RADV vs. AMDVLK vs. Radeon Software Vulkan Driver Performance - October 2018 Linux Gaming

[faldzip]

That is exactly what I thought is going to happen when I saw RADV news for the first time - that we are going to have two not-so-good AMD open source Vulkan drivers instead of one better driver. Having two drivers is not twice better... They shouldn't be competing but rather unite but it seems to be very unlikely...

[GruenSein]

That is exactly what I thought is going to happen when I saw RADV news for the first time - that we are going to have two not-so-good AMD open source Vulkan drivers instead of one better driver. Having two drivers is not twice better... They shouldn't be competing but rather unite but it seems to be very unlikely...

Where are you seeing two "not so good" drivers? I am seeing three good drivers. RADV has it quirks when it comes to Vega and AMDVLK certainly has some headroom to improve to reach the -Pro level of performance due to the different shader compilers but in general, this is an amazing situation for linux GPU drivers. Potentially the best it has ever been. Just think about it: You actually have open source drivers at performance levels comparable to the proprietary one. Five or so years ago, this was hardly imaginable.

[Strunkenbold]

Well AMDVLK made a huge leap in terms of performance and compatibility.
This wasnt the case the last time Michael made a comparison.

So maybe the time has come to speak about suspending Radv and let amdvlk integrate into Mesa. Cause while Radv is marvelous work and has very good reason to exist over the last years simply because AMD was too late and amdvlk wasnt mature enough, it doesnt make sense to offer multiple implementations and develop an extra driver which simply doesnt share the same code base among Windows and Linux. Duplicating efforts is not needed and history shows that every driver needs extensive testing. So every manpower available is badly needed.

[humbug]

Well AMDVLK made a huge leap in terms of performance and compatibility.
This wasnt the case the last time Michael made a comparison.

So maybe the time has come to speak about suspending Radv and let amdvlk integrate into Mesa. Cause while Radv is marvelous work and has very good reason to exist over the last years simply because AMD was too late and amdvlk wasnt mature enough, it doesnt make sense to offer multiple implementations and develop an extra driver which simply doesnt share the same code base among Windows and Linux. Duplicating efforts is not needed and history shows that every driver needs extensive testing. So every manpower available is badly needed.

Agreed. AMDVLK seems to be coming along nicely... Would like to see it in Mesa.

But people are saying it's not so easy for outsiders to contribute. So the guys from Red Hat, Valve, Feral etc prefer working on RadV...

[darkbasic]

Why do we need RADV when we already have the official opensource implementation AMDVLK? RADV devs should just drop the development of it and help AMDVLK instead. There is no reason to have 2 opensource drivers and split efforts.

All of this specially considering how the propietary driver is faster, they are in no position to affort reinventing the hot water with 2 opensource drivers.

That's the exact opposite.

[pal666]

A lot of effort went into the amdvlk release, where AMD had to refactor their driver to allow them to opensource it

but surely less than went into radv to write it from scratch
and lol, you are talking like opensourcing is not useful in itself. all software should be opensourced, it makes it better from technical pov

[duby229]

but surely less than went into radv to write it from scratch
and lol, you are talking like opensourcing is not useful in itself. all software should be opensourced, it makes it better from technical pov

Radv wasn't written from scratch, mesa already existed, radeonsi already existed, all the rendering techniques, all the compilers, everything Radv needed was already done. That's why amdvlk was retarded from the first place, AMD already everything they needed. And it -not- better to throw away money and time and talented peoples resources, that's -exactly- what amdvlk does. It's a waste by every measure. AMD is not going to cripple Mesa or LLVM in order to harm Radv, it is here to stay and I wouldn't be surprised one little bit if it surpasses amdvlk some day. It's already close to equal, it doesn't cheat to do so, and has far more active developers contributing to it.

[smitty3268]

but surely less than went into radv to write it from scratch

Judging by the timeframes that each driver took, I don't think that's true. And as duby mentioned, about 95% of the radv driver was pre-existing code that AMD had already opensourced. The kernel driver, the llvm backend, the allocator, existing radeonsi code, etc.

and lol, you are talking like opensourcing is not useful in itself. all software should be opensourced, it makes it better from technical pov

I never said anything like that. I think it's good amdvlk is open source vs not being open sourced.

[hakzsam]

phoronix Btw, I can't reproduce the vkmark performance issue on Vega 56. Just launched the benchmark at 4k: radv: 5487 | amdvlk: 5415