#
✕ nick:chadmed
#asahi · 2022-03-29 10:25
chadmedboth 8 cores per EU, both same FLOPS/MHz
#asahi · 2022-03-29 10:25
chadmedhmmm based on some very rough maths, it seems like the cores in the A15 GPU are the same as those in the G13/X
#asahi · 2022-03-29 09:54
chadmedbut do bear in mind the last comment on that PR, we want people to actually fix their software rather than rely on dirty kernel patches
#asahi · 2022-03-29 09:52
chadmedclone the repo, makepkg -csi
#asahi · 2022-03-29 09:52
chadmedits literally exactly the same
#asahi · 2022-03-29 08:03
chadmeddid you tell NetworkManager to use it as the backend?
#asahi · 2022-03-29 08:03
chadmedhm phoronix benchmarks from kernel 5.14 shows it being significantly slower for IO-heavy tasks
#asahi · 2022-03-29 08:01
chadmedi have not noticed any slowdown using btrfs on my nvme drives
#asahi · 2022-03-29 07:55
chadmedyou can put 4k sectors in 16k pages, you cant pack a 16k sector into a 4k page
#asahi · 2022-03-29 07:55
chadmedyeah because it will use 4k sectors
#asahi · 2022-03-29 07:15
chadmedbecause if the former, then that would use iwd. if the latter then installing gnome would have probably pulled in wpa_supplicant anyway
#asahi · 2022-03-29 07:15
chadmed the asahi image or debian?
#asahi · 2022-03-29 06:50
chadmedyou might have to build it youtself from source though
#asahi · 2022-03-29 06:50
chadmedtheres an android version so it definitely works on arm
#asahi · 2022-03-29 06:44
chadmedseems to be more reliable on these machines, integrates better with systemd and we support it
#asahi · 2022-03-29 06:44
chadmedyou should instruct users to use iwd instead of wpa_supplicant
#asahi · 2022-03-29 03:17
chadmedwhich is why software should not assume any particular alignment
#asahi · 2022-03-29 03:17
chadmedbut the aarch64 spec says that 4k, 16k and 64k pages are all fair game
#asahi · 2022-03-29 03:17
chadmed16k is faster for reasons
#asahi · 2022-03-29 03:17
chadmedthis particular hardware supports both 16k and 4k pages natively
#asahi · 2022-03-29 03:12
chadmedno software for aarch64 should have ever assumed a page size since it breaks spec, but we dont live in a fair or good world
#asahi · 2022-03-29 03:11
chadmedmuch harder to fix when your codebase relies on those assumptions to run fast
#asahi · 2022-03-29 03:11
chadmedthe fix is theoretically easy, stop using hardcoded page alignments
#asahi · 2022-03-29 03:11
chadmedthe latter is much the same but harder to fix
#asahi · 2022-03-29 03:10
chadmedin the former example, they might be statically linked against a malloc that assumes 4k pages, so when it goes to allocate the next page it goes page[0]+4k, and finds that its collided with page[0]
#asahi · 2022-03-29 03:10
chadmedits usually because they have their own malloc implementations or weird optimisations that rely on assuming a certain page alignment
#asahi · 2022-03-29 03:02
chadmednah not even the machine, just been neglecting work to focus on study and research this year
#asahi · 2022-03-29 03:01
chadmedbit povvo at the moment :P
#asahi · 2022-03-29 02:57
chadmedid also be compelled to replay deus ex mankind divided which i found bitterly disappointing
#asahi · 2022-03-29 02:56
chadmedwell i have a couple of games that have native metal ports, but to make room to install them id have to nuke this linux install again and i dont really fancy doing that
#asahi · 2022-03-29 02:37
chadmedA14 got 1.2
#asahi · 2022-03-29 02:37
chadmedaccording to wikipedia the A15 gets about 1.5 TFLOPS out of 5 cores in its power constrained envelope
#asahi · 2022-03-29 02:36
chadmedthe AGX block is going to be more interesting than the CPUs imo
#asahi · 2022-03-29 02:35
chadmedyeah
#asahi · 2022-03-29 02:34
chadmedi expect the base M2 would probably handily outdo an M1 Max in single threaded and lightly threaded applications
#asahi · 2022-03-29 02:33
chadmedmore interesting will be how little power they draw
#asahi · 2022-03-29 02:33
chadmedAvalanche and Blizzard are both about a 10% jump in IPC so its a decent uplift
#asahi · 2022-03-29 02:32
chadmedwhich makes sense because the A14 Firestorm and Icestorm were 0x21 and 0x20 respectively
#asahi · 2022-03-29 02:32
chadmedyeah id guess it means theyre different revisions
#asahi · 2022-03-29 02:31
chadmedMIDR_EL1 is a per-core thing
#asahi · 2022-03-29 02:30
chadmedMIDR_EL1 has the pro/max as 0x24 and 0x25
#asahi · 2022-03-29 02:30
chadmed0x22 and 0x23 are meant to be firestorm and icestorm
#asahi · 2022-03-29 02:30
chadmedthe cores in the m1 pro/max are different variants
#asahi · 2022-03-29 02:30
chadmedhm interesting
#asahi · 2022-03-29 02:26
chadmedimplementer id* sorry
#asahi · 2022-03-29 02:26
chadmedyeah apple have their own vendor id in the arm scheme
#asahi · 2022-03-29 02:21
chadmedfitting name but it can only work on Apple hardware
#asahi · 2022-03-29 02:14
chadmedyeah one thing that would be nice is mapping ctrl+shift to cmd in konsole even
#asahi · 2022-03-29 02:10
chadmedkwin already has a ridiculous amount of customisation though
#asahi · 2022-03-29 02:10
chadmedif you want that level of keybind customisation just use dwm or something

plus ›