Announcement

Collapse
No announcement yet.

Mesa Radeon Vulkan Driver Sees ~30% Performance Boost For APUs

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

  • Mesa Radeon Vulkan Driver Sees ~30% Performance Boost For APUs

    Phoronix: Mesa Radeon Vulkan Driver Sees ~30% Performance Boost For APUs

    Mesa's RADV Radeon Vulkan driver just saw a big performance optimization land to benefit APUs like Raven Ridge and Picasso, simply systems with no dedicated video memory...

    http://www.phoronix.com/scan.php?pag...t-No-vRAM-Type

  • gukin
    replied
    Originally posted by dungeon

    Is that typo, twisted numbers? If not then this does not sound like definite improvment to me

    Let alone that i don't see these 30% anywhere for you, maybe it was at different settings tested by dev... as by these numbers game does not sound like it is playable

    I didn't actually try playing but I've been playing on ROTR with the presets set to HIGH and at 1920x1200 resolution. For my benchmarks I just used it with the settings I'd been playing under with my RX 570.

    Nevertheless, it's still pretty cool to see this sort of improvement with such a small patch.

    As for playable, you young'ins never had to play Unreal (1998) with single digit FPS using the "slomo" cheat.

    Leave a comment:


  • aejsmith
    replied
    Originally posted by andrei_me View Post

    So we should understand the improvements as a "up to 30%" then? But at the least the 2 digits improvements seems consistent
    That was the exact wording in my commit message

    Leave a comment:


  • andrei_me
    replied
    Originally posted by aejsmith View Post

    Yeah it's going to differ by system/settings/game. Depends on how much the GPU is bottlenecked by memory access. IIRC where I got the 30% difference was with minimum settings @ 720p, but that was on a different APU (my 2500U laptop).
    So we should understand the improvements as a "up to 30%" then? But at the least the 2 digits improvements seems consistent

    Leave a comment:


  • aejsmith
    replied
    Originally posted by dungeon

    Is that typo, twisted numbers? If not then this does not sound like definite improvment to me

    Let alone that i don't see these 30% anywhere for you, maybe it was at different settings tested by dev... as by these numbers game does not sound like it is playable
    Yeah it's going to differ by system/settings/game. Depends on how much the GPU is bottlenecked by memory access. IIRC where I got the 30% difference was with minimum settings @ 720p, but that was on a different APU (my 2500U laptop).

    Leave a comment:


  • gukin
    replied
    Running ROTR on Mageia 7 2400g Mesa 19.1.4:

    Before Mountain 22.09, (min 12.85, max 42.50) After 27.46 (min 18.13, max 47.67)
    Before Syria 22.09 (min 12.85, max 42.50) After 19.08, min 6.61, max 43.32
    Before Geothermal Valley 17.59 (min 9.23, max 44.87) After 19.23 (min 15.75, max 24.37)
    Before Overall 19.04, after 22.12.

    It's a definite improvement, mins are usually higher and that is indeed a very good thing.

    Leave a comment:


  • timofonic
    replied
    Originally posted by ryao View Post
    It would be really interesting to see the effect of this on a PS4 running Linux.
    Is this possible in newer PS4 consoles?

    Were the mesa and kernel fail0verflow patches merged upstream? Last time I heard, the console hacking group didn't take the effort and AMD developers weren't happy about it nor took the effort.

    AMD responds to Linux kernel maintainer's rejection of AMDGPU patch

    Originally posted by MastaG View Post
    Yeah about the ps4.. are there any public patch sets available (mesa and the kernel) for running linux?
    Yes, but not merged upstream :'(

    https://github.com/fail0verflow/ps4-radeon-patches
    https://github.com/fail0verflow/ps4-linux
    https://github.com/Ps3itaTeam/ps4linux-video-drivers
    https://github.com/Ps3itaTeam/ps4-linux​​​​​​​
    https://github.com/ps4gentoo/ps4-overlay https://github.com/ps4gentoo/ps4-kernel​​​​​​​
    Last edited by timofonic; 08-25-2019, 11:35 AM.

    Leave a comment:


  • Marc.2377
    replied
    Great news, good work.

    (If anyone is curious, GTT in this context stands for Graphics Translation Table. https://en.wikipedia.org/wiki/Graphi...emapping_table)

    Leave a comment:


  • uc_sam
    replied
    Originally posted by FireBurn View Post

    Yes, there are
    Where can one find them?

    Leave a comment:


  • cb88
    replied
    Originally posted by dungeon
    Of course APU memory managment is not ideal, and only question is does this breaks perf of anything else? Does it introduce more stutter, some hangs and so on and so forth... OK, was not the only question

    Optimal is the best anyway, results comes after a lot of testing of as much as possible different apps, otherwise profile



    Nah, that is tested just on laptop. These are weird as that could be an issue with low carve, as laptops AFAIK have UMA Memory set on lower like even 256 MB (not sure if that is FP5 APU default or some laptop vendors decide to set it this low) so suffer here and there as default and some says it is not changable like on Desktop APUs...

    It is questionable, if this improve anything for Desktops... likely testing is needed
    Probably not as this is unbreaking a perf problem... since the entries were in a cacheable area, they were probably using up cache entries... .even though they were uncachable. So what you do by putting it outside the memory area that is cachable (since it doesn't benefit from cache anyway).... is more cachable things can stay in cache... since they aren't getting pushed out by this.

    Leave a comment:

Working...
X