Finally did a test, got an improvement from 107.4 to 110.1 H/s.

I'm not sure why we have such a big difference in numbers. The only thing running on my PI is PiHole

You mistyped (missed M), need to type sudo ./xmrig --bench=1M --randomx-1gb-pages

I don't know why it's slower for you. My RV2 has only heatsinks but no fan like on yours, so your RV2 is definitely not overheating. But you had 101.8 h/s all the way back here: #3725 (comment)
Maybe OS reinstall will help you.

I'll explain what I've done. I've deleted the repository, redownloaded it, switched to this pr like so gh pr checkout 3746. Then just proceeded as normal, mkdir build → cd build/→ cmake .. -DCMAKE_BUILD_TYPE=Release -DARCH=native → make -j8 → sudo ./xmrig --bench=1M --randomx-1gb-pages .

Am I missing something?

I don't know why it's slower for you. My RV2 has only heatsinks but no fan like on yours, so your RV2 is definitely not overheating. But you had 101.8 h/s all the way back here: #3725 (comment) Maybe OS reinstall will help you.

Everything looks correct. I think either something is using a lot of CPU in background, or you set power governor to "power saving" mode or something like this.

@Slayingripper I now have the same problem. Hashrate dropped to 85 H/s and I can't do anything to get it back to 110 H/s. --randomx-1gb-pages doesn't help, in fact hashrate stays the same as with the regular huge pages. So I suspect that 1 GB huge pages stopped working, and OS is allocating the regular huge pages in the background.

@Slayingripper I now have the same problem. Hashrate dropped to 85 H/s and I can't do anything to get it back to 110 H/s. --randomx-1gb-pages doesn't help, in fact hashrate stays the same as with the regular huge pages. So I suspect that 1 GB huge pages stopped working, and OS is allocating the regular huge pages in the background.

At least I now know I didn't do anything on purpose. I tried disabling everything (Docker, PiHole, etc) to see if it would do anything, but nothing happened. Maybe Huge pages is something we should look into. I don't think this is a RISC-V problem, more likely an OS issue.

@Slayingripper I don't know what changed, but now I get 118.3 h/s with the latest XMRig. It seems that it's random between reboots, and I think it depends on which parts of physical memory get allocated for the dataset: