Announcement

Collapse
No announcement yet.

HasVK Trims Some Fat For This Old Intel Hardware Vulkan Driver

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

  • NM64
    replied
    Originally posted by Dukenukemx View Post
    Core™ i5-5300U
    That's Broadwell, which is considered "gen8 graphics" and therefore uses the current Iris driver (the same one used by even the newest Intel graphics).

    Be glad that Broadwell actually had a substantial release in mobile, otherwise you'd probably be stuck in the same boat as those of us on LGA1150! (as I mentioned on the first page, mainstream desktop Broadwell barely even existed)

    Leave a comment:


  • Dukenukemx
    replied
    Great, now my laptop with it's Core™ i5-5300U is so bad that I shouldn't even use Vulkan on it. I guess I should playing Switch games on it using Vulkan. Why didn't someone tell me this sooner?

    Leave a comment:


  • mangeek
    replied
    Originally posted by Tian View Post

    Can't enter to mesamatrix.org so can't peek. There should be no env variable set to show vulkan 1.3, I have bee researching and it seems like its a labeling error given that the icd file shows vulkan 1.3.230 (mangohud reports 1.2.230), to be sure I wanted to check if the extensions worked (dynamicrendering is showed on hasvk thus the doubt), someone on gitlab told me about vk-gl-cts but its a pain to setup.
    My apologies, I meant https://mesamatrix.net/

    Leave a comment:


  • Tian
    replied
    Originally posted by mangeek View Post

    I'm pretty sure the ANV driver was just reporting a version it was capable of as a driver, but not really when it was actually running on your hardware. Take a peek at mesamatrix.org, bits and pieces of Vulkan 1.2 and 1.1 weren't implemented on older Intel GPUs.

    Check if things actually work the same, and if not, try seeing if there's a MESA env variable that reports Vulkan 1.3 conformance even if you don't actually have it.
    Can't enter to mesamatrix.org so can't peek. There should be no env variable set to show vulkan 1.3, I have bee researching and it seems like its a labeling error given that the icd file shows vulkan 1.3.230 (mangohud reports 1.2.230), to be sure I wanted to check if the extensions worked (dynamicrendering is showed on hasvk thus the doubt), someone on gitlab told me about vk-gl-cts but its a pain to setup.

    Leave a comment:


  • mangeek
    replied
    Originally posted by Tian View Post
    I just installed mesa 22.3 to test the hasvk driver and... WHY THE F@CK IS IT A VULKAN 1.2 DRIVER?! I had vulkan 1.3.224 on mesa 22.2 and it works perfectly, why did they removed crucial features (remember that dxvk requires vulkan 1.3)? Does this means that now I must keep the vulkan driver on 22.2 without updating it? I thought that the whole point of hasvk was to keep the driver "as is" (1.3) not make it worse (1.2).
    I'm pretty sure the ANV driver was just reporting a version it was capable of as a driver, but not really when it was actually running on your hardware. Take a peek at mesamatrix.org, bits and pieces of Vulkan 1.2 and 1.1 weren't implemented on older Intel GPUs.

    Check if things actually work the same, and if not, try seeing if there's a MESA env variable that reports Vulkan 1.3 conformance even if you don't actually have it.

    Leave a comment:


  • Tian
    replied
    I just installed mesa 22.3 to test the hasvk driver and... WHY THE F@CK IS IT A VULKAN 1.2 DRIVER?! I had vulkan 1.3.224 on mesa 22.2 and it works perfectly, why did they removed crucial features (remember that dxvk requires vulkan 1.3)? Does this means that now I must keep the vulkan driver on 22.2 without updating it? I thought that the whole point of hasvk was to keep the driver "as is" (1.3) not make it worse (1.2).

    Leave a comment:


  • Tian
    replied
    Originally posted by NM64 View Post
    That's probably because the e3500 is a wolfdale Core 2 Duo chip that predates having an iGPU on the CPU itself, so the graphics feature-set of the motherboard's integrated graphics will be woeful (for example, all three of my Core 2 Duo laptops use Intel 965GMA which doesn't even support OpenGL 2.0 on Windows, though it can on Linux but that's still not enough for modern versions of Dolphin).

    DeSmuME by comparison has a software rasterizer that is uses by default which tends to be more accurate than its OpenGL renderer so, as long as you have a GPU that can present a desktop image at all, then Desmume should work.

    And to be clear, Core 2 Duo CPUs tended to have better IPC than the Atom-derived CPU cores up until very recently, like Intel 12th gen E-core recently, while even Celeron CPUs derived from Core 2 Duos had IPC comparable to an Athlon 64 which itself is only maybe 12% behind the IPC of a Core 2 Duo (though emulation seems to favor the Core 2 Duo more, making it more than that general 12%).

    BTW protip: last I checked, melonDS ran faster and more accurately than DeSmuME.
    will have to check on melonds!

    Leave a comment:


  • Developer12
    replied
    Originally posted by Leopard View Post

    Having Vulkan support on Windows is not a critical feature at all. Like 99 percent of the users of those chips won't know what Vulkan is or what a graphics api is but it won't be important as basically everything that is essential would be working.
    lots of stuff now runs on top of vilkan, from zink to rusticl to even some dialects of pytorch. having a vulkan driver is becoming a minimum-requirement for even the lowest-performing of smartphone chips.

    Leave a comment:


  • NM64
    replied
    Originally posted by Tian View Post
    about cpu performance... yes its lacking, on my old e3500 desmume runs faster on windows than on my laptop with linux (e3500 has 2.7ghz vs 2.1 on n3050) however dolphin runs fine on the n3050 but not at all on the e3500.
    That's probably because the e3500 is a wolfdale Core 2 Duo chip that predates having an iGPU on the CPU itself, so the graphics feature-set of the motherboard's integrated graphics will be woeful (for example, all three of my Core 2 Duo laptops use Intel 965GMA which doesn't even support OpenGL 2.0 on Windows, though it can on Linux but that's still not enough for modern versions of Dolphin).

    DeSmuME by comparison has a software rasterizer that is uses by default which tends to be more accurate than its OpenGL renderer so, as long as you have a GPU that can present a desktop image at all, then Desmume should work.

    And to be clear, Core 2 Duo CPUs tended to have better IPC than the Atom-derived CPU cores up until very recently, like Intel 12th gen E-core recently, while even Celeron CPUs derived from Core 2 Duos had IPC comparable to an Athlon 64 which itself is only maybe 12% behind the IPC of a Core 2 Duo (though emulation seems to favor the Core 2 Duo more, making it more than that general 12%).

    BTW protip: last I checked, melonDS ran faster and more accurately than DeSmuME.
    Last edited by NM64; 02 December 2022, 08:03 PM.

    Leave a comment:


  • Tian
    replied
    Originally posted by NM64 View Post
    Though, for my emulation-related interests, braswell is quite lacking on the single-threaded CPU front, so that's a use-case you can eliminate since the emulators that its CPU would be fast enough for (e.g. consoles before 2000) are the kind that are more likely to use OpenGL anyway.
    what about slang shaders? they are for retroarch (emulation) but only work on vulkan, does that count for your interest? about cpu performance... yes its lacking, on my old e3500 desmume runs faster on windows than on my laptop with linux (e3500 has 2.7ghz vs 2.1 on n3050) however dolphin runs fine on the n3050 but not at all on the e3500.

    Leave a comment:

Working...
X