader exhaustion, but yeah - it's a great little card, and has the added bonus of not needing external power.

While you can (and should) generally run a server headless and just ssh or X into it, if it's having a hardware problem or systemd has effed up the fstab, you NEED some kind of onboard video out, and cards like that are ideal for the job. A GTX720 or whatever may be passively cooled and use fractionally less power, but if you don't already have one it's a pretty big increase in e-waste to get one just for that rather than repurpose old gear for the task, and downright stupid to do so when that old gear is already extremely well suited to it.

Do you know that you can get a pcie serial card on ebay ?