Monero mining Windows vs Linux

[Kayote]

gents see for yourself Linux comes pretty bad and I didn't even overclocked windows which I do both on linux oc'ing the cpu with Zenstates and the gpu with overdrive.
The system is a gigabyte ab350 8gb of ddr4 2600mhz memory, Ryzen 1600, two Radeon rx 460, unlocked bios mod.
I'm using latest driver from amd on windows, on Linux I'm using standard amdgpu kernel module with the opencl libraries from amdgpu-pro (blockchain beta version).
That's over 150 h/s if considering the totals, 140 h/s is you consider the highest .
The question is, I can understand that opencl is not at the same level, but ¿wtf are doing my Ryzen cores to get 20 h/s less per core??

https://i.imgur.com/YgOK7tu.png

https://i.imgur.com/gnpxHXS.png

(sorry I can't add attachments)

[Kayote]

I used Xmr Stak https://github.com/fireice-uk/xmr-stak

the configuration is pretty much the same on linux you have to set affinity to even cores, on windows to odd cores. On Linux you can crank intensity a bit more, I set up 352 on windows you can go up to 900 intensity on linux, you might also want to increase worksize on Linux.
Large pages enabled on both systems, amdgpu.vm_fragment_size=9 on grub.

[Kayote]

I'm still investigating this. My guess is that Windows is using some turbo clocks that Linux is not using. At least I was able to replicate the same hashes by upping the Pstates0,1,and 3. Now looking much better now practically the same hashrate for the cpu. I still got to find out why there is that difference on gpu/opencl

this is cpupower frequency-info,

sudo cpupower frequency-info  
analyzing CPU 0:
 driver: acpi-cpufreq
 CPUs which run at the same hardware frequency: Not Available
 CPUs which need to have their frequency coordinated by software: Not Available
 maximum transition latency:  Cannot determine or is not supported.
 hardware limits: 1.55 GHz - 3.20 GHz
 available frequency steps:  3.20 GHz, 2.80 GHz, 1.55 GHz
 available cpufreq governors: conservative ondemand performance
 current policy: frequency should be within 1.55 GHz and 2.80 GHz.
                 The governor "performance" may decide which speed to use
                 within this range.
 current CPU frequency: 2.80 GHz (asserted by call to hardware)
 boost state support:
   Supported: no
   Active: no
   Boost States: 0
   Total States: 3
   Pstate-P0:  3700MHz
   Pstate-P1:  3600MHz
   Pstate-P2:  2000MHz

In case anybody wonders, this is how I set up ZenStates on Ryzen 1600 to get the same performance than Windows. I think the most important is Pstate-P1

zenstates.py -p 1 -f B4

PState-P0

zenstates.py -p 0 -f 94

PState-P2

zenstates.py -p 2 -f A0

temp looking good 35º!

[Brisse]

Which kernel version are you using? There was a fix in 4.14 regarding power states which ensures it can boost all the way to it's maximum clocks. Earlier kernels would not hit XFR clocks.

[Kayote]

I've been using 4.14.14 that's exactly what I thought off kernel is not hitting XFR boost clocks. Anyway I think there has been another fix, now running on 4.15-rc9 (didn't upgraded to final yet) and there is some improvement regarding OpenCL on amdgpu, not much but anyway, still behind windows performance but not for much now.

[Danks124]

Monero mining is based on a proof-of-work algorithm called CryptoNight. This algorithm was especially designed to be as inmuned to ASIC miners as possible; therefore, it is feasible to mine Monero using consumer CPU and GPU hardware. This is one of the big advantages of Monero over Bitcoin WalgreensListens