<rwmjones>
biab, the binutils build is still going
<rwmjones>
it's "stuck" (not actually stuck) on the LTO step
<rwmjones>
I wonder if disabling LTO would be a good idea for our builds
<rwmjones>
I doubt it gets us much benefit, and it tends to serialize the builds, and is very slow on these relatively low memory systems
<davidlt[m]>
LTO has a huge impact, but I tried to deviate as little as we can these days.
<rwmjones>
I was wondering if we could do a global setting, but you're right it's probably better to stick to what Fedora is doing
<rwmjones>
so it's *still* doing LTO stuff ...
<rwmjones>
I'll copy and paste the binutils time into here when it finishes, entertaining guests today so I'll not be around too much
<rwmjones>
& also glibc afterwards
<davidlt[m]>
Yeah, LTO is really expensive for us. There are issues/bugs with some packages, at some point I should report them.
<davidlt[m]>
If we disable LTO globally we also go blind here until it's enabled again (e.g. once we become a proper arch).
<davidlt[m]>
There are 3 things that typically cause problems: (1) Slow single-thread/core performance (visible with LTO); (2) Slow storage (low 4K random IOPS) and DNF are not friends; (3) memory capacity for debug stuff, running stuff, LTO, etc.
<davidlt[m]>
Also compression is affected by slow single-thread/core performance.
<davidlt[m]>
So that also goes to (1).
<davidlt[m]>
News: new GCC 13 build + annobin landed.
<davidlt[m]>
TH1520 final PCB will have 16GB option, and it should be 1.3-1.5X faster than JH7110, and some things might be way more faster with their OoO cores (TBD, I guess).
<davidlt[m]>
Storage we could fix with USB-A to NVMe adapter. That's the best available option.
<davidlt[m]>
Software support is another thing. VF2 will probably be good enough within a few months (hopefully), while TH1520 god a good start, but lack of vendor involvement is annoying as usual.