#asahi-dev · 2026-01-27 11:13
integralpilotim just curious how they changed the ISA, it's probably a lot i'm thinking#asahi-dev · 2026-01-27 11:12
noopwafelthere is going to be more than enough to do :-D#asahi-dev · 2026-01-27 11:12
integralpiloti really like compilers#asahi-dev · 2026-01-27 11:12
integralpilotif you want to do the kenrel side of things im happy to do the userspace side of things#asahi-dev · 2026-01-27 11:12
integralpilotoh okay that sounds cool!#asahi-dev · 2026-01-27 11:12
jannauthe m2 ultra is a little short on PCIe lanes (16 + 8) for its case / number of slots. the massive PCIe switch mitigstes that. only being PCIe 4 doesn't help either#asahi-dev · 2026-01-27 11:11
noopwafelmy intention is to do M3 gpu bringup (and the rest of this is just a distraction because I want to be able to run code :p)#asahi-dev · 2026-01-27 11:11
noopwafelintegralpilot: I have a bunch of work on the AGX tracer btw#asahi-dev · 2026-01-27 11:10
jannauampere has a workaround of trapping unaligned access#asahi-dev · 2026-01-27 11:08
jannaunvidias kernel and user space driver work on the mac pro, I'd expect they work via thunderbolt as well#asahi-dev · 2026-01-27 11:08
integralpilothigh performance drivers on macOS wouldn't really work either as they've "deprecated" (and made nearly impossible to guide a user to install) kernelspace drivers, and only offer a limited userspace driver api#asahi-dev · 2026-01-27 11:07
nickchanon marcan's socials of course which is deleted#asahi-dev · 2026-01-27 11:07
jannauthe mac pro is supported except for the main installer since the fedora images are too old. Will be fine with Fedora 43#asahi-dev · 2026-01-27 11:07
chadmedthe other problem is, no one is writing a driverkit/iokit/whateverkit pcie driver for their widget unless apple pays them, and apples not going to pay them because apple's got their own hardware#asahi-dev · 2026-01-27 11:07
nickchani thought tencent had a workaround for the memory model thing#asahi-dev · 2026-01-27 11:06
jannauthe base m3 macbook pro and the m3 imacs are still using i2c usb-pd controllers#asahi-dev · 2026-01-27 11:06
chadmediirc the rpi 4 suffered from a similar problem, and even with revolting disgusting amdgpu kernel driver hacks to make it "work" it was abysmally slow #asahi-dev · 2026-01-27 11:05
chadmedits not that pcie over tbt doesnt work, its that the thing doesnt support the memory model that gpus assume exists for performance hacks #asahi-dev · 2026-01-27 11:05
jannauisp/webcam can probably a optional initial feature as well for m3 devices#asahi-dev · 2026-01-27 11:05
integralpilotmaybe they'll explore that prehaps#asahi-dev · 2026-01-27 11:05
integralpilotis tbt pcie still not workable on newer socs?#asahi-dev · 2026-01-27 11:04
chadmedyeah#asahi-dev · 2026-01-27 11:04
chaos_princessoh, i expect that that thing fully dies. studio only from now on#asahi-dev · 2026-01-27 11:04
NickChan[m]i don't expect anything either other than raw specs more suitable for a macpro (but still no PCI GPU etc)#asahi-dev · 2026-01-27 11:03
chaos_princessi don't expect anything#asahi-dev · 2026-01-27 11:03
chadmedwell they glued on a whole second pcie controller and that still couldnt save it. i dont expect to see another cheese grater. the m5 ultra will probably go in a studio#asahi-dev · 2026-01-27 11:02
NickChan[m]m2 ultra is clearly not a good fit for macpro and a fallback, and we will see about m5 ultra#asahi-dev · 2026-01-27 11:02
chadmedyeah that ^ #asahi-dev · 2026-01-27 11:02
chadmedpretty sure their private compute stuff is rackmounted apple silicon #asahi-dev · 2026-01-27 11:02
integralpilotthey're doing it to help their servers#asahi-dev · 2026-01-27 11:02
chaos_princessand server-ized macs for "private clould compute"#asahi-dev · 2026-01-27 11:02
NickChan[m]or for PCC rack m2 ultra#asahi-dev · 2026-01-27 11:02
integralpilotyeah @apple.com ppl contribute a lot to like kernel core stuff, so i guess that makes sense#asahi-dev · 2026-01-27 11:02
chaos_princessgeneric linux boxes#asahi-dev · 2026-01-27 11:01
integralpiloti wonder what apple uses for their own servers?#asahi-dev · 2026-01-27 11:01
chadmedmaybe if it had more real pcie lanes it wouldve been interesting #asahi-dev · 2026-01-27 11:00
chadmedi half expected it to be a test bed for a new server platform but they never followed through on that#asahi-dev · 2026-01-27 11:00
chaos_princessyes#asahi-dev · 2026-01-27 11:00
integralpilotthat's the cheese grater right?#asahi-dev · 2026-01-27 11:00
chaos_princessprobably because they promised a full arm transition#asahi-dev · 2026-01-27 11:00
chadmedidek why they released it tbh its so obvious the studio is supposed to replace it#asahi-dev · 2026-01-27 10:59
chadmedlast time i checked there were single digit install attempts so....#asahi-dev · 2026-01-27 10:58
chaos_princessjan_ne has one, and the device tree#asahi-dev · 2026-01-27 10:58
NickChan[m]on the other hand, j180d is still not supported (who actually uses that machine)#asahi-dev · 2026-01-27 10:57
chadmedi wonder if their rf chips will be as black boxy#asahi-dev · 2026-01-27 10:56
sventhe magic all happens inside atcphy#asahi-dev · 2026-01-27 10:56
sveni think it's two additional ones for the co-processor and NHI#asahi-dev · 2026-01-27 10:55
sventbt? nope#asahi-dev · 2026-01-27 10:55
NickChan[m]that thing must have plenty of tunables#asahi-dev · 2026-01-27 10:55
chadmedi dont think theres any real reason to stage trivial changes downstream anymore unless the whole driver is also still downstream