search

ishitatsuyuki / acominer

An experimental ETH miner powered by Vulkan
Apache License 2.0
18 stars 6 forks source link

Low hashrate on Fedora 35/RX 6800 XT #7

Closed nettersrl closed 2 years ago

nettersrl commented 2 years ago

Hi, I compiled acominer from the source, without recompiling Mesa nor applying your hrtimer kernel patch, and, using AMDVLK, I get a low hashrate 32.21 MH/s 22.96ms 32.07 MH/s 22.97ms

compared to 61Mh/s I can get with teamredminer. I would like to leave OpenCL... do you have any idea about the reason? May I get a better performance with AMDGPU PRO or is there any flag/parameter I should set when I run acominer?

This is my launch command: ./acominer stratum://0x76cB31Fdb28Ddf887AF3cA4A0f7786D02C86032a.Provaio@asia1.ethermine.org:4444

Thank you in advance.

ishitatsuyuki commented 2 years ago

Do not use AMDVLK. Use the binaries provided at https://github.com/ishitatsuyuki/acominer/releases/tag/v0.4.0.

nettersrl commented 2 years ago

./acominer --subgroup-size 128 stratum://ADDR.workername@eu1.ethermine.org:4444 2021-12-23T15:16:45.426476Z INFO acominer::stratum: Pool set difficulty difficulty=4.30G 2021-12-23T15:16:46.159825Z INFO acominer: device=AMD Radeon RX 6800 XT type=DiscreteGpu 2021-12-23T15:16:46.163612Z INFO acominer: subgroup_size=128 2021-12-23T15:16:46.171956Z INFO acominer: Generating light cache (CPU)... 2021-12-23T15:16:50.412397Z INFO acominer: Light cache generation done. 2021-12-23T15:16:50.530106Z INFO acominer: Light cache transferred to GPU. 2021-12-23T15:16:50.531210Z INFO acominer: Building DAG... epoch=462 ███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████ 1208320/1208320 2021-12-23T15:16:51.864310Z INFO acominer: DAG built duration=1.333104401s 96.90 MH/s 7.59ms 96.77 MH/s 7.59ms 96.77 MH/s 7.59ms 96.78 MH/s 7.59ms 96.77 MH/s 7.58ms 96.78 MH/s 7.59ms 96.75 MH/s 7.60ms 96.58 MH/s 7.61ms 96.53 MH/s 7.61ms 96.48 MH/s 7.61ms 96.43 MH/s 7.61ms 96.43 MH/s 7.62ms 96.43 MH/s 7.61ms 96.44 MH/s 7.61ms

But just few shares are accepted. I'm still with AMDVLK. If I use a lower subgroup-size value I get much lower hashrate. Also, I used to see 61Mh/s with teamredminer on this card, is it feasible to see such a bump in terms of hashrate?

ishitatsuyuki commented 2 years ago

You're glitching out the calculation with subgroup-size=128. That's not a valid option and what you are seeing is just fake doubled hashrate.

As I said, the only driver supported right now is the Mesa fork. AMDVLK has its own issue with shader compilation, and only runs at roughly half the theoretical capacity. If you don't want to build it, grab the prebuilt version from releases.

nettersrl commented 2 years ago

You mean that running run.sh will load your Mesa fork vulkan driver? Which kind of performance uplift should I expect using your miner? Do I need AMDGPU PRO proprietary driver?

ishitatsuyuki commented 2 years ago

You mean that running run.sh will load your Mesa fork vulkan driver?

Yes, it loads the bundled Mesa fork build through an environment variable override.

Which kind of performance uplift should I expect using your miner?

On RDNA, it should be roughly on par with teamredminer.

Do I need AMDGPU PRO proprietary driver?

You need not. The proprietary AMD Vulkan driver, in fact, performs even worse.

nettersrl commented 2 years ago

ok nice, using the prebuilt package, it works fine as you mentioned! Thank you and have a nice holiday