#
✕ nick:marcan
#asahi · 2022-11-01 12:56
marcandoes that show a 'cp' command?
#asahi · 2022-11-01 12:56
marcanzcat /boot/initramfs-linux-asahi.img | grep -a vendorfw.manifest
#asahi · 2022-11-01 12:55
marcan(or wherever your ESP is mounted)
#asahi · 2022-11-01 12:55
marcanalso, does /boot/vendorfw/firmware.cpio exist?
#asahi · 2022-11-01 12:54
marcandoes /proc/device-tree/chosen/asahi,efi-system-partition exist?
#asahi · 2022-11-01 12:52
marcanafter upgrading, and is your bootloader actually using the updated initramfs?
#asahi · 2022-11-01 12:51
marcanand did you update your initramfs?
#asahi · 2022-11-01 12:50
marcanare you using the official boot stuff?
#asahi · 2022-11-01 12:50
marcanas of now, yes
#asahi · 2022-10-31 12:47
marcan77% -> 29% is actually kind of sensible for overnight s2idle too, so I don't know what made it crash but assuming the 29% on startup was real, that's not unreasonable power draw either
#asahi · 2022-10-31 12:45
marcanI never said it was like macOS already, the next big ticket item is cpu deep idle
#asahi · 2022-10-31 12:45
marcanbut also 30% battery with my numbers means it'll only last ~9h or so, so assuming you get a good night's sleep, it's entirely expected that you'd wake up to it empty
#asahi · 2022-10-31 12:44
marcans2idle is only tested not to crash, nobody has done power measurements in s2idle yet
#asahi · 2022-10-31 12:44
marcanit's entirely possible and likely that s2idle is triggering issues that actually drain the battery under whatever circumstances
#asahi · 2022-10-31 12:44
marcanI only tested simply leaving the machine idle
#asahi · 2022-10-31 12:43
marcanjoske: I never said s2idle actually worked
#asahi · 2022-10-29 12:02
marcansighs
#asahi · 2022-10-27 08:46
marcannor whether there is anything for the link
#asahi · 2022-10-27 08:46
marcanthis is deep memory controller shenanigans
#asahi · 2022-10-27 08:46
marcanbut I wouldn't know how to find out
#asahi · 2022-10-27 08:45
marcanbut yeah, traditional ECC is dead with LPDDR, though conversely I think ECC is becoming a necessity with LPDDR5 densities, so there's a good chance there's RAM-internal ECC support
#asahi · 2022-10-27 08:45
marcanthe caches do have ECC though, that I know
#asahi · 2022-10-27 08:44
marcanhowever, there could be ECC internal to the RAM itself, and some form of ECC used at the protocol level
#asahi · 2022-10-27 08:44
marcanthe channels are too narrow
#asahi · 2022-10-27 08:43
marcanrann: I don't think traditional ECC is or can be supported in LPDDR systems
#asahi · 2022-10-27 07:27
marcan(we still need to do something about them...)
#asahi · 2022-10-27 07:10
marcanthose errors are normal
#asahi · 2022-10-27 05:10
marcan*secondaries
#asahi · 2022-10-27 05:10
marcanchadmed: make sure you enable the MMU on the seconaries, otherwise all mem accesses will be super slow
#asahi · 2022-10-27 04:47
marcanah, these are from SMC? or Ecounters?
#asahi · 2022-10-27 04:46
marcanthe rest look reasonable though!
#asahi · 2022-10-27 04:46
marcanchadmed: the first couple lines there for the E cores are suspicious, I wouldn't expect efficiency to drop at the lowest Pstate (the lowest pstate needs to be the peak efficiency, otherwise it makes no sense)
#asahi · 2022-10-26 14:12
marcanborisk: that makes sense, mostly because macOS is quite a bit worse at VM/FS stuff than Linux AIUI. lots of people report better performance under Linux than macOS.
#asahi · 2022-10-25 15:06
marcan(much better OoO on Apple)
#asahi · 2022-10-25 15:06
marcanj`ey: yeah (though that probably helps intel close the IPC gap with Apple more than anything else, since Apple are much better at that)
#asahi · 2022-10-25 15:05
marcanborisk: ^
#asahi · 2022-10-25 15:05
marcanif anything the Max being only 30% slower is pretty awesome
#asahi · 2022-10-25 15:05
marcanyou're comparing a 16 core processor that can turbo at 5.2 GHz and burn power vs. a 10 core processor that maxes out at 3 GHz multicore and uses 1/3 of the power
#asahi · 2022-10-25 15:03
marcanso I really don't see what the problem is
#asahi · 2022-10-25 15:03
marcan30% slower sounds completely expected for these CPUs
#asahi · 2022-10-25 14:31
marcanI imagine all cores doing stuff at once should be sufficient to keep the RAM awake anyway
#asahi · 2022-10-25 14:30
marcan(increasing the memory power management timeout to reduce latency on first access after some idle time when the CPU is busy)
#asahi · 2022-10-25 14:30
marcanOTOH, there is a memory latency related tweak we are not applying right now, because it didn't make much of a difference for the tests I did. I think it mostly only affects certain single-core type workloads though
#asahi · 2022-10-25 14:28
marcanit should be memory *latency* bound
#asahi · 2022-10-25 14:28
marcanalso, I don't think compilation is memory bandwidth bound at all?
#asahi · 2022-10-25 14:25
marcanthus: get an M1 ultra and it'll be better than that intel and use less power :p
#asahi · 2022-10-25 14:25
marcanconsidering it has more cores and easily eats 190W at full CPU throttle, while the M1 Max will push 60W or so, that makes the M1 Max more than 2x as efficient as the intel
#asahi · 2022-10-25 14:22
marcangeekbench says the intel has about 50% better performance
#asahi · 2022-10-24 16:25
marcannull: we should probbly move to #asahi-dev
#asahi · 2022-10-24 15:10
marcanso it's either come up with a new standard to stay in EL2, or drop the kernel to EL1 and then you lose virtualization support

plus ›