Conversation

i fell asleep sitting up for like an hour and then woke up and started compiling an optimized linux kernel and made a totally sausage sandwich of tuna and kewpie and rice vinegar on toastered soursdough. lets go 3 am

1
0
0

the latest nvidia driver (590.48.01) does not currently compile on the latest linux kernel (6.19) btw so im doin it over again

2
0
0

@qualia don't make me yeet an amd gpu at you

1
0
0

@vikxin I MEAN IF YOU GOT ONE LYIN' AROUND ,

i am so frequently redlining this 2080 super so hard you'd think i was trying to conjure an embarrassing agentic dr. sbaitso waifu

1
0
0

@qualia I don't have a spare one no, but I do have two in use at the moment. Thinking about going to RePC tomorrow to grab some Xbox controllers again. Sigh. It's that time of month.

1
0
0

yayy yippee 6.18 lets go be BOREing

*rebooting sounds, explosions, screaming, toilet flushing, crowd boos, CMOS beep, floppy seek noise*

1
0
0
Linux tailpulse 6.18.10-qualicachy #1 SMP PREEMPT_DYNAMIC Sat Feb 14 03:23:06 PST 2026 x86_64 GNU/Linux

done did it

1
0
0

@qualia how long does it take you to compile a kernel?? I've been doing it a bunch lately and from scratch is the better part of an hour. And I give it a bunch of cores and everything. Maybe I'm IO-bound?

2
0
0

@vikxin how are you going through xbox controllers

what on earth kind of hormonal cycle is this

1
0
0

@qualia while I did imply that as a joke the fact of the matter is that I keep needing weird and obscure variants of controllers. This time I need a wireless 360 controller and an adapter for it.

I can show you my collection of controllers I already have. Most of them still work. Most of them are also junk. Please clap

0
0
0

@vikxin @qualia how much have you trimmed it down?

it builds like every driver for every device in existence with distro racconfigs and that takes a huge chunk of that

1
0
1

@charlotte @qualia I'm using a very specific config that's somewhat but not hugely trimmed down. I know if I use a very minimal config it's much faster, but just building amdgpu alone adds a lot of compile time

0
0
1

@vikxin idk 15m? i didn't time it. i'm on a 14-core xeon (no hyperthreading), quad-channel ddr4. it's the CachyOS kernel so it's chock full of tweaky bullshit, it's also got a bunch of optimizations turned on (Thin LTO, -O3) which probably isn't helping build times. mayb i unset my goofy CFLAGS and try a vanilla 6.19 real quick

1
0
0

@qualia jegus. I have more threads than that, but it's only dual channel DDR4 and it's running in a VM with questionable IO performance

1
0
0

@vikxin oh yeag this is on modest nvme bare metal

1
0
0

@qualia oh yeah I'm not using NVMe here, I'm using ZFS backed by SATA SSDs. That's probably a big part of it

1
0
0

@vikxin oh fuck okay. yeah. fuckin' unity basically forced me to get on the nvme drive train to not lose my sanity, and it was like going from . well, not dialup to dsl. more like dsl to cable. during off-peak hours. yknow

it's like tangibly difficult to go back to SATA now, i thought it wouldn't be as significant of a difference as it is

2
0
0

@qualia yeah I'm only using SATA under duress. Read: my VM host doesn't have enough lanes for an NVMe zpool. The whole rootfs for this VM is encrypted via dm-crypt, and I don't really have anywhere faster I can put it. I've tried.

I'm not about to spend thousands to rebuild this on an Epyc box either ugh

0
0
0

@vikxin defconfig, -march=x86-64 -mtune=generic

Kernel: arch/x86/boot/bzImage is ready  (#1)
make -j14 1412.35s user 71.20s system 1032% cpu 2:23.68 total
1
0
0

@vikxin

small parallel random I/O

fio --name=random_rw --ioengine=libaio --rw=randrw --direct=1 --bs=4k --numjobs=14 --size=1G --runtime=30 --time_based --group_reporting --unlink=1

   iops        : min=41498, max=53467, avg=47941.68, stdev=181.72, samples=840
[...]
Run status group 0 (all jobs):
READ: bw=187MiB/s (196MB/s), 187MiB/s-187MiB/s (196MB/s-196MB/s), io=5609MiB (5881MB), run=30002-30002msec
WRITE: bw=187MiB/s (196MB/s), 187MiB/s-187MiB/s (196MB/s-196MB/s), io=5618MiB (5891MB), run=30002-30002msec

48k IOPS probably helps lol. i don't suppose you have enough free RAM you could build it in a ramdisk, do u

1
0
0

@qualia I do not know this fio program

Anyway the biggest problem is that I'm building via a PKGBUILD in a git tree. I'd have to make that whole subdirectory secretly mount a ramdisk without makepkg noticing and...I might be able to do that, but idk

Lemme check how much RAM that vm has

1
0
0

@qualia nah I'd need like 3x the amount of RAM I have haha. The build artifacts and everything add up to about 26 GB, and the VM only has 10

1
0
0

@qualia I have been so tempted to build a super overkill server for like 3 years but I really cannot afford it. It'd be $3k-$10k depending on how stupid I get and good lord can I not justify that

1
0
0

@vikxin honestly terrible time for it too with ram how it is

fio is an I/O benchmark tool, it's been around forever and does the thing pretty nicely

2
0
0

@qualia i could sell ram I'm not even using anymore that I bought in 2019 for a profit. That's fucked up

0
0
0

@vikxin yknow also if you've got a slot for it, sticking an NVMe drive on a PCIe riser and passing just the one fucker through raw to the VM might not be the worst idea for a scratchspace

1
0
0

@qualia the problem is lanes, not slots. Basically everything is in use, except maybe one NVMe slot on the back of the board? Idk. I also bought this board under duress when the chipset on my much better motherboard died on Christmas a few years ago

ughhhh

1
0
0

@qualia if you come over at some point I can open the thing up and show you just how much of a nightmare it is. I installed liquid cooling recently. It was the only way I could keep the 5950X from constantly overheating

0
0
0