<alj[m]>
Would the method of virtualization used in spectrum improve performance of applications trying to use hardware graphics rendering (like chrome)? Qubes gives me some headaches with that and since spectrum seems to be the silver bullet for a lot of my issues, I wondered if it might help in this regard too
<qyliss>
alj[m]: the answer, as usual, is possibly
<qyliss>
the technology Spectrum is being built on does support GPU sharing with applications in a way Qubes does not
<qyliss>
there will be a security tradeoff to enabling that, because GPUs do not care one bit about your security
<alj[m]>
That is enough for me. I love this project
jb55 has quit [Remote host closed the connection]
jb55 has joined #spectrum
tilpner_ has joined #spectrum
tilpner has quit [Ping timeout: 246 seconds]
tilpner_ is now known as tilpner
<qyliss>
update on virtio-vhost-user: does not cause kernel panic on another computer I tried
<qyliss>
so I'm guessing it's a processor difference
<puck>
hrmm, kvm vs non-kvm maybe??
ehmry has joined #spectrum
<qyliss>
puck: both using kvm
<qyliss>
I did seperately create the VM on each machine since it's an ubuntu thing I set up imperatively
<qyliss>
there may be differences between them
<qyliss>
so now I'm copying the panicking VM to the other computer to check that it actually is the difference between the hosts
<qyliss>
but that'll take a couple of hours
jb55 has quit [Ping timeout: 240 seconds]
jb55 has joined #spectrum
<qyliss>
puck: so I can only reproduce it on my laptop, with kvm
<puck>
huh.
<qyliss>
works fine on laptop with no kvm, works on server with kvm
<puck>
iinteresting
<puck>
do you have the corresponding drv somewhere?
<qyliss>
oh there's no drv here
<qyliss>
eeeeverything is impure
<qyliss>
puck: if you want to look at it I'm happy to put something together though
<puck>
is this the same code that you're building tho?
<qyliss>
is what the same code?
<puck>
or well, maybe the same binary even
<qyliss>
oh
<qyliss>
no I built the binary fresh
<qyliss>
everything else is the same. kernel, initrd, disk.
<puck>
mmm
<qyliss>
I'll copy the binary to double check but highly doubt that's the difference
<qyliss>
wow, non-kvm sure is slow :D
<qyliss>
puck: confirmed with a nix qemu build that the same binary reproduces the issue on my laptop but not uhura
<puck>
qyliss: pfff.
<puck>
like, now i'm wondering which host kernels are used here
<puck>
and if it's reproducible on another intel machine
<qyliss>
puck: got an intel machine? want to try?
<puck>
i have an x240 i could test
<qyliss>
puck: uhura: 5.4.12, x220: 5.4.46
<puck>
not sure when i can test, brain's currently a bit toast, so probably somewhen tomorrow
<puck>
hmm.
<puck>
i'm running 5.6.15
<qyliss>
puck: /nix/store/260pzlfdx96cj4qknz4wlpc7xb5dzywy-qemu-4.2.0, available on uhura
<qyliss>
should be reproduceable with basically any normal-ish VM, using the arguments in the email I posted a few days ago
<puck>
mhm
<qyliss>
If anybody else in here has an intel computer and would like to help test this, it would be really useful to have the data and I'd be happy to provide more detailed instructions and/or a Nix derivation you could just build and run
<qyliss>
I'm going to write another email to qemu-devel with these findings
<zgrep>
I've got an X270, but am completely oblivious with regards to actually doing anything (and would need instructions).
<qyliss>
I'll start putting an easy test case together
<zgrep>
(Or a nix derivation to build and run, that'd probably be easiest for me, at least.)