Author

Topic: best Polaris sgminer eth options? (Read 632 times)

full member
Activity: 190
Merit: 100
March 31, 2017, 12:15:50 PM
#4
that may be why on the pool side I had bad results. Hashrate is not all that counts...

I turned back to claymore, dual mining is currently better as long as DCR price is good.
sr. member
Activity: 588
Merit: 251
March 31, 2017, 09:19:28 AM
#3
Hi,
I run worksize 256, it gives me better hashrate than 192 on both hawaii and polaris cards all of them 4G.
As xI I tried a lot but could'nt get anything better than with 4620.
And 2 threads was better than 1.

Strange.  worksize 256 gives me the same problems as 128; lots of hardware errors and WU: 0.
full member
Activity: 190
Merit: 100
March 31, 2017, 07:24:37 AM
#2
Hi,
I run worksize 256, it gives me better hashrate than 192 on both hawaii and polaris cards all of them 4G.
As xI I tried a lot but could'nt get anything better than with 4620.
And 2 threads was better than 1.

sr. member
Activity: 588
Merit: 251
March 29, 2017, 10:14:25 PM
#1
With a few different sgminer forks (GG, sgminer-gm, ...), there's not just one place to discuss sgminer tweaks.  Hence this post.

What I've already figured out is that the default gpu-threads:1 and xi:1024 don't give the best performance.  gpu-threads:2 gives a small but material improvement.  I tried xi:1053 instead of 1024 based on a recommendation from Eliovp, but didn't notice any improvement.  After looking at GG's sample params for equihash-new using xi:4620, I tried xi:2310 and noticed a tiny improvement over 1024.  However the HW error rate has gone up, especially on my Asus Strix with Hynix (30HW errors after 20 minutes running).  I've tried worksize:128, and it doesn't work at all (no valid shares).

I'm running kernel 4.10.5 with AMDGPU-Pro 16.60, and sgminer-gm 5.5.5.  I'm running my Strix at 1167/2025 and my Sapphire (Samsung K4G4) at 1169/2100, getting ~28.2 and ~29.5Mh respectively.  With Claymore 8.1 I get ~28.7 and ~30.2Mh.  So does anyone have some well-tested sgminer params that are better (by at least 1%) than worksize:192, gpu-threads:2, and xi:2310?

Jump to: