Pages:
Author

Topic: Sapphire 7950 @ 650-800Kh/sec! lulz - page 2. (Read 6271 times)

full member
Activity: 208
Merit: 100
May 17, 2013, 12:56:51 AM
#19
Ok found the sweet spot:

GPU clock - 1075Mhz
Mem clock - 1500Mhz
Vddc 1150 mv
Thread concurrency - 16384
Intensity - 20
GPU threads - 2
Vectors 1
Power consumption: 2x7950s = 550watts
hero member
Activity: 532
Merit: 500
bearded, drunk, fat, naked
May 11, 2013, 08:36:48 AM
#18
430 W from the wall for 3 7950s is insane. How is that even possible? I cannot get my sapphire OCs under 968 voltage in sapphire trixx at stock voltage (925/1250) or they will become unstable after a few hours. The rig will draw 590-600 W at that point.
full member
Activity: 182
Merit: 100
May 10, 2013, 08:56:29 PM
#17
You got to find the sweet spot that's most profitable. 430W is good. I run mine 1000/1500/1060 and it runs about 700W for three 7950.
full member
Activity: 208
Merit: 100
May 10, 2013, 07:05:59 PM
#16
cgminer just crashed, think i'll just run with e:1000 m:1300 vddc:1050.

With the overclocker before the cgminer crash (e:1030 m:1550 vddc:1050) the miner was drawing 590 watts, with the current overclock its drawing 435 watts.

The overclock that led to cgminer failing after 2hrs, would only get us an extra 60khash/sec. I think with the reduce power usage and still mining about 600Khah/sec, the reduced power consumption makes up for the loss in hash rate.
full member
Activity: 208
Merit: 100
May 10, 2013, 06:04:26 PM
#15
Me too

I can live with 1200Khash/sec (total).

Now just need to let it do its thing for 3 months and it'll pay itself off so long as the price of ltc/btc doesn't drop.
hero member
Activity: 854
Merit: 1000
May 10, 2013, 05:45:53 PM
#14
630 is good.  Glad to see you raised the bar from 100k!
full member
Activity: 208
Merit: 100
May 10, 2013, 05:32:47 PM
#13
with 22400 TC, I 20, I can do 1030/1550.

I got 8GB of system ram.

Seems to do approx 630~Khash/sec.

Thanks for your help. Smiley
hero member
Activity: 854
Merit: 1000
May 10, 2013, 05:16:06 PM
#12
"can't run 1125/1550."

You can at thread concurrency 8192.  Start small.


You need to re-read thread.  Start at stock core and memory.  Have intensity at 13.  Raise threads as high as you can.  Start at 8192 and increase 2048 at a time till you can get as high as you can.  Then increase core & memory.  Last...increase intensity and powertune.

Edit...you also need to make sure you have enough memory on board to do scrypt hashing.
full member
Activity: 208
Merit: 100
May 10, 2013, 04:27:46 PM
#11
can't run tc higher than 22400, thats what it autodetects anyway.

Need to figure out if core or mem is too high, either way can't run 1125/1550.
hero member
Activity: 854
Merit: 1000
May 10, 2013, 01:43:56 AM
#10
With intensity at 13, powertune 0.  try this:

 start with
--thread-concurrency 8192 and add 2048 to it at a time till you find the highest
value it will start successfully at.

Then once stable increase intensity...should get up to 18 19 or even 20. Then increase powertune.

Source:
https://github.com/ckolivas/cgminer/blob/master/SCRYPT-README


Hw error because card is too much overclocked for settings.  You need to get stable threads at stock clock.  Then start increasing core & memory.  Finally after everything is good increase powertune and intensity.  

Keep trying...you will get it above 600.  Good luck.
full member
Activity: 208
Merit: 100
May 10, 2013, 12:32:38 AM
#9
oops had powertune set to 10 and within about 30-45mins cgminer crashed.

will set it back to 15 and also set TC to 14366

setting TC to 14366 results in HW errors (i delete the bin file before re-running cgminer if making changes), so I've dropped off the TC from the batch file.

hmm pc might need a reboot, as now going back to what i used one of the gpus can't seem to break past 350khash/sec.
full member
Activity: 208
Merit: 100
May 10, 2013, 12:24:25 AM
#8
Thanks for the link and info Smiley

The heat was running high due to how cgminer was configured to run, looks like the person who set it wanted low fanspeed and ok with it sitting on 81C.

