Author

Topic: Different hash speed of GPU threads (Read 1948 times)

legendary
Activity: 952
Merit: 1000
May 20, 2013, 02:06:49 PM
#6
--intensity 10 --vectors 1 --worksize 256  --lookup-gap 0 --thread-concurrency 0  --shaders 0 --gpu-engine 1000 --auto-fan --gpu-memclock 900  --gpu-memdiff 0 --gpu-powertune 0 --gpu-vddc 1.150 --temp-cutoff 90 --temp-overheat 85 --temp-target 72 --api-port 4028 --expiry 120 --gpu-dyninterval 7 --gpu-platform 0 --gpu-threads 2
Overclocking will not void your warranty at all. Some of those settings are redundant, and some are uncecessary, and others still may actually be hurting performance. Try simplifying it down to this:

--auto-gpu --auto-fan --temp-target 72 --gpu-engine 600-1100 --gpu-fan 40-90 --gpu-memdiff -150 -I 7
full member
Activity: 134
Merit: 100
May 20, 2013, 01:11:39 PM
#5
Problem can not be solved with restart, because I restarted computer many times anyway and results are pretty constant.
Some say that driver 12.8 is better than 13.4, I never tried and use 13.4. I guess I am too lazy to change driver.
member
Activity: 98
Merit: 10
May 20, 2013, 09:29:19 AM
#4
Sometimes when the driver resets a GPU will underperform until reset (sometimes software is fine sometimes reboot).

full member
Activity: 134
Merit: 100
May 20, 2013, 09:26:56 AM
#3
crazyates,

Thanks for reply. I did not overclock hor undervoltage card, only underclocked the memory, cause I do not want to lose the warranty.
My settings are in the command line of bfgminer 3.0.0.

  --intensity 10 --vectors 1 --worksize 256  --lookup-gap 0 --thread-concurrency 0  --shaders 0 --gpu-engine 1000 --auto-fan --gpu-memclock 900  --gpu-memdiff 0 --gpu-powertune 0 --gpu-vddc 1.150 --temp-cutoff 90 --temp-overheat 85 --temp-target 72 --api-port 4028 --expiry 120 --gpu-dyninterval 7 --gpu-platform 0 --gpu-threads 2

I left parameter --gpu-vddc 1.150 there, but it does not do anything, because of voltage is locked, I did not bother to remove parameter. Next time I restart the miner, probably I will remove it just to be command line consistent with reality.

I tried higher intensity, but it did not bring any good except of bumping CPU load from 1-4% to 50-90%.

I saw some speed posted closer to 300-320, but most of them shows temperature in the area of 80C and higher or fan running really heavy.
But sure, I do like to find optimum parameters for the card if they are not.
legendary
Activity: 952
Merit: 1000
May 20, 2013, 12:39:28 AM
#2
They don't need to run at the same. As long as the two combined are about right for your card, that's all that matters.

My 7970 is running 2 threads, and the first one is 545MH/s, and the second is 158MH/s. Combined, that makes 703MH/s, which is right about right for my clocks.

That said, 222MH/s is pretty low for a 7850. You should be closer to 300MH/s. Can you post your settings?
full member
Activity: 134
Merit: 100
May 19, 2013, 11:02:00 PM
#1
Single videocard 7850 on a motherboard, running with intensity 10. GPU engine clock 1000, memory clock 900. Temperature 70C, fun on 40-42% load. Average performance about 222MH/S.
I noticed that thread 0 shows 91.4 MH/S, and thread 1 shows 131.4 MH/S.
How I can both threads run about the same, around 131 MH/S?

Thank for any advice.
Jump to: