<rwmjones>
which is not that interesting for us really (but appeals to researchers)
<davidlt>
So for researches it's probably about RV International, how it's royalty free, open source cores, ability to influence development in RVI, or have their own custom extensions for specific workloads?
<davidlt>
That's basically the main unique point compared to x86 and aarch64
<rwmjones>
yup, also I want to talk a bit about RISE
<davidlt>
Mentioning RISE would be good, but again not really a research thing (but everyone is mentioning RISE in their slides it seems).
<davidlt>
rwmjones, would you be interested in talking to bootflow folks at RH? ;)
<rwmjones>
who is doing that now?
<davidlt>
No idea
<rwmjones>
is that pjones
<davidlt>
What I need is for Fedora to start updating their GRUB2 and gnu-efi.
<davidlt>
The RH shim depends also on gnu-efi. Also thers is old PR to add riscv to the shim, which didn't go anywhere.
<rwmjones>
what's the PR link?
<davidlt>
GRUB2 is at RC, and gnu-efi 3.0.17 + riscv backports from master would be nice.
<davidlt>
No
<rwmjones>
I've never heard of gnu-efi until just now
<davidlt>
I am destroying our infra to attempt to do Pungi composes, and these things showing up :)
<davidlt>
High perf cluster is A76 level, and efficiency one is A75 level perf.
<davidlt>
If that's 8-core design, it would be epic.
esv has joined #fedora-riscv
<sorear>
was there news about jh8100?
jcajka has quit [Quit: Leaving]
somlo_ is now known as somlo
<davidlt>
sorear, not really. They talked again about their in house developed IP and said that JH8100 implements dual cluster, with high-perf and efficiency.
zsun has joined #fedora-riscv
zsun has quit [Ping timeout: 255 seconds]
zsun has joined #fedora-riscv
<conchuod>
1500+, lolno
<davidlt>
conchuod, it might cheaper than buying multiple SBCs to achieve the same perf.
<conchuod>
For your use case, yeah
<davidlt>
GCC is still cooking
<davidlt>
but it's like roller coaster
<davidlt>
Those link stage arrives and it's time for LTO it's just one process (one core) for even 1+ hour.
<conchuod>
For me as a punter that just wants to test things etc its far too expensive
jednorozec has quit [Ping timeout: 248 seconds]
<davidlt>
conchuod, oh yeah, for sure
<conchuod>
I'm really surprised though, I didn't expect it to be cheap.
<davidlt>
I kinda hope JH8100 happens sooner than later, and it requires a lot less upstreaming compared to JH7110
<conchuod>
I'm hoping that the starfive people learn a lot from the 7110 & apply it to that.
<davidlt>
I bet it will be 8-core design, with A76 and A75 perf that would be a killer too (I expect similar price point)
jednorozec has joined #fedora-riscv
<conchuod>
They seem to have so far, so I'm pretty hopeful :)
<davidlt>
I am not sure TH1520 will get similar amount of upstreaming effort
<davidlt>
and I love vendors that upstream their stuff
<conchuod>
thead seem completely uninterested in it
<davidlt>
we don't have Horse Creek, not sure if it's happening this year (based on recent Intel comments)
<conchuod>
they have their own reyvos distro & seem to just care about that
<davidlt>
The thing about T-HEAD is that Alibaba have tons of money. They can do whatever custom internally and support it.
<conchuod>
It prob saves me heartache arging about random bindings if it is done by community people.
<davidlt>
It probably works for them and that's fine.
<sorear>
i'm not sure "learn from the 7110" is useful framing
<davidlt>
It would be nice to see Ventana stuff. I doubt anyone will be able to afford it.
<conchuod>
what do you mean by that sorear?
<conchuod>
I dunno what "useful framing" means.
<sorear>
if it requires less upstreaming it'll be because the peripherals are shared with the 7110 and already upstream, not because the peripherals themselves are somehow magically better
<davidlt>
I also saw a news article about T-HEAD, StarFive, Nuclei and some others forming IP alliance in China regarding RISCV.
<conchuod>
Oh I didn't mean in terms of hardware.
<davidlt>
StarFive had their 1st time experience working upstream. Basically 101.
<conchuod>
I meant in terms of the process of upstreaming things to the various projects, they'll hopefully have an easier time of things for the 8100.
<davidlt>
Originally they didn't plan doing that themselves.
<sorear>
it would be nice if we had open standards for the uncore stuff so every new vendor doesn't have to upstream a bunch of clock drivers
<conchuod>
Maybe drew & jisheng will end up getting through a lot of the 1520, since it is mostly using dw peripherals.
<conchuod>
:copium: The RPMI spec will negate the need for 728 clock drivers.
<davidlt>
Yeah, TH1520 is mostly 3rd party IP, just a few bits have custom DT strings (probably for quirks?).