| 19 Jul 2024 |
ElvishJerricco | I suspect we're having different issues | 23:50:42 |
woobilicious | I don't use ibus, so probably not. | 23:50:49 |
ElvishJerricco | (ibus is just started by gnome no matter what, so I don't intentionally use it :P) | 23:51:15 |
woobilicious | Ahh right, wonder if sway is similar lol | 23:51:32 |
ElvishJerricco | I'm mostly surprised that other gnome users haven't been reporting the same issue. There was an issue report linked earlier in this room but I'm not entirely convinced it's the same | 23:52:32 |
woobilicious | lmao, I think I remember something similar from like 2 years ago when gnome migrated to wayland, the halo plasma pistol would charge, instead of fire single shots if you rapidly tapped mouse 1. | 23:58:01 |
| 20 Jul 2024 |
| CodeMichael joined the room. | 12:44:29 |
mjm | yayy just installed the new 7800xt, and it doesn't suffer the weird fps lag when obs is running like my nvidia card did | 14:42:15 |
K900 | Enjoy drivers that are still not great but at least you can debug them | 14:46:43 |
mjm | it's something. seems to be working better overall but time will tell | 14:47:51 |
mjm | just being able to actually get 60fps while using OBS is big to me | 14:52:16 |
| 21 Jul 2024 |
woobilicious | GPUs aren't preemptively multi-tasking so you'll always have issues from what I gather. high load games tend to overly stress sway on my system on occasion. | 00:08:30 |
woobilicious | https://www.youtube.com/watch?v=09vxZJ7HeC4 | 00:17:14 |
K900 | In reply to @woobilicious:matrix.org GPUs aren't preemptively multi-tasking so you'll always have issues from what I gather. high load games tend to overly stress sway on my system on occasion. They actually absolutely are! | 06:26:47 |
K900 | Any modern GPU has a priority system | 06:26:52 |
woobilicious | priority systems can work with co-operative systems, the question is can you stop a while (true) loop externally (i.e. via an interrupt). There's some research papers from like 2014 but no info I can find indicates that a modern GPU has hardware features to prevent tasks taking too much GPU due to bugs etc. | 07:07:48 |
K900 | There is most definitely preemption on modern GPUs | 07:16:38 |
K900 | https://learn.microsoft.com/en-us/windows-hardware/drivers/display/gpu-preemption | 07:16:39 |
K900 | Windows because that was the first thing I found | 07:16:46 |
woobilicious | I'm not convinced the granularity level requirement is enough to prevent stalls at the same level as a CPU thread would get via timer-based interrupts. though I'm not a GPU driver writer shrug. | 07:47:48 |
woobilicious | I found this oldish quote: "There are also pages later in the Pascal whitepaper that provide some details on its thread-level preemption capability, which is a big improvement on the draw-level preemption supported by the earlier Maxwell 2.0 architecture. Interestingly it seems their hardware also features instruction-level preemption, but only for CUDA." | 07:57:51 |
K900 | In reply to @woobilicious:matrix.org I'm not convinced the granularity level requirement is enough to prevent stalls at the same level as a CPU thread would get via timer-based interrupts. though I'm not a GPU driver writer shrug. It's pretty close, because shaders are generally VERY small | 08:11:46 |
woobilicious | I guess it depends on what you mean by small, I have many games that drop below 90fps and affect the fps of sway itself. | 09:27:05 |
K900 | That's likely just an implicit sync issue | 09:36:06 |
K900 | Which is distinct from preemption | 09:36:11 |
woobilicious | I'm not even sure what shader level granularity means on EP workloads with only a fraction of the frame buffer in flight at any one time. Are we rendering the entire frame buffer and then yielding or does the GPU submit each region separately | 09:40:41 |
K900 | The GPU has much higher granularity anyway | 09:43:11 |
K900 | The time of "all the shaders at the same time" is long gone | 09:43:20 |
K900 | https://dl.acm.org/doi/fullHtml/10.1145/3453417.3453432 | 09:43:46 |
woobilicious | I guess I don't have nearly the same level of trust on the average game dev and writing your average Turing complete shader to yield quickly lol | 09:49:19 |