Bitcoin Forum

Hi,

So here is my problem, I hope someone can help me out.

First of all, my rig is composed of 3 7950s at stock clocks with one 1000w KingWing platinum PSU. One video card hooked up to a display and the 2 others to a dummy plug. It is running on Ubuntu server 12.04, cgminer 3.1.1 with latest AMD drivers.

Everything is running without a glitch except when I go over an intensity of 17, problems start at 18. What happens is the card hooked to the external display runs fine and stable but the two others are throttled: once every approximately 10 seconds, they will drop from 515 kh/s to approx. 420 kh/s for a couple of seconds, and then get back to 515 in this particular exemple. I can easily get a single 7950 to 630 kh/s when overclocked but I am far from getting there with all my cards.

I suppose the problem is not one coming from the PSU since at stock clocks and intensity of 18, the power from the wall reads at max 815 w. I should have enough room.

It is neither a heat problems since the cards at those settings are running at 82 celcius max fans under 4000 rpms.

I thought this could be a driver issue, not letting me use 100% of all GPUs all the time to protect them... dunno. I didn't want to start testing bunch of drivers version since this would be really time consuming. Maybe someone knows of a version working better than others? Tried adjusting --gpu-powertune without much luck, though it seems to help a little bit...

No HW errors, rejected shares of 1 to 1.5%.

Any ideas of what would cause that?

Thanks for your help!

edit: export DISPLAY=:0, export GPU_MAX_ALLOC_PERCENT=100 and export GPU_USE_SYNC_OBJECTS=1 are all set.
here are the settings I use in this example: --shaders 1792 --thread-concurrency 21712 -g 1 --lookup-gap 2 --gpu-vddc 1.03
so nothing to fancy, just trying to get stable at high intensity... thx :)

why do yo uhave the other two hooked to a dummy plug?

you need to post some more specs.

also post your conf file.

you need to add
--gpu-powertune 20
to your config

I edited my original post for config. Heard it would help with dummy plugs so I tried. I am having the same problems with or without them.

Tried that, same results.

What specs would you like me to provide?

Ok do this - its like mixing music - just set everything to zero and start again.

so you run a .Bat file obviously - set it to just start CGminer and only set intensity – I 18

and tell me what happens - so copy your .bat and just run a vanilla one take out the screwing with the voltage and TH and Sharders - if it works , slowly bring things back until you find the problem.

** you can leve the SET group i.e GPU_MAX_ALLOC_PERCENT=100 etc.

So dont run intensity 17?

I run mine at 12, anything higher just throttles or crasches.

Now with --gpu-powertune 20 PSU is sucking up to 910w from the wall. I am having the same unstable behavior only from the 2 GPUs not hooked. The one hooked to the display is fine.

was that in response to me?

also I run 4 cards at -I 19,20,20,20

no problems - once every couple of days the pc would freeze - but i'm working on that , have it solved.

also you are very close to your Watt limit i'd say.

The thing is this rig *should* be able to run full intensity and get better hash rate, so I am doing what I can to achieve it, you know :) anything over 17 intensity not working isnt normal.


thx digitalindustry, will try that and get back when I have results.

Not exactly, just providing more info. I know I am close, I will daisychain another one soon but would like to get everything stabilized by then since I thought I had enough room so the psu wouldnt cause this problem...

"1000w KingWing" - also be a little concerned about any PSU that makes me think of parts of a chicken. and/or hungry .

Haha :P yeah, well I thought it was a good deal (180$) for a 1000w platinum that had good reviews on the web... But I get you.

Ok so even with nothing but SET groups and intensity, I get the same behavior. Maybe a little bit better, but still same problem.

And again, the only GPU that is not affected is the one hooked to the display. Dummy plugs on the others change nothing. --gpu-powertune 20 neither.

what brand are your GPU?... i can help you  with some info  if you have MSI tf3. pacpac

They are powercolors. If you have any clues, don't hesitate to share :)

with aticonfig --od-getclocks --adapter=all I can see that the gpuload gets down to 64%, then back to 99, the back to 64... like that.

annoying!

And this is with --gpu-powertune 20.

How much Ram do you have?  You could be maxing that out possibly

I have 8 GB. Didn't check the clock at which the system put it though. Could that change something? I have an Athlon 2

Still thinking about a potential driver issue... Because the default adapter always stays at 99% load while the two others go down, then up, than down... like that. Anybody mining on ubuntu 12.04? If so, what driver version are you using?

The best Power Supply that I have found is the XFX ProSeries 1050W
http://www.newegg.com/Product/Product.aspx?Item=N82E16817207021

Currently running 3x 7950s and have overhead for 1 more atleast.

$190 and its fully modular, I'm very impressed by this power supply.

I'm using XFX Power supply and XFX video cards. 5 year warranty on the power supply and lifetime on the cards. I don't know how the other brands compete.

I've seen that happen with too low a voltage, if you haven't changed the voltage I'd look at trying another PSU.

I use corsair HX 850 and it powers 3x 7950's with ease.

I also have seen this happen with voltage/over-clocking issues. Sometimes there's a spot that's just unstable, how do you have the cards clocked?

You could also mess with some of the settings from here https://github.com/litecoin-project/litecoin/wiki/Mining-hardware-comparison

Tried stock clocks and voltage, undervoltage, overvoltage, can't do. I think I tried all the config possible :P

I will get this to work on a new windows7 install and get back to you to see if the drivers I was using under ubuntu server had any problems that the windows' ones don't.

If same problem arises, will try with another psu.

I run 4x7950 http://www.newegg.com/Product/Product.aspx?Item=N82E16814125414 at ~610khash/s each with stock config and with temp of : 61°C, 77°C, 79°C, 81°C
--shaders 2048 --intensity 20 --worksize 256 --thread-concurrency 21712 -g 1

Still same problem. Thx for sharing your config tho!

Are you overclocking your cards? With my KingWing lazer platinum 1000w I was reading a little over 1000w from the wall when I would start overclocking a bit, that is with 3 7950s. What are your cards? How do you measure that you would have room for a 4th one?

Thank you :)

P.S: I am looking to buy a PC Power & Cooling 1200w Silencer MK 3 platinum for 4 7950s. I returned the KingWing.

you might think it's driver but i'm calling Power on this one, either drawing though the PCIe or just right on out of the Kingwing !

your PSU says 1000W right? you know a KingWing isn't going to Run 1000W - you are running 3x cards and a mobo, and a HD?   

cya later KingWing, likely the draw on the Pcie is too much , also does it outright just freeze sometimes?

if so that's power , its happened to me , how i actually tested it was to turn my vacume on , it acted like it was drivers , but its a power draw issue.

that's my best guess,

pull one card out and see if it still does it , (or have you done that already? ) 

so just run two , see what happens.

That IS a heat problem. If you're reaching that high on stock clocks, then you can't push those cards any further and need some better cooling.

I returned the KingWing today, I'm getting another one and once it's received, I will then post the results. I am thinking PC Power & Cooling Silencer MK 3 1200w. Not sure if it's worth going with the corsair ax1200wi... it's 40$ more. Still thinking about it. hendo420 also told me about the XFX ProSeries, which I am considering too. Found the corsair ax1200wi for 300$ on NCIX, I'm getting this one. As of now, this rig is not mining until I receive the new psu. Also, I wasn't using powered risers since the motherboard was 3way Crossfire ready, but in my next attempt I will have all cards on powered risers.


Why would the default adapter work flawlessly then? It is not any cooler than the other 2. I've run those cards to 90+ celcius without seeing throttling on my other rig... But I am looking into trying to have better cooling, I know they are a little hot. I've reapplied thermal paste and removed their plastic cases, which helped them get a couple of degrees.

Thx for your insights!