04:02 fdobridge_: <p​lagman> https://cdn.discordapp.com/attachments/1034184951790305330/1185794063388123156/HatInTime_Blob.png?ex=6590e7c7&is=657e72c7&hm=b0abf92de93dcf8a523904b24e074a0af92913ca94fffc6198026057214086ec&
04:02 fdobridge_: <p​lagman> https://cdn.discordapp.com/attachments/1034184951790305330/1185794064004694036/HatInTime_NVK.png?ex=6590e7c7&is=657e72c7&hm=bb211e20357556cfa2831178c3c0d02ec9c4e9f727e819c78b991207acd8bd23&
04:02 fdobridge_: <p​lagman> same scene on a GA102, NVK: 210FPS Blob: 165FPS
04:02 fdobridge_: <p​lagman> the blob is getting less gpu occupancy i think
04:05 HdkR: Smells like winning to me
05:03 fdobridge_: <s​amantas5855> 165 fps is an oddly specific number, how many hz is the screen
05:06 fdobridge_: <p​lagman> 120, it's also oscillating a bunch, it's 140-165
05:12 fdobridge_: <s​amantas5855> insane
06:26 fdobridge_: <g​fxstrand> Damn!
06:27 fdobridge_: <g​fxstrand> I'll take it! 😅
13:58 fdobridge_: <d​jdeath3483> https://tenor.com/view/intel-gif-25435404
13:58 fdobridge_: <d​jdeath3483> Ahah
14:16 fdobridge_: <b​utterflies> impossible?
14:20 fdobridge_: <e​sdrastarsis> Theres a discussion about that here: https://discord.com/channels/1033216351990456371/1034184951790305330/1177638160193306704
14:21 fdobridge_: <b​utterflies> ugh, D3D12 and Vulkan ended up fulfilling only a small part of that low level API promise except the memory management pain 😦
14:23 fdobridge_: <b​utterflies> to sum it up: No.
14:23 fdobridge_: <b​utterflies> there's a really big impedance mismatch between a6xx and D3D12
14:26 fdobridge_: <r​hed0x> interesting, whats the mismatch?
14:32 fdobridge_: <b​utterflies> RenderPass on D3D12 was originally just broken in practice... (so Adreno spends ~100% of its time on Windows in IMR mode) is a big one
14:32 fdobridge_: <b​utterflies> but there are others too 😬
14:32 fdobridge_: <r​hed0x> doesnt really matter, nothing uses that anyway 🙃
14:33 fdobridge_: <r​hed0x> and hopefully their renderpass api changes fix that
14:34 fdobridge_: <b​utterflies> it's sad to see the perf difference when comparing D3D11 to D3D12 on A6xx
14:35 fdobridge_: <b​utterflies> that said things have slowly been getting better
15:39 fdobridge_: <g​fxstrand> Not impossible, but really annoying.
16:01 fdobridge_: <b​ylaws> Stray performs better with Tu and dxvk than stock dx11/12 drivers
16:01 fdobridge_: <b​ylaws> on x13s
16:03 fdobridge_: <r​hed0x> 12 too?
16:03 fdobridge_: <r​hed0x> oh wait, stray always uses d3d11
16:03 fdobridge_: <r​hed0x> does qualcomm have an actual d3d11 driver or do they use 11on12?
16:04 fdobridge_: <r​hed0x> because 9on12 and 11on12 get annihilated by DXVK 🙃
16:06 fdobridge_: <b​ylaws> 12 too
16:06 fdobridge_: <b​ylaws> At least with -dx12
16:31 fdobridge_: <!​DodoNVK (she) 🇱🇹> Will DX12 be supported on Cassia? 🤔
16:37 fdobridge_: <r​hed0x> there's no reason not to support it on Qualcomm GPUs with turnip running
16:38 fdobridge_: <!​DodoNVK (she) 🇱🇹> Is sparse a problem though? Intel suffered a lot with that
16:56 fdobridge_: <b​ylaws> And blob :)
16:56 fdobridge_: <b​ylaws> Adreno 7xx blob has sparse
16:57 fdobridge_: <b​ylaws> Tu lacks it cause msmdrm has no interface rn
17:45 fdobridge_: <k​arolherbst🐧🦀> ~~probably higher fps because of lower rendering quality~~ (reference to a discussion I had in the past :ferrisUpsideDown: )
20:17 fdobridge_: <S​id> tried out nvk again on 6.7-rc5 with whatever commit was head on master an hour ago and I'm still facing the full-system-freezes I'd mentioned a while ago when dxvk is in use. turing card, GSP enabled
20:21 fdobridge_: <S​id> probably worth mentioning it's a prime setup running on plasma wayland as well
20:24 fdobridge_: <S​id> unable to reproduce the issue when gsp is not in use, so I have good reason to believe this is a kernel thing
20:26 fdobridge_: <a​irlied> nothing in the logs after a reboot?
