In reply to @aciceri:nixos.dev
Your work there is great, do you keep images updated? Don't you host a cache also?

In reply to @aciceri:nixos.dev
Hey folks!
I'm working with some RISC-V derivations and I'm a bit annoyed by the building speed (I'm using binfmt on a quite powerful x86_64 machine). I know that there is this cache (indeed I'm using it, with the same exact nixpkgs used to populate it) but still many derivations are not cached and at the same time I would like a more updated nixpkgs.
Isn't there a community builder, right? Also I tried looking for cloud providers that offer RISC-V without luck, maybe there is one instead?
I was also considering buying an SBC but they don't look so powerful (I checked the VisionFive2 and the Star64), I fear that I would get performances similar to my emulated RISC-V.

In reply to @skeuchel:matrix.org
I have a VisionFive2 and a Lichee Pi 4A. The VF2 is slower than QEMU user mode emulation on recent x86 CPUs (AMD 5950X / 3990X). The Lichee Pi 4A is slightly faster than QEMU.

In reply to @thefossguy:matrix.org
Yeah, JH8100 should be as fast as RK3588, at least in raw computer. Which is what we need for compilation.      

Somewhat of a follow-up to that message:
JH8110 SoC: 2x dubhe-80 (low-perf, energy efficient cores) and 4x dubhe-90 (high-perf, said to be equivalent to ARM's A76 cores (what RK3588 has as high-perf cores))

Even though the CPU IP has the V 1.0 (vector) extension, the SoC doesn't. I don't know why. The H (hypervisor) extension is missing too, so forget h/w accelerated KVM on this.

It has 16 cores and up-to 96 GB of RAM with 16x lanes of PCIe Gen 4 goodness.

All 16 of them are SiFive's P670 cores. From SiFive's estimate, they're "equivalent" to ARM's A78 cores.

12 cores clock up-to 2.5 GHz as perf-cores and 4 cores clock up-to 1.6 GHz as high-efficiency cores.