I've since adjusted the batch file and so configured cgminer to run at 85% fan which keeps temps at 69C & 59C for GPU0 & GPU1.

In the future I plan to replace the motherboard with the gigabyte one that has like 5-7 pci-e slots on it so I can better space the gpus, as the 10C difference is no doubt due to just the one expansion card space between each GPU.

hero member
Activity: 854
Merit: 1000
May 09, 2013, 11:51:06 PM
#7
Powertune explained:

https://bitcointalksearch.org/topic/cgminer-powertune-explained-192210


I also believe when you increase thread concurrency it will make the card more efficient (less heat more hash).
full member
Activity: 208
Merit: 100
May 09, 2013, 11:47:38 PM
#6
Cheers, what does powertune do? I only have it in the batch file coz of what I found worked for others...

I'm guessing its managing the vddc and boosting it by 20%?
hero member
Activity: 854
Merit: 1000
May 09, 2013, 11:42:13 PM
#5
Back down powertune to 15 and make sure thread concurrency is set.  I have mine set at 14366.
full member
Activity: 208
Merit: 100
May 09, 2013, 11:33:46 PM
#4
Running:
1000 core, 1250 mem, 1050 vddc, I 20. Now getting 580-600Khash/sec.

Temp sitting at 81C~

I don't think I could o/c it any further without temp rising to an unsustainable amount.

Will let it run to test stability.
hero member
Activity: 854
Merit: 1000
May 09, 2013, 11:27:02 PM
#3
You can get up to around 650.  Add thread concurrency to your bat file.  Also run core at 1125 and memory 1550
full member
Activity: 208
Merit: 100
May 09, 2013, 11:08:15 PM
#2
Fixed it,

Ran the amd removal utility to cleanout the catalyst/amd software.

Installed Catalyst 12.8 and AMD SDK 2.7.

Installed CGminer 2.11.4.

First batch file:
setgpu.bat
setx GPU_MAX_ALLOC_PERCENT 100
setx GPU_USE_SYNC_OBJECTS 1

Second batch file:
cgminer_run.bat
cgminer -o server -u blah -p blah --scrypt -s 2 --lookup-gap 2 -I 18 -g 1 -w 256 --auto-fan --gpu-fan 0-80 --temp-target 81 --temp-overheat 87 --temp-cutoff 93 --gpu-powertune 20 --no-submit-stale

And get the following (currently tweaking undervolting, sitting the card on 1031 vdcc atm.

For both 7950 gpus, temp is sitting just under 80C.



Is it possible to better 535Khash/sec ? Or should I just stop now lest I get back to where I was before not getting more than 100Khash/sec and thinking I killed a card! Smiley
full member
Activity: 208
Merit: 100
May 09, 2013, 09:43:02 PM
#1
G'day

Been playing with my system trying to eek out the best hash rate and now, no matter what I try it won't hash more than 100Khash/sec.

I was using 13.4 Cat + 2.8 SDK. Tho from researching to try and sort my issue looks like I should be using 12.7 + 2.7...

I'm going to try installing 12.8 CAT + 2.7 SDK and see how this goes. I've reset everything back to defaults. I have a bad feeling my card is "fried" even tho I haven't overvolted the thing. All I was doing was undervolting. I'm worried I may have "fried" it to some extent as it got quite hot and ended up crashing. But when I run MSI Kombustor no artiffacts or anything in its 3d testing... Its just cgminer that I have an issue with.

Any ideas? This is 1 GPU out of a 2x7950 setup with the 2nd 7950 working fine and getting greater than 550Khash/sec. Thus why I think the card has become borked. Also tried booting off it with the bios switch in the 2nd position, still no better Sad

Well wish me luck a reinstall of catalyst and to a different version may help fix it?

If anyone has any ideas, please let me know. Thanks!
Pages:
Jump to: