#
…

2023-11-07

00:03Armlin has quit [Ping timeout: 480 seconds]
00:06Tramtristmarcan which M3 model did you buy?
00:08j`ey14", but not sure of the other specs
00:08Tramtristyeah wondering based on that SS
00:09Tramtristif i decide to buy one .. should match whatever he got 😬
00:09jannegI'd assume a 16 core M3 Max as it is planned to be used as main "workstation"
00:35Armlin has joined #asahi
00:43Armlin has quit [Ping timeout: 480 seconds]
00:51dgb has quit [Quit: dgb]
00:51dgb has joined #asahi
00:57dgb has quit [Quit: dgb]
00:57dgb has joined #asahi
01:30Armlin has joined #asahi
01:31Armlin has quit [Remote host closed the connection]
01:33Armlin has joined #asahi
01:35Armlin has quit [Remote host closed the connection]
01:50Armlin has joined #asahi
01:57ciiur^ has joined #asahi
01:58Armlin has quit [Ping timeout: 480 seconds]
02:03Armlin has joined #asahi
02:05Armlin has quit [Remote host closed the connection]
02:23hightower3 has joined #asahi
02:30hightower2 has quit [Ping timeout: 480 seconds]
03:16Armlin has joined #asahi
03:37Armlin has quit [Remote host closed the connection]
03:38crabbedhaloablut has joined #asahi
03:43Armlin has joined #asahi
03:50Armlin has quit [Remote host closed the connection]
03:51Armlin has joined #asahi
04:08marvin24 has joined #asahi
04:11marvin24_ has quit [Ping timeout: 480 seconds]
04:12Armlin has quit [Remote host closed the connection]
04:41Armlin has joined #asahi
04:50Armlin has quit [Ping timeout: 480 seconds]
05:23ptudor has quit [Read error: Connection reset by peer]
05:34gspbirel56 has quit [Quit: Ping timeout (120 seconds)]
05:50Armlin has joined #asahi
05:54gspbirel56 has joined #asahi
06:00lena6 has quit [Ping timeout: 480 seconds]
06:06john-cabaj has joined #asahi
06:20Armlin has quit [Remote host closed the connection]
06:25Armlin has joined #asahi
06:28Armlin has quit [Remote host closed the connection]
06:31nicolas17 has quit [Ping timeout: 480 seconds]
06:33Armlin has joined #asahi
06:34Armlin has quit [Remote host closed the connection]
06:35Armlin has joined #asahi
06:38amwI think there is a generic bug in simpledrm driver - as this fixes it and allows X to work again for me - https://paste.debian.net/1297396/
06:40amwAs per my report - https://oftc.irclog.whitequark.org/asahi/2023-11-05#32627097 - simpledrm was failing to allocate memory due to conflict and thus causing X not to have a fb and fail
06:42amwAccording to https://github.com/AsahiLinux/linux/blob/asahi/drivers/video/aperture.c#L27 a FB driver should call aperture_remove_conflicting_devices() in it's probe function to release it from an earlier driver
06:44amwFor some reason my 6.4 kernel worked but in 6.5 efifb driver has grabs the memory first and then simpledrm tries to grab it again later and fails and results in NO FB device available at all...
06:45amwIf anyone has a better place or way to report this issue I'm all ears - thanks
06:45Armlin has quit [Remote host closed the connection]
06:46Armlin has joined #asahi
06:49capta1nt0ad has joined #asahi
06:54Armlin has quit [Ping timeout: 480 seconds]
07:02Armlin has joined #asahi
07:05john-cabaj1 has joined #asahi
07:05john-cabaj has quit [Read error: Connection reset by peer]
07:05john-cabaj1 is now known as john-cabaj
07:11Armlin has quit [Ping timeout: 480 seconds]
07:22Armlin has joined #asahi
07:29john-cabaj has quit [Quit: john-cabaj]
07:30john-cabaj has joined #asahi
07:32jannegamw: simpledrm can't use aperture_remove_conflicting_devices since that might remove the the aperture from the hw specific device driver depending on probe order
07:33amwok - but it works for me - so perhaps someone who understands this has a better fix...?
07:33jannegnot sure what we can do to resolve the conflict between simpledrm and efifb
07:33janneguse the dcp driver
07:34jannega better workaround would be either not to compile the efifb driver or block its loading
07:35Armlin has quit [Ping timeout: 480 seconds]
07:36amwjanneg: Those all seem like patching over the issue not really addressing it properly - but perhaps that's too hard. Thnks
07:38janneghow is using the proper display driver papering over the issue? both efifb and simpledrm are good enough for their purpose: displaying something until the real display driver can be loaded
07:40kunev has quit [Ping timeout: 480 seconds]
07:40Armlin has joined #asahi
07:43jannegsolving usage priority issues between efifb and simpledrm seems mostly a non-problem for startup
07:46jannegfeel free to ask in #dri-devel or on https://lore.kernel.org/dri-devel/
07:46jannegbut I think the short/mid term plan for fedora is to disable efifb. see https://lore.kernel.org/dri-devel/20230719081544.741051-1-javierm@redhat.com/
07:55mkurz_ has joined #asahi
08:35farghal has joined #asahi
08:36ciiur^ has quit [Ping timeout: 480 seconds]
08:36ciiur^ has joined #asahi
08:37j`eyamw: I came up with the same patch while investigating this before :)
08:40Armlin has quit []
08:40kunev has joined #asahi
08:43farghal has quit [Ping timeout: 480 seconds]
08:44qyliss has quit [Ping timeout: 480 seconds]
08:48le0n_ is now known as le0n
08:55capta1nt0ad has quit [Ping timeout: 480 seconds]
09:15PaulFertsermarcan: another user reports that all-upstream code supports WPA3 Personal on raspberrypi 4: https://github.com/raspberrypi/linux/issues/4718#issuecomment-1686971085 . I can also confirm that it works with same SDIO module on a firefly rk3288 board.
09:17PaulFertserRPi users mostly complained about lack of WPA3 because they're using some odd distros (like armbian) which ship broadcom firmware not from the linux-firmware repository, and it just lacks sae support altogether. Also, it wasn't obvious at all that only iwd supported that part of the API.
09:17marcanmaybe Cypress or someone from rPi should actually reply to my emails then
09:18marcanI already send my tree to the rPi folks and asked whether it breaks anything and haven't gotten any reports of breakage yet
09:18marcanmaybe the Cypress firmware *also* supports the WSEC protocol that Broadcom uses
09:18PaulFertserSomeone from rpi replied on that github ticket asking for more details, I provided everything I knew.
09:19marcanunless someone tests *my* patch and confirms it *actively* breaks Cypress devices with upstream kernel, there's nothing I need to do here
09:19holiday has joined #asahi
09:19PaulFertserMy point is that the old codepath also works but that doesn't matter if the new one is no worse.
09:19marcanif it also works with Cypress then it is correct
09:19marcanif it doesn't then someone better engage in the ML, because so far nobody involved with Cypress chips actively has
09:19PaulFertserYeah, right, you shouldn't care more than the rpi foundation themselves.
09:20PaulFertserI'm just saying the old codepath actually works.
09:20marcanI mean I figured it would work with *some* firmware *somewhere*, otherwise it wouldn't exist :p
09:20marcanbut it sure doesn't seem to work for everyone in the lovely mess that is rPi distros...
09:21PaulFertserBecause they're shipping odd firmware versions instead of sticking to whatever is in the Linux git firmware tree.
09:21PaulFertserThey just lack the -sae- function.
09:21capta1nt0ad has joined #asahi
09:22marcanit would be helpful to, you know, get direct word from Cypress as to what happened here and how they ended up with that weird prop instead of the upstream Broadcom implementation
09:22marcanbut so far, crickets
09:22marcanthey've ignored all my emails
09:22marcanand it doesn't seem anyone cares, so it's probably a good idea to *actively* break some people and see if anyone notices
09:22PaulFertserWe all know that.
09:22marcanotherwise we're just doing their job for them and walking on eggshells
09:23PaulFertserYour exchange on the mailing list was pretty convincing and apparently upstream wireless subsystem maintainers agreed it's the way forward.
09:24PaulFertserBTW, is firmware you extract from macOS compiled with -sae- , what's result of command similar to strings /lib/firmware/brcm/brcmfmac43455-sdio.bin | tail -n 2 ?
09:25marcan<FW-TAG>4387c2-roml/config_pcie_perf_udm_nosar Version=20.96.31.0.8.7.148 Date=2023-05-27T02:22:01Z FWID=01-1f7d4484
09:25marcan<UCODE-TAG>path=.../dot11_firmware/branches/OlympicAXUcodeTwig_1478_26400@44762 major=1478 minor=26400 diff_ct=0 prebuilt=0 build=/var/xbs/bwa/F6C191EE-9DD2-479E-A1E3-445222C22A78/BCMFirmware_4387_all-343/srcroot/build/dongle/4387c2-roml/config_pcie_perf_udm_nosar/ucode_build
09:26marcanthere are hits for 'sae' elsewhere
09:27PaulFertserThis lacks the feature list though, something like 43455c0-roml/43455_sdio-pno-aoe-pktfilter-pktctx-wfds-mfp-dfsradar-wowlpf-idsup-idauth-noclminc-clm_min-obss-obssdump-swdiv-gtkoe-roamprof-txbf-ve-sae-dpp-sr-okc-bpd Version: 7.45.221 ...
09:27marcanyeah I think they don't do that for apple
09:28marcanbut there are strings that look like a feature list: https://mrcn.st/p/NPsQ7A0l
09:29PaulFertserThe API to get the feature list should still work, brcmf_feat_firmware_capabilities() would print it on probe if debug is enabled.
09:30marcanyes I'm pretty sure it would show up
09:31marcantoo lazy to reboot with debug right now :p
09:31PaulFertserApple asked them to be generous with the features, that's plenty of them :)
09:31marcanincluding the custom apple ones (awdl)
09:33PaulFertserBTW, with this offload SAE works but the corresponding AKM is never reported in "Supported Ciphers" section of "iw list".
09:34PaulFertserI also do not know if anyone tried it with Fast Transition enabled (it's yet another AKM).
09:37dgb has quit [Ping timeout: 480 seconds]
09:39dgb has joined #asahi
09:39marcanthere's so much broken in that driver...
09:42PaulFertserThe firmware also indicates support for OWE, that's nice.
09:44svenfwiw, the Apple Bluetooth firmware lies in the supported capabilities hci reply because xnu just completely ignores that anyway
09:44PaulFertserSo /that/ was the reason you needed that quirk. Heh.
09:45svenyup :(
09:45svenI later realized that it just claims to support everything
09:45svenso far we only hit two things it lied about but it would not surprise me if we end up with a few more
09:52holiday has quit [Ping timeout: 480 seconds]
09:54Dementor3 has joined #asahi
10:00Dementor has quit [Ping timeout: 480 seconds]
10:01Dementor has joined #asahi
10:06Dementor3 has quit [Ping timeout: 480 seconds]
10:25dgb1 has joined #asahi
10:28zumi has quit [Remote host closed the connection]
10:29zumi has joined #asahi
10:31dgb has quit [Ping timeout: 480 seconds]
10:31dgb1 is now known as dgb
10:31capta1nt0ad has quit [Ping timeout: 480 seconds]
10:36EmantorCypress/Infineon seem to have no intrest whatsoever in upstreaming support for their wifi/bt chipsets. I have a Laird LWB5+ here where wifi using USB requires a clm blob to work and the downstream BT driver does some additional pokes to get BT to work…
10:40EmantorI guess cypress/infineon don't want to engage with mainline since they provide a 211 patch series for their chipsets in their driver releases: https://community.infineon.com/t5/Wi-Fi-Bluetooth-for-Linux/Cypress-Linux-WiFi-Driver-Release-FMAC-2023-09-01/td-p/492862
10:43EmantorThe mentality may still be a "we patch stuff anyway, lets just maintain our adjustments downstream"
10:46wille-io has quit [Quit: The Lounge - https://thelounge.chat]
10:46wille-io has joined #asahi
10:52wille-io has quit [Quit: bye!]
10:53wille-io has joined #asahi
10:53capta1nt0ad has joined #asahi
11:12capta1nt0ad has quit [Ping timeout: 480 seconds]
11:51rkr11 has joined #asahi
11:54rkr11hey, is it possible to access network cards from grub? Trying to make setup to network boot linux from grub over http, but net_ls_cards returns empty in grub. Loading modules `efinet` and `net` does not help, so I was wondering if its not supported by current drivers/modules and is known
11:54dgb has quit [Remote host closed the connection]
11:54dgb has joined #asahi
11:54Zeroine has quit []
11:55Zeroine has joined #asahi
12:00jannegrkr11: usb ethernet adapters (with u-boot drivers) supposedly work. Not sure if u-boot has drivers for the brooadcom (1GB) or aquantia (10GB) chipsets. if it has those need firmware
12:01jannegwlan is powered off at boot (could be changed) but I don't think u-boot has drivers or wlan support
12:02rkr11 has quit [Remote host closed the connection]
12:03zumi_ has joined #asahi
12:10zumi has quit [Ping timeout: 480 seconds]
12:13zumi_ is now known as zumi
12:16mkurz_ has quit [Ping timeout: 480 seconds]
12:21levixxxx has joined #asahi
12:22john-cabaj has quit [Remote host closed the connection]
12:22john-cabaj has joined #asahi
12:23levixxxx has quit []
12:28levixxxx has joined #asahi
12:28levixxxx has quit []
12:41marcannor firmware
12:43john-cabaj has quit [Remote host closed the connection]
12:43john-cabaj has joined #asahi
12:44Moprius has joined #asahi
12:46mkurz_ has joined #asahi
12:54rkr1 has joined #asahi
12:54rkr1 has quit []
12:54rkr11 has joined #asahi
12:59kidplayer666 has joined #asahi
13:09rkr11janneg: ah, I see, thanks. So the only way to do that would be with usb dongle then if netboot from either grub or u-boot... Gonna try that. But for wlan itself, I see there are some broadcom drivers in u-boot, but not sure if its relevant https://github.com/AsahiLinux/u-boot/blob/46d6873a2428a9e037660312d756c0ca412b1bdf/drivers/net/phy/broadcom.c
13:10ciiur^ has quit [Ping timeout: 480 seconds]
13:10ciiur^ has joined #asahi
13:10rkr11marcan: is it possible to maybe custom-build u-boot with that firmware to make wlan work?
13:11marcanu-boot absolutely does not have a brcmfmac driver
13:11marcanforget about it
13:12marcanwe also aren't going to provision wifi firmware for u-boot like we do for aspeed
13:12marcanif you want to netboot the only practical options that could be supported are wired ethernet (internal or usb)
13:13marcanthere is no wlan stack in u-boot at all as far as I know
13:13marcanand dare I say there should never be
13:13arthere's wlan stack in edk2, but that's a bit cursed
13:14marcanyes, I was about to say, down that path lies edk2
13:14marcanwe are using u-boot and not edk2 for a reason
13:14Szadek636 has joined #asahi
13:14marcan(mostly for our own sanity)
13:14rkr11marcan: yep, I wanna use wired ethernet (internal)
13:14marcanoh but you said wlan?
13:15marcanwlan = wireless lan
13:15rkr11yep, totally my bad, sorry! need lan
13:16marcanwhat device?
13:16rkr11m1 mac mini, my problem is wired ethernet is not detected currently in grub, no cards are visible basically
13:17PaulFertserSome other practical way might be to boot Linux+initramfs as first payload, and kexec from that.
13:17rkr11have yet to try if it will detect usb ethernet dongle tomorrow
13:17marcanPaulFertser: no, kexec is not supported and never will be
13:17marcanthere is no way to reinitialize firmware
13:17marcanthis is a platform limitation
13:18Moprius has quit [Quit: bye]
13:18PaulFertserOK, that makes this platform special.
13:18marcanit took you this long to figure that out? :p
13:20armarcan: "reinitialize firmware", as in the things that happen before m1n1, or in/after?
13:20marcanrkr11: as far as I can tell u-boot does not currently have drivers for any of the ethernet chipsets apple uses, no
13:20marcanar: before m1n1
13:20marcanthere is no way to fully reset any of the coprocessors to its boot state, because its boot state is loaded by iBoot
13:20PaulFertserNot able to kexec feels considerably more special than I could expect, yes :)
13:21Szadek63 has quit [Ping timeout: 480 seconds]
13:22armarcan: and adding a code path (triggered by a cmdline arg?) "expect these things to be already configured" would be too fragile, right?
13:22marcanI mean you might be able to once we have PSCI, just forget about GPU, video encoder, touchpad/keyboard (on M2+), and a few other things if the first kernel loaded any of those drivers
13:22marcanwe already do a hideous handoff hack for the keyboard in u-boot and I'm not particularly keen on also implementing that for kexec
13:22marcanGPU is a lost cause
13:23rkr11marcan: thank you, that does save me lots of time digging into it. Gonna test the dongle way for now and see if it works for my use case. Thanks again!
13:23marcanar: "being configured" = sharing memory with the kernel
13:23marcanhanding that off is insanity
13:23aryeah, doesn't sound like something that could be handled with kexec in a reasonable way
13:26marcanthis is why apple reboots after the boot picker, they *can't* do it any other way
13:26marcanyou could implement kexec by setting some environment/file flags and putting the new kernel/initramfs somewhere in the ESP and rebooting with a flag that tells u-boot to load that instead
13:27Axenntio has joined #asahi
13:27marcanapple style ;)
13:27aranother, more realistic probably, approach would be to load a generic kernel+initrd, but instead of kexec-ing, it would pivot_root + init reexec (which is a common thing initrds do already). but that limits you to only netbooting different linux distros
13:27marcanand then you have a problem with modules
13:27arright
13:27marcan(and kernel configs etc, like fedora uses selinux on by default)
13:28ChaosPrincessisnt kexec on it's way out on consumer hw anyway?
13:28ChaosPrincessthe whole thing is incompatible with secure boot, and x86 oems are pushing that
13:28marcandoes it even work properly for full consumer hw use cases anyway?
13:28marcanlike I have an AMD machine with a Radeon and it absolutely doesn't know how to reset that upstream
13:29marcanI have to load an out of tree module to get vfio passthrough to work more than once
13:29marcanno way kexec works with that
13:29ChaosPrincessi have a nv card in my x86 box, can't kexec that sanely either
13:29marcankexec always felt like a hack that is only useful for very restricted bootloader-type use cases and for kernel panic dumping purposes, not as a general-purpose thing
13:29ChaosPrincessyou can easily kexec vms :P
13:30marcanoh sure :p
13:30marcanbut that's also useless
13:30marcansince half the reason to kexec is speed
13:30marcanand VMs boot fast :p
13:30ari don't remember ever getting kexec working on anything but VMs and *some* server hardware
13:30aron x86
13:31ChaosPrincessi actually have a reason-ish to kexec vms, but i only need this due to some really stupid workarounds for a certain hypervisor
13:31PaulFertserRHEL has kexec-based dumping on panics enabled by default.
13:31arChaosPrincess: whose name starts with "V" and ends with "mware"? ;)
13:31ChaosPrincessmight be that :P
13:33arone usecase I have for kexec, is replacing distros in vm instances with things that a given vm provider doesn't officially support
13:33arbut that's also flaky
13:33marcanPaulFertser: that one is plausible because the NVMe controller is one of two things that are designed to be handed off, though it will probably still need some horrible code to make it work when the first kernel *panicked*.
13:33marcan(the other one being DCP)
13:34marcan(well okay and SMC I guess)
13:34ChaosPrincessmake pstore over nvram, dump panic data into pstore :P
13:34marcanbut as a strict panic dumper kexec thing only, with a special kernel with all the other drivers disabled
13:34marcanChaosPrincess: I think macos actually dumps it into one of the extra nvme namespaces
13:36ChaosPrincessit is a neat trick when you control the platform
13:37qyliss has joined #asahi
13:37Axenntio has quit [Quit: Axenntio]
13:38Axenntio has joined #asahi
13:38Axenntio has quit []
13:40rkr11 has quit [Remote host closed the connection]
13:41sveni'm still not sure how they do that dump since iirc that namespace is read-only with our driver. probably needs some special commands
13:45mikee3000 has quit [Quit: The Lounge - https://thelounge.chat]
13:46mikee3000 has joined #asahi
13:49sdomi has quit [Quit: meow, Quitting.]
13:50sdomi has joined #asahi
13:55ciiur^ has quit [Ping timeout: 480 seconds]
13:55ciiur^ has joined #asahi
13:57sefidel has quit [Remote host closed the connection]
14:00sefidel has joined #asahi
14:09holiday has joined #asahi
14:09nicolas17 has joined #asahi
14:09SalimTerryLi has quit [Read error: No route to host]
14:10SalimTerryLi has joined #asahi
14:10SalimTerryLi has quit [Remote host closed the connection]
14:12SalimTerryLi has joined #asahi
14:19holiday has quit [Quit: WeeChat 4.1.1]
14:19holiday has joined #asahi
14:34holiday has quit [Quit: WeeChat 4.1.1]
14:35john-cabaj has quit [Remote host closed the connection]
14:36rvalue has quit [Read error: Connection reset by peer]
14:37rvalue has joined #asahi
14:46john-cabaj has joined #asahi
15:03joshheyse has joined #asahi
15:04john-cabaj has quit [Ping timeout: 480 seconds]
15:04delsol has joined #asahi
15:08lucyllewy has joined #asahi
15:43Chai-T-Rex has quit [Quit: Chai-T-Rex]
15:59nicolas17looks like all all M3 Macs ship with macOS 13.5 22G2074, and all M3 Pro/Max models ship with 14.1 23B2073
16:07nicolas17if anyone gets a machine I'd like a filesystem dump >.>
16:08nicolas17and let me know if you get (or hear of) anything different, such as any M3 coming with 14.1 (which build?), or any Mac at all coming with 13.5 22G2080 (it exists but we don't know where)
16:11ciiur^ has quit [Remote host closed the connection]
16:26Rgn has joined #asahi
16:34Rgn has quit [Quit: Page closed]
16:42mkurz_ has quit [Ping timeout: 480 seconds]
16:45fossdd_ is now known as fossdd
16:47kidplayer666That sounds weird
16:47kidplayer666Kinda random as they got launched at the same time?
16:48kidplayer666Maybe they started manufacturing the more popular normal models before the pro ones?
16:48delsolkidplayer666: seems reasonable.
16:49kidplayer666Cause if I’m not mistaken this process should’ve started a while ago
16:49delsolespecially since rumors were suggesting possibly only M3 released, but discuss/introduce M3Pro, M3Max... shipping at later date (likely after holidays)
16:49kidplayer666So all should have old software
16:49kidplayer666Maybe that’s it
16:50kidplayer666Huh last minute change?
16:50kidplayer666Maybe they cancelled 13” at last minute?
16:50delsolkidplayer666: I think its possible that yields on M3Pro/M3Max were shit, and expected to take far longer to get in qty needed.....
16:50kidplayer666So it wouldn’t make sense to debut the m3 exclusively on the IMAC?
16:50delsolother questions is whats the QTY?
16:51delsolpeople bitching about pro being 6p6e instead of 8p4e... but the ramifications are far far bigger.
16:52delsolbefore EVERY single M1Pro/M2Pro that they shipped... was a crippled M1Max/M2Max that they cut off half of the GPU.
16:52kidplayer666The problem is more
16:53kidplayer666That they now kinda hide a bit more the reduction in p/ e cores
16:53delsolif you need to sell 100k M2, 20k M2Pro and 15k M2Max.....
16:53delsolthats more than 50% of the M2Max dies you are cutting 40% off of...
16:53delsoland the people bitching about memory bandwidth are fools.
16:54delsolM2Max may have 400GB/s of theoretical memory bandwidth... but real world from memory to cores.... is only about 225GB/s
16:54kidplayer666Most certainly
16:54kidplayer666Unless you’re a 4060 ti owner :P
16:54delsolSapphire Rapids with its 8 channel DDR5 controller... and socket TDP of 380w.... only gets real world 250GB/s from memory to cores.
16:54kidplayer666In which case I understand the trauma
16:55delsoland whats worse, is to get the 250GB/s from memory to cores on sapphire rapids... you need 32 or more cores.
16:55nicolas17kidplayer666: yeah it seems like M3 was intended to launch earlier and for whatever reason didn't?
16:55kidplayer666But it doesn’t include a GPU
16:55kidplayer666Which is way more bandwidth dependent
16:55delsolkidplayer666: yes and no.
16:55kidplayer666nicholas17: yeah, kinda odd
16:56kidplayer666The simple answer is waiting for actual reviews
16:56kidplayer666Instead of whatever crap Apple might be spweing
16:56delsolkidplayer666: thats the other gotcha.... the previous setup GPU wasn't able to use all the memory bandwidth either.
16:56kidplayer666I remember on the M2 pro or something release
16:56kidplayer666How Apple claimed 6x better performance
16:57kidplayer666On rendering
16:57ChaosPrincessdelsol: the bitching about pro not having 8+4 is that people who mostly care about cpu perf above all else, and the gpu is "meh whatever" now have to buy the max
16:57kidplayer666(Cause they had a project that took advantage of those extra few gigs of ram)
16:57delsolChaosPrincess: other than how many workloads are going to be actually more performant on 8p4e vs 6p6e?
16:58kidplayer666delsol: if the theoretical limit is lower, the proper limit is probably lower too
16:58delsoland how much die space do you want to spend on it....
16:58ChaosPrincessdelsol: all of my workloads, and throw all the die at it, tia
16:58delsolChaosPrincess: then you don't want an M3Pro or M3Max...
16:59delsolyou want an M3-ShortIntelStockToday.... 4 M3-Max dies bolted to an IOdie with another couple memory channels for those huge memory footprint workloads, along with a pile of PCIe lanes....
16:59ChaosPrincesswell, yes, i want a threadripper, but guess how that line was going the previous years
17:00delsolin all seriousness, most apple workloads wouldn't benefit much from an -Extreme 4x die setup.....
17:00delsolbut if linux was better supported on it, plenty of big linux server workloads would run fucking amazeballs on it, and justify the 10k/15k/20k pricetag such a machine would demand.
17:00ChaosPrincess"Apple workloads" sounds like cope from the "macs are for video only" crowd :P
17:01delsolSee also Ampere Altra benchmarks....
17:01ChaosPrincessShit singlecore, ancient arm cores
17:01kidplayer666If Apple wants to be taken seriously then it does have to cater to those power users
17:01delsolsingle threaded blows chunks, but if you are doing a metric fuckload of separate VM's that don't need single threaded..... throughput is huge for the price.
17:01kidplayer666I’m kinda curious about windows on arm
17:02kidplayer666Cause Qualcomm deal is expiring soon
17:02delsolkidplayer666: Amdahl's law basically says that those workloads are few and far between.
17:02kidplayer666And mediatek has been closer and closer to Qualcomm in the mobile space
17:02kidplayer666So I expect on a laptop chip that is slightly less power constrained a bunch of competition
17:02kidplayer666Especially cause NVIDIA might be thinking about it too
17:03delsoland at the clockspeeds apple is now pushing on M3 cores... if the cores are as powerful as expected, plenty of "big xeon" workloads would likely be faster on an M3-Max than the xeon.
17:04delsolbecause the big xeons don't clock for crap, chug power, and even four times as many cores doesn't make a difference if 10-15% of your workload execution time is single threaded and your single threaded performance is 20% lower.....
17:05delsolSee also: no consumer 5000 series threadripper..... because at the pricepoint the Threadripper Pro was plenty beefy to handle most of the single-workload setups as fast if not faster than a Dual EPYC....
17:07delsol(memory bandwidth and cache is also a large consumer of power.... and for a laptop chip, it makes a reasonable difference in potential battery life)
17:07billak has joined #asahi
17:07areven the AM4 5950x was quite popular as a server chip, despite having "only" 16 cores
17:07arand the 3950X before that too
17:07ChaiTRex has joined #asahi
17:08delsolYes, on non-bandwidth limited workloads.
17:08kidplayer666Now there is 7000 series tho for threadripper
17:08delsolsingle threaded performance increases of higher clocks made up the difference....
17:08kidplayer666How long it’ll last, it’s anyone’s guess
17:08delsolBut look at the 7000 series....
17:08delsoltake a goood look at the clock speeds. ;)
17:09kidplayer666It really feels like
17:09kidplayer666We finally got out of the 2010s boredom
17:09kidplayer666Intel is getting kicked in the nuts, but slowly trying to respond
17:09kidplayer666Apple made their own chips, for crying out loud
17:09kidplayer666Windows has launched a new version
17:10kidplayer666Despite promising 10 was the last one
17:10kidplayer666Windows 12 rumours exist!
17:10delsolkidplayer666: you see people that have run the dev builds of windows for ARM on M1/M2?
17:11kidplayer666I heard it kicks qualcomms chips in the nuts
17:11kidplayer666Beats their best chips by a fair margin on a gm
17:11ydalton has joined #asahi
17:12ydaltonkidplayer666: yes, the m1/m2 are the most powerful consumer level arm chips available
17:12kidplayer666Cause m3 is coming soon lol
17:12delsolnon-GPU stuff appears to blow the pants off of windows on x86 laptops....
17:12delsolbut look at what it was doing.....
17:13kidplayer666But Qualcomm x elite may be interesting
17:13delsollook at benchmarks of random tasks..... M2 at 3.4ghz was competing with intel P cores at 5.5ghz.
17:13kidplayer666And there’s rumours mediatek may partner with NVIDIA for graphics
17:14delsolkidplayer666: I don't think so. I think its the flipside.... NVIDIA is going to just build their own arm shit with the own GPU.
17:14delsol(and potentially license a bunch of the rest of the SoC stuff from mediatek)
17:14kidplayer666That’s also possible
17:14ydalton"non-GPU"? i'm pretty sure agx is more powerful than integrated graphics on intel
17:14kidplayer666They probably don’t need to licence anything
17:14kidplayer666ydalton: Not on a vm
17:15delsolit already is, see also Nvidia Grace/Grace-Hopper
17:15kidplayer666Just a matter of remembering first surface was NVIDIA powered
17:15kidplayer666And so is the switch
17:15ydaltonkidplayer666: what do you mean not on a vm
17:15kidplayer666ydalton: not sure how well the GPU runs through the vm
17:15ydaltoni thought you meant in general
17:15delsolydalton: windows GPU drivers to run as a VM on M2-Ultra even.... not so great.
17:15delsolbecause you're emulating a GPU with CPU....
17:16delsolbasically an emulated Matrox G200e
17:16delsolhey, don't fuck with the G200e, G200e is a great GPU for all sorts of shit.
17:16delsolseveral of my hackintosh's had G200e's
17:16kidplayer666You could always brute force your way by having many cpu cores
17:17ydaltonhow does the hypervisor framework work though? cuz i know you can gpu accelerate with that
17:17delsolnot really, that would take EPIC/VLIW style architecture.
17:17delsolydalton: still need drivers.
17:17kidplayer666ydalton: your first project :P
17:18ydaltonam not interested in windows
17:18delsolyou can take a Geforce 4090 back to 1997, plug it into a PCI to PCIe bridge chip and put it on your overclocked dual socket celeron machine... and it still aint gonna do shit.
17:18kidplayer666Sounds like a fun project
17:19delsolkidplayer666: no, a fun project was an ALR Revolution 6x6... with 6 pentium pro's (3 socket 8's per CPU daughterboard...)
17:19ydaltonsomething for ltt
17:19delsolwith 6 socket 370 to socket 8 adapters.....
17:19delsoland six P2's, P3's or celerons.....
17:20ydaltoni'm too young to remember any of those except the celeron
17:20* delsolactually worked on one with a buddy in college. We only ever got 4 CPU's operational with the adapters....
17:22kidplayer666I’m too young to remember any of that
17:23ydaltoni heard rumours the m2 was supposed to be what m3 is now
17:23delsolso what was your era of experimenting with shit to see what you could accomplish?
17:24delsolI'm basically backwards in this channel..... did most of my stuff on mac, later OSX.... and did a shitload of hackintoshing for a while....
17:24ydaltondunno, the closest thing was getting gpu passthrough working on my desktop
17:24delsolin ESXi or Proxmox?
17:24ydaltonjust libvirt/kvm
17:25ydaltonkinda got me into the linux space after seeing mutahar demonstrating it
17:25delsolI had plans of using Proxmox to finally solve a goal that I've been trying to solve for like 20 years....
17:26delsolMulti-headed Multi-seat......... one computer to rule them all, with dual monitors in this room, that room, and the other room..... 3 sets of keyboard and mice... and be able to log in and out, and take home directory/environment/etc with you from one room to another.
17:27ydaltoncan't you do that in linux anyways?
17:27ydaltonwell you'
17:27delsolnot really.....
17:27ydaltonyou'd need quite long cables to connect from one room to the other
17:27delsolyou can have 1 machine act as 3..... with 3 separate VM's....
17:27ydaltonmulti seat setups
17:27ydaltonmultiple monitors and sets of keyboard and mice with linux
17:28delsolYes.... but they don't work right.
17:28delsolbut you can't lot out of kitchen, log into bedroom and get your environment back.
17:28ydaltonwhat do you mean your environment back?
17:28delsolnot fully log out rather, but kick to screensaver login......
17:28delsoland then re-connect to running session on other seat.
17:29delsolwatching video side by side IRC. pause, go to loginwindow...... move to different room that is on loginwindow, log back in, have IRC windows still connected and video ready to hit play again.
17:32delsolAbout the closest thing that truly exists as a real solution was Sun Microsystems "hoteling" system..... where every desktop/cubicle had a machine, a phone, and a smartcard reader.
17:33delsolInsert your smartcard, and it logs you in as you, with your (over network home environment), forwards YOUR phone extension to the phone on the desk, and marks your position on their own internal user-tracking map thingy.
17:33delsolbut that still required you to save and close all of your applications.....
17:33delsolto re-open them on a different instance later.
17:35delsol(Citrix also almost does it.)
17:36ydalton has left #asahi [ERC 5.5.0.29.1 (IRC client for GNU Emacs 29.0.92)]
17:36delsolthe way to do it with Proxmox that I was planning on doing was to basically use one instance to trigger webserver to pause my VM, unconfigure its passthrough, and then on login, trigger webserver to reconfigure device passthrough for new seat (new USB bus, new GPU, new displays) and resume the VM.
17:37delsolbut there are some GPU considerations and gotchas there.....
17:37dgb has quit [Remote host closed the connection]
17:37dgb has joined #asahi
17:39delsol(Namely GPU state and GPU memory are lost on switch..... and while Nvidia has some special tools to deal with that with Grid enabled GPU's.... AMD's GPU's don't have the whole toolset you truly need to make it seamless.... and grid enabled GPU's weren't in my budget or supported in all the OS's I wanted....
18:05zumi_ has joined #asahi
18:09zumi has quit [Ping timeout: 480 seconds]
18:10zumi_ is now known as zumi
18:26delsol has quit [Remote host closed the connection]
18:27kidplayer666what were "all the OS'S you wanted?
18:28billak has quit [Quit: Konversation terminated!]
18:34chadmed has quit [Ping timeout: 480 seconds]
18:40nicolas17https://twitter.com/jonathan_gblt/status/1721930212785475727 looks like erasing an M3 Mac and trying to reinstall it from recovery doesn't work x_x
18:40kidplayer666jesus christ
18:40kidplayer666what the hell
18:41kidplayer666oh wait, its now x.com
18:42nicolas17I thought they'd fix this M3 issue together with releasing 14.2 b2 an hour ago, but neither of those happened
18:42Axenntio has joined #asahi
18:51kidplayer666also, quick question, why IRs
18:51kidplayer666IRC and not a protocol like matrix or xmpp
18:54Axenntio has quit [Quit: Axenntio]
18:54dgb1 has joined #asahi
18:55j`eykidplayer666: there is a fedora-asahi matrix
18:56kidplayer666im guessing this is the main thing tho
18:56PaulFertserAnd xmpp is kind of unconventional for multi user chats, and requires registrations. While IRC works right away for everybody.
18:56dgb has quit [Ping timeout: 480 seconds]
18:56dgb1 is now known as dgb
18:57kidplayer666really just cause most clients for irc feel kind of dated
18:59chadmed has joined #asahi
19:00___nick___ has joined #asahi
19:01kidplayer666having to check logs for past messages
19:07j`eykidplayer666: lots of us just stay connected
19:08kidplayer666Would mean configuring maybe a computer to stay on all the time
19:08kidplayer666Getting a server
19:08kidplayer666Given that my main pc consumes 200w on idle
19:08kidplayer666Not an option :p
19:09j`eythere's stuff like irccloud
19:11___nick___ has quit []
19:15___nick___ has joined #asahi
19:15Axenntio has joined #asahi
19:15Axenntio has quit []
19:15___nick___ has quit []
19:17ydalton has joined #asahi
19:17ydaltonnot to mention i can chat directly from emacs!
19:18___nick___ has joined #asahi
19:19nicolas17<ydalton> how does the hypervisor framework work though? cuz i know you can gpu accelerate with that
19:19nicolas17ydalton: Metal paravirtualizized graphics
19:21nicolas17so afaik getting GPU acceleration on a Linux-on-macOS VM, or macOS-on-Linux VM, would both need pretty complex Metal-specific stuff...
19:22ydaltonnicolas17: not remotely similar to opengl emulation with qemu?
19:22nicolas17maybe similar... but Metal instead of OpenGL, and undocumented :P
19:25ydaltoni was hoping something similar to GVT-d where it creates a virtual GPU and you can pass that to a vm instead
19:29svenall we know is that it's para-virtualized and not any kind of direct HW passthrough
19:31delsol has joined #asahi
19:32cozydeer has joined #asahi
19:40ydalton has left #asahi [ERC 5.5.0.29.1 (IRC client for GNU Emacs 29.0.92)]
19:41cozydeer has quit [Remote host closed the connection]
19:56atipls_ has quit [Ping timeout: 480 seconds]
20:29jashton has joined #asahi
20:33kidplayer666j`ey: that’s what I’m using lol
20:35dbm5 has joined #asahi
20:36dbm5refreshing to find a group using IRC instead of discord
20:56kidplayer666We gotta do discord but OpenSource and not sucking
20:56kidplayer666Just fix the entire matrix protocol :P
21:00nicolas17lmao it's a legal holiday in the US?
21:02ChaosPrincessthere are legal holidays over there?
21:02atipls has joined #asahi
21:03nicolas17ok not a holiday
21:03nicolas17"The 2023 United States elections are scheduled to be held, in large part, on Tuesday, November 7, 2023. The off-year election includes gubernatorial and state legislative elections in a few states, as well as numerous citizen initiatives, mayoral races, and a variety of other local offices on the ballot."
21:05ChaosPrincessyep, that is happening, and a certain state has a "do we backslide further into authoritarianism" on the ballot
21:05nicolas17why the heck did Apple release new hardware today and then not leave anyone to fix the trash fire in the update servers
21:05ydalton has joined #asahi
21:05ydaltonwell, it's not much of a holiday if ya still gotta go out and vote :P
21:07___nick___ has quit [Ping timeout: 480 seconds]
21:45dbm5 has quit [Quit: Textual IRC Client: www.textualapp.com]
21:46dbm5 has joined #asahi
21:48dbm5 has quit []
21:54opticronThe USA does not get holidays on voting days, just FYI. You generally have to take off work to do it or go outside of work hours.
21:57ydaltonit's not mandatory to vote there is it?
21:57jni guess it would be too democratic to simply vote on sundays
22:03c10l has joined #asahi
22:16kidplayer666here in portugal its always on a sunday
22:20opticronydalton, no, it's not mandatory
22:21nicolas17ok Apple released 13.6.2 and 14.1.1 and fixed the mess
22:34ydaltonnicolas17: the ProMotion bug and recoveryOS not being updated bug?
22:35nicolas17ydalton: no, the "M3 Macs ship with 13.5 and can't be updated" bug
22:35nicolas17no idea if they did anything with ProMotion
22:36nicolas17ydalton: actually I think they may have fixed ProMotion
22:36nicolas17because 13.6.2 is *only* for "MacBook Pro (2021 and later) and iMac (2023)"
22:51ydaltonwould marcan receive official confirmation from apple?
22:51ydalton has left #asahi [ERC 5.5.0.29.1 (IRC client for GNU Emacs 29.0.92)]
23:23cy8aerSilly in the early days of linux I waited for a month when a new kernel was published before I installed it. Today you need to wait for something like new macOS versions for a month before install it.
23:24nicolas17ydalton: confirmed!
23:24nicolas17marcan: they fixed the ProMotion issue!
23:26mini_ has quit [Quit: ZNC closing...]
23:31jashton has quit [Ping timeout: 480 seconds]
23:36j`eyI wonder what they did
23:37delsol_ has joined #asahi
23:37nicolas17j`ey: where do you think the fix would be? kernel or bootloader?
23:39nicolas17someone might be able to get me iBoot decryption keys, if someone's interested in diffing 13.6.1 and 13.6.2 :P
23:41j`eynicolas17: not sure
23:41delsol has quit [Ping timeout: 480 seconds]
23:45delsol_ has quit [Remote host closed the connection]
23:54mini_ has joined #asahi
23:54mini_ has quit [Remote host closed the connection]
23:59mini_ has joined #asahi
…