#
#asahi · 2025-01-17 23:26
JamesB192HDMI alt mode exists in theory; nobody licenses it, though, except for a few non-production devices. DP alt mode and a generic DP->HDMI converter power the MacBook Pro HDMI ports.
#asahi · 2025-01-17 23:21
bmc77Unless I'm overlooking some part of that
#asahi · 2025-01-17 23:21
bmc77JamesB192: That seems to be just about DP Alt Mode (which I know doesnt work at all)... I'm talking about HDMI connection, which works very well (besides the sleep/resume thing I mentioned)...
#asahi · 2025-01-17 23:10
JamesB192Bmc77: https://www.reddit.com/r/AsahiLinux/comments/1i0h3do/when_simple_linux_subsystems_collide_with_complex/
#asahi · 2025-01-17 23:09
JamesB192Pitust: IIUC those counters have been problematic for a while.
#asahi · 2025-01-17 23:08
bmc77bummer. do you have that reddit link?
#asahi · 2025-01-17 23:07
JamesB192Bmc77: it's answered in the subreddit already. Though I saw it on Mastodon first. TLDR: it only worked before because of bugs interacting.
#asahi · 2025-01-17 21:31
pitustThis doesn't seem to match my understanding of the apple M1 PMC? Which appears to support them just fine
#asahi · 2025-01-17 21:31
pitustI'm having some trouble with profling a KVM virtual machine; perf kvm record fails with 'apple_firestorm_pmu/cycles/Pu: PMU Hardware doesn't support sampling/overflow-interrupts'
#asahi · 2025-01-17 21:04
bmc77Okay, I've already asked this twice. I'm going to ask again and if no answer then go to the subreddit... Has there been any update on the instability of external monitors re: sleep and resume / connect and reconnect? (HDMI, M2)
#asahi · 2025-01-17 20:39
fl0_idbmc77 oh and with a recent version of podman, you can also run linux containers on macos that use the metal gpu
#asahi · 2025-01-17 20:32
fl0_idinstead of trying to use nvidia
#asahi · 2025-01-17 20:32
fl0_idideally, setting device on pytorch to 'mps' will use the apple gpu
#asahi · 2025-01-17 20:32
fl0_idyou can f.e. also get jupyter lab notebooks that are mostly set up for using the gpu. they can teach you a lot more
#asahi · 2025-01-17 20:31
bmc77Thats what I meant / why I was drawn to it when I saw it! haa
#asahi · 2025-01-17 20:31
fl0_idI would forget about ANE (neural engine) though, unless you are deploying a product/app.
#asahi · 2025-01-17 20:30
fl0_idbmc77 you don't need any math for playing around with ollama. in facts it's probably counterproductive ;)
#asahi · 2025-01-17 20:21
bmc77Yeah, just sort of trying to get the foot in the door for the whole machine learning thing. I figure going the practical/hands on way of learning would be best. I knew all the math in college but I would be embarassed to reveal how little I remember
#asahi · 2025-01-17 20:19
chaos_princessbut if it is for learning, i'd say stick to macos for now, less chances that you will have to fix irrelevant nonsense.
#asahi · 2025-01-17 20:19
chaos_princessJust don't bother, stochastic parrots are overrated anyway :P
#asahi · 2025-01-17 20:18
bmc77Or would learning be precisely what I would want to do on macOS
#asahi · 2025-01-17 20:14
bmc77So I would be fine with either macOS or asahi for learning?
#asahi · 2025-01-17 20:12
bmc77*discussion flies over my head*
#asahi · 2025-01-17 11:25
fl0_idJamesB192 do it haha
#asahi · 2025-01-17 11:12
JamesB192is almost tempted to get the ANE driver and ollama.
#asahi · 2025-01-17 11:09
JamesB192ANE code was last touched 2023 July 9, after that the 'docs' were removed from the tree
#asahi · 2025-01-17 11:04
fl0_idthere are also some ANE/CoreML models on HF https://huggingface.co/apple
#asahi · 2025-01-17 11:01
fl0_idwhile ANE is more efficient, for learning about it gpu is better imo, as it is more easily transferrable to cuda etc
#asahi · 2025-01-17 11:01
fl0_idhttps://www.reddit.com/r/LocalLLaMA/comments/1de0fkx/coreml_vs_ollama_performance_on_macos/ has some discussion on this
#asahi · 2025-01-17 10:59
fl0_idwhich uses gpu
#asahi · 2025-01-17 10:59
fl0_idpretty sure ollama etc does not use ANE just metal
#asahi · 2025-01-17 10:46
mazpitust: the timers are working correctly. the vgic implementation is a nasty piece of shit.
#asahi · 2025-01-17 09:45
PaulFertserpitust: but does Ollama support ANE?
#asahi · 2025-01-17 09:29
pitustbmc77: macOS would be better; linux does not have support for ANE at the moment
#asahi · 2025-01-17 09:28
pitustnice9cel: DTrace works, so, yes, eBPF is supported
#asahi · 2025-01-17 05:14
bmc77(wanting to learn more about LLMs and so on)
#asahi · 2025-01-17 05:13
bmc77If I want to use something like Ollama would it be best to use macOS (with the 16 core neural engine) or linux? or would there not be much of a difference?
#asahi · 2025-01-17 03:02
JamesB192CONFIG_HAVE_EBPF_JIT=y ?
#asahi · 2025-01-17 00:50
nine9celi see some of the BPF options are enabled in that config but some are absent
#asahi · 2025-01-17 00:49
nine9celforgive me, i'm a novice when it comes to this stuff. do you know if asahi supports eBPF by default? like so i can use bpftrace and BCC
#asahi · 2025-01-17 00:49
nine9celthanks
#asahi · 2025-01-17 00:46
j`eyhttps://copr-dist-git.fedorainfracloud.org/cgit/@asahi/kernel/kernel.git/tree/kernel-aarch64-16k-fedora.config
#asahi · 2025-01-17 00:43
nine9celi found this https://github.com/AsahiLinux/linux/blob/asahi/arch/arm64/configs/defconfig but i don't know if that's actually what gets used
#asahi · 2025-01-17 00:42
nine9celis there somewhere online i can see the kernel config used for asahi
#asahi · 2025-01-17 00:41
nine9celhey bros
#asahi · 2025-01-16 23:24
pituster, virtual timer, not physical
#asahi · 2025-01-16 23:23
pitustSetting it to *any* lower value does not
#asahi · 2025-01-16 23:22
pitustBackground: i'm debugging an OS, and resetting the physical timer threshold to 0xffff...ffff (which the arm spec says "might never trigger", makes the interrupt trigger constantly
#asahi · 2025-01-16 23:21
pitustIs there any known M1/arm non-compliance issues around physical timers/vGIC?
#asahi · 2025-01-16 12:50
Juestso apparently the website and presence of the kernel-mc guy are down because of the australia nsw storms

plus ›