20:28 fdobridge_: <S​id> nope, the journal for that boot simply vanishes if it freezes and I hard reboot
20:28 fdobridge_: <a​irlied> can you ssh in and leave a dmesg -w running?
20:28 fdobridge_: <S​id> it pretty much is killing everything, because even ssh stops receiving/sending anything
20:36 fdobridge_: <S​id> ...this is definitely a kernel thing but likely not a GSP thing
20:36 fdobridge_: <a​irlied> yeah a dmesg -w might capture the start of it crashing before it takes out the system
20:37 fdobridge_: <S​id> upon further inspection DRI_PRIME=1 is still putting the load on my iGPU (checked with nvtop)
20:37 fdobridge_: <S​id> so possibly i915 bug?
20:37 fdobridge_: <S​id> I'll try specifying the ICD
20:51 fdobridge_: <S​id> yeah, i915 thing
20:51 fdobridge_: <S​id> ```
20:51 fdobridge_: <S​id> [ 1462.942511] msg: 00000000: 22 00 00 00 "...
20:51 fdobridge_: <S​id> [ 1463.082124] nouveau 0000:01:00.0: gsp:msg fn:4123 len:0x24/0x4 res:0x0 resp:0x0
20:52 fdobridge_: <S​id> [ 1463.082139] msg: 00000000: 23 00 00 00 #...
20:52 fdobridge_: <S​id> [ 1479.507711] Asynchronous wait on fence 0000:00:02.0:kwin_wayland[885]:2d0a2 timed out (hint:intel_atomic_commit_ready [i915])
20:52 fdobridge_: <S​id> [ 1479.507961] Asynchronous wait on fence drm_sched:nouveau_sched:328 timed out (hint:submit_notify [i915])
20:52 fdobridge_: <S​id> [ 1490.607737] Asynchronous wait on fence 0000:00:02.0:kwin_wayland[885]:2d0b2 timed out (hint:intel_atomic_commit_ready [i915])
20:52 fdobridge_: <S​id> [ 1490.607993] Asynchronous wait on fence drm_sched:nouveau_sched:329 timed out (hint:submit_notify [i915])
20:52 fdobridge_: <S​id> [ 1501.707763] Asynchronous wait on fence 0000:00:02.0:kwin_wayland[885]:2d0ca timed out (hint:intel_atomic_commit_ready [i915])
20:52 fdobridge_: <S​id> [ 1501.708019] Asynchronous wait on fence drm_sched:nouveau_sched:329 timed out (hint:submit_notify [i915])
20:52 fdobridge_: <S​id> [ 1512.574454] Asynchronous wait on fence 0000:00:02.0:kwin_wayland[885]:2d0e0 timed out (hint:intel_atomic_commit_ready [i915])
20:52 fdobridge_: <S​id> [ 1512.574713] Asynchronous wait on fence drm_sched:nouveau_sched:329 timed out (hint:submit_notify [i915])
20:52 fdobridge_: <S​id> [ 1523.674481] Asynchronous wait on fence 0000:00:02.0:kwin_wayland[885]:2d0e4 timed out (hint:intel_atomic_commit_ready [i915])
20:52 fdobridge_: <S​id> [ 1523.674741] Asynchronous wait on fence drm_sched:nouveau_sched:329 timed out (hint:submit_notify [i915])
20:52 fdobridge_: <S​id> ```
20:53 fdobridge_: <S​id> I'll take a better look tomorrow, almost 2:30am already...
20:53 airlied: need to track down what is happening there, if nouveau craaps out and blocks i915
20:54 fdobridge_: <S​id> I've had something similar happen on the proprietary driver a few times without gsp as well, but with nouveau it's extremely consistent
21:13 fdobridge_: <b​utterflies> d3d9 goes through d3d9on12. d3d11 and d3d12 are native
21:26 fdobridge_: <!​DodoNVK (she) 🇱🇹> I definitely experience this too