19:31 jekstrand: Of course nvidia has VERTEX_BEGIN_GL in hardware....
19:31 jekstrand:starts drafting a vkCmdVertex4f extension....
21:11 HdkR: jekstrand: Nvidia has tons of random garbage in hardware :P
21:11 HdkR: Gotta go fast
21:12 karolherbst: it's really weird, some stuff they don't have _at_all_ even though it's reasonable, and other things...
21:13 karolherbst: but the entire 2D class is fun, though one might argue you could just throw it away because "3D can do it" :P
21:13 HdkR: 2D, not the blitters specificallay?
21:13 karolherbst: well.. the blitter is quite limited :P
21:14 HdkR: But it's async, so you can do other stuff :D
21:14 karolherbst: :D
21:14 karolherbst: good to know I guess
21:22 jekstrand: karolherbst: You keep saying this and I have net to come across anything that's all that insane. :)
21:23 karolherbst: maybe I just never had to work with insane things
21:23 karolherbst: except maybe CL :P
21:24 karolherbst: but for whatever reason, nvidia also seems to be the only vendor putting the actual "architecture" id into mmio reg 0x0 where everything else just messes with stupid PCI IDs? so maybe I should just assume everything else is just that worse?
21:41 karolherbst: HdkR: soooo.. class numbers.. the upper byte, is that actual strict per generation or is it fine that c461 is turing, but c4c0 is volta?
21:41 karolherbst: I always though it's _fixed_ per gpu
21:41 karolherbst: uhm.. architecture
21:42 airlied: yeah I think they pick and choose sometimes
21:43 karolherbst: well... it's pretty much consistent overall, just c4 is "weird"
21:43 HdkR: Assume a mixed mess
21:44 karolherbst: :(
21:44 karolherbst: _of_course_
21:44 HdkR: A lot of those things are autogenerated so it can end up being whatever
21:44 karolherbst: ohh sure
21:45 karolherbst: but I expected that at least each generation has the "version" byte fixed... or something
21:45 karolherbst: maybe I expect too much
21:45 HdkR: Really doesn't matter when you just compile half a million lines of autogenerated code and don't think about it. Just look at amdgpu autogenerated headers :P
21:46 karolherbst: uhhh
21:46 karolherbst: I wanted to have code like if (dev->cls > 0xc3) .. use NVC3... stuff else {not supported or old stuff}
21:47 karolherbst: but I suspect we might need to do that on a per engine decision at some point
21:47 HdkR: Yea, you get a mix of engines so it doesn't really work
21:47 karolherbst: the awesome thing is, the kernel doesn't even tell us what's actually end up being used
21:47 HdkR: Like some parts using a generation old nvenc but with a new GPU core
21:47 karolherbst: ohh sure
21:48 karolherbst: but that's fine
21:48 karolherbst: so using older stuff is usually possible, unless you have a newer version of thing
21:48 HdkR: I'm still of the mind that if it looks stable, then hardcode until proven wrong. Just a smidge of refactoring
21:48 karolherbst: but I suspected that at least c5 is _always_ turing or ampere or whatever
21:51 HdkR: If you walk the entire family of GPUs, GA100, GA102, GA104, GA106, GA10B and it is stable, then it's stable right? :)
21:51 karolherbst: ....
21:51 karolherbst: dunno :P
21:52 HdkR: Worst case you get someone with some wacky GTX 3030 in the future or something that managed to change it
21:53 HdkR: oop, "RTX"
21:53 karolherbst: nah, 3030 would be GTX :P
21:53 karolherbst: oh nah
21:53 karolherbst: it would be 2630
21:53 karolherbst: sorry
21:54 HdkR: blegh
21:54 karolherbst: GTX 2630 the new entry level GPU for just $200
21:54 HdkR: GT 1010 weeping in a corner somewhere
21:57 karolherbst: I am already wondering what nvidia does after the 60 series
21:57 karolherbst: uhm
21:57 karolherbst: 50 actually
21:57 karolherbst: the new GeFource RTX 6800 Ultra Extreme Edition
21:58 HdkR: Go the Intel route, wrap around again and make it even longer
21:58 HdkR: 11987HX or something
21:58 karolherbst: "I bouth an GeFource 6800 GS, but somehow I can't play Fallout 5?!?! what's wrong"
21:58 karolherbst: *bought
21:59 karolherbst: mhhhhh
21:59 HdkR: hah
21:59 anarsoul: karolherbst: slim version, passive cooling
21:59 anarsoul: :P
22:00 karolherbst: you never had a 6800 do you :P
22:00 i509VCB: Just start using uuids for model numbers lol
22:00 karolherbst: back then GPUs had style
22:01 HdkR: and mermaids
22:01 karolherbst: and dragons
22:01 karolherbst: mine has an MSI fighter jet
22:02 anarsoul: karolherbst: nope, but there were 6800 with passive cooling
22:02 karolherbst: ehhh
22:02 karolherbst: welll
22:02 karolherbst: _maybe_?
22:02 HdkR: If you rip a fan off something it'll be passive
22:02 anarsoul: HdkR: you also need bulkier heatsink
22:02 karolherbst: I guess so
22:03 karolherbst: I guess passively cooling with a target of 125º is easier than 95ª
22:03 anarsoul: and a case with well-designed air flow
22:04 karolherbst: I don't know what's up with this gen, but those GPUs ran frigging hot
22:04 karolherbst: I have this mac mini and the GPU idles at 90º :'(