Announcement

Collapse
No announcement yet.

AMDVLK Still Has Some Performance Advantages Over Mesa's Radeon RADV Vulkan Driver, But It's A Good Fight

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • AMDVLK Still Has Some Performance Advantages Over Mesa's Radeon RADV Vulkan Driver, But It's A Good Fight

    Phoronix: AMDVLK Still Has Some Performance Advantages Over Mesa's Radeon RADV Vulkan Driver, But It's A Good Fight

    As it's been a while since our last comparison of the two AMD Vulkan drivers for Linux gaming and with getting the Radeon VII situation straightened out here are some fresh benchmarks of the latest AMDVLK and RADV Vulkan drivers when running various Ubuntu gaming benchmarks with Radeon RX Vega 64 and Radeon VII graphics cards.

    http://www.phoronix.com/vr.php?view=27926

  • ntropy
    replied
    Originally posted by airlied View Post

    Sounds like you are using pro not vlk
    I extracted the json and icd files from the amdvlk github repo precompiled Deb for Ubuntu.
    As far as I know tho, the amdvlk json and icd files are the same used in the pro driver, only older versions and with a longer update interval between versions.

    Leave a comment:


  • L_A_G
    replied
    Originally posted by airlied View Post
    But some people and companies want improvements in the amd vulkan drivers that AMD aren't prioritising, so AMD hasn't covered what they want, so when they put development effort in they want it to count for something and get a useful return on investment. So working on nouveau would be pointless for those people since that isn't going to get the result of a first class Linux vulkan driver as opposed to a Windows driver dumped over the wall every few weeks to be second class Linux driver.
    Well considering RADV is a community effort and thus not privy to all the important information that an in-house development team would be privy to I'd argue that it's even more of a second class driver than AMDVLK. Obviously AMD is way better at providing this information than Nvidia, who is still kind of deserving of the rather harsh words Linus had for them a few years ago. However it's not like the RADV developers have the same kind of access to the hardware documentation and specially the hardware specialists that Bridgman & Co have access to.

    But if you're actually getting paid by customer companies to work on a dead end NIH syndrome project like RADV then I guess the waste of engineering resources does have a point as you're obviously going to have a profit margin in that. Red Hat, now a division of IBM, is a private company whose first and foremost goal is to generate a profit, not a non-profit like Stallman's Free Software Foundation or Linus' Linux Foundation.

    Originally posted by smitty3268 View Post
    You're sure posting a lot about something you supposedly don't care about.
    Well I guess that depends on your standards. Before this thread I don't think I've posted about RADV in quite a few months and every post in this thread after the initial one has been replying to people responding to that comment.

    Originally posted by smitty3268 View Post
    You're sure posting a lot about something you supposedly don't care about.
    It my not have come off like it was supposed to, but that comment was a tongue-in-cheek response to a similarly tongue-in-cheek comment from Dave.
    Last edited by L_A_G; 06-04-2019, 07:38 AM.

    Leave a comment:


  • bridgman
    replied
    Originally posted by L_A_G View Post
    If you're free to work on something even more pointless than something already completely pointless it sounds like you've got the kind of job security where you can do pretty much anything, maybe even groping co-workers just for fun, and not get fired.
    At the risk of stating the obvious, I don't think airlied was suggesting that either he or his employer felt that his new project was pointless.

    Leave a comment:


  • smitty3268
    replied
    Originally posted by L_A_G View Post
    The fact that I didn't even know that you've moved on to something that isn't pointless should say how little I actually care about RADV.
    You're sure posting a lot about something you supposedly don't care about.

    Leave a comment:


  • airlied
    replied
    Originally posted by L_A_G View Post

    If you're free to work on something even more pointless than something already completely pointless it sounds like you've got the kind of job security where you can do pretty much anything, maybe even groping co-workers just for fun, and not get fired. Lucky you, but you probably shouldn't try your luck with the groping.



    Being a second class dev may not be that much fun, however I'd argue it's still more rewarding than working on a completely pointless NIH syndrome project. That's probably what the people who worked on Mir felt towards the end.

    However what I actually meant was that the developer effort is better spent on something other than open source Vulkan drivers for AMD graphics when AMD has that covered already. If you're asking about specific projects, then the first thing that springs to mind is Noveau. It's unfortunately been tragically under-staffed for way too long and could definitely use some additional helping hands.
    But some people and companies want improvements in the amd vulkan drivers that AMD aren't prioritising, so AMD hasn't covered what they want, so when they put development effort in they want it to count for something and get a useful return on investment. So working on nouveau would be pointless for those people since that isn't going to get the result of a first class Linux vulkan driver as opposed to a Windows driver dumped over the wall every few weeks to be second class Linux driver.

    Dave.

    Leave a comment:


  • airlied
    replied
    Originally posted by humbug View Post
    What's wrong with the Radeon 7 in Dota 2 at 1440p? It's performing slower than the Vega 64. How can both RadV and AMDVLK drivers have the same regression?

    When Navi comes out in July AMDVLK is going to be at a big advantage because internal AMD devs would have had access to hardware for testing for months. Will studying the AMDVLK code help RadV developers to speed up their Navi work?
    More likely we'd just be copying from radeonsi as usual, but yes having vlk code out there would definitely make finding the hw bugs easier :-P

    Leave a comment:


  • L_A_G
    replied
    Originally posted by airlied View Post
    ​​​​​​I've moved on to something you would consider even more wasteful than radv.
    If you're free to work on something even more pointless than something already completely pointless it sounds like you've got the kind of job security where you can do pretty much anything, maybe even groping co-workers just for fun, and not get fired. Lucky you, but you probably shouldn't try your luck with the groping.

    You can't develop anything more than fixes for vlk, features are really hard since you have no insight to the internal dev, whether they already have something written. Nobody wants to be second class devs.
    Being a second class dev may not be that much fun, however I'd argue it's still more rewarding than working on a completely pointless NIH syndrome project. That's probably what the people who worked on Mir felt towards the end.

    However what I actually meant was that the developer effort is better spent on something other than open source Vulkan drivers for AMD graphics when AMD has that covered already. If you're asking about specific projects, then the first thing that springs to mind is Noveau. It's unfortunately been tragically under-staffed for way too long and could definitely use some additional helping hands.

    Leave a comment:


  • humbug
    replied
    What's wrong with the Radeon 7 in Dota 2 at 1440p? It's performing slower than the Vega 64. How can both RadV and AMDVLK drivers have the same regression?

    When Navi comes out in July AMDVLK is going to be at a big advantage because internal AMD devs would have had access to hardware for testing for months. Will studying the AMDVLK code help RadV developers to speed up their Navi work?

    Leave a comment:


  • airlied
    replied
    Originally posted by L_A_G View Post

    The fact that people don't push patches to them doesn't mean that they're not going to accept them. Most people interested in contributing something to the effect of the two projects will obviously have gotten involved in RADV in the time it took to properly open source AMDVLK. Also considering how open they've been in their other open source efforts I highly doubt that they're not going to accept other peoples' contributions.



    The fact that I didn't even know that you've moved on to something that isn't pointless should say how little I actually care about RADV. Seeing how even you've moved on to something that isn't a wasted effort, isn't that more than enough reason to tell the people still wasting their time to stop?

    Or is the NIH syndrome too strong with you?

    ​​​​​​I've moved on to something you would consider even more wasteful than radv.

    You can't develop anything more than fixes for vlk, features are really hard since you have no insight to the internal dev, whether they already have something written. Nobody wants to be second class devs.

    Dave.

    Leave a comment:

Working...
X