AMDVLK vs. Mesa RADV Radeon Vulkan Driver Performance On Linux

[phoronix]

Phoronix: AMDVLK vs. Mesa RADV Radeon Vulkan Driver Performance On Linux

Kicking off a number of interesting articles over the week ahead for the Phoronix 19th birthday week is a fresh look at how AMD's official open-source Linux Vulkan driver "AMDVLK" compares to Mesa's RADV Vulkan driver that tends to be more popular with Linux gamers and is the driver backed by Valve, Red Hat, and other stakeholders.

https://www.phoronix.com/review/amdvlk-radv-rx7900

[rhysperry111]

Can't wait for Mr Blumenkrantz to see this, and seethe over all the AMDVLK wins.

Unless of course it's to do with the clocking side of drivers.... I don't think he knows how that works.... or clocks in general.... or the concept of time as a whole

[Quackdoc]

I personally use vk_icd overrides for lots of apps, I use radv by default and opt into amdvlk depending on the app dont see this changing anytime soon

[shmerl]

I wonder why there are such wild discrepancies between 7900 XT and 7900 XTX like in case of Hitman 3. Does amount of VRAM make such a huge difference for radv or it's something else?

[loganj]

i didn't know that arch install both amdvlk and radv with that archinstall tool. so i had so many stutters in games and steam kept compiling shader for the same game even i relaunch the game in the same session of steam. after removing amdvlk everything is so smooth and steam stop compiling shaders with almost every launch of the game.and when it still does is so much faster. like 2-5 min compared to 20-30 minutes before

[mbriar]

I wonder why there are such wild discrepancies between 7900 XT and 7900 XTX like in case of Hitman 3. Does amount of VRAM make such a huge difference for radv or it's something else?

I looks weird and I don't really know why it would behave that way, but i'm pretty sure it has nothing to do with VRAM. There is no way Hitman 3 exhausts 20GB of VRAM, and if it did, then you'd see it perform in the 20-30 fps range.

[ET3D]

I wonder why there are such wild discrepancies between 7900 XT and 7900 XTX like in case of Hitman 3. Does amount of VRAM make such a huge difference for radv or it's something else?

Obviously there's no real reason for the performance difference, because the better performing drivers, RADV in the case of F1 22 and AMDVLK in the case of Hitman 3, show a much smaller difference. It looks like it's clearly a bug. However, why both drivers will have such a bug in different games is certainly perplexing.

[pieman]

7900xtx:
radv wins by 1%
7900xt:
radv wins by 6%
7600xt:
radv wins because amdvlk doesn't even support it (really embarrassing for amd's official vulkan driver)

amdvlk is obviously not bad, but the open source driven radv is doing very well considering its not a strictly amd in house vulkan driver for amd's own products. it is nice that we have alternatives even though personally i've only ever used radv for my 6900xt. a few times i tried amdvlk i had instability problems...

[brucethemoose]

Yeah, I think we can't take much from these averages because certain games are bugged... not just in general, but on one card and not the other even though both are extremely similar.


But what? Phoronix Test Suite should catch weird launch anomolies... So it must be a consistent bug?

[cipper]

stupid question, I cannot understand the advantage in investing in two different drivers. Why not merging amdvlk and radv and focusing on a single, well optimized, driver?