6 gpu mining rig shows considerable loss of hashing power. is this typical?

[rannici]

hey. i'll probably post this again to mining, once my account stops being noobified, and if i don't get the answer here,

but. i've just built my first 6 gpu rig, running with 3 radeon 5970s.

but, if i run one card, i get what i expect to be normal hash rates in the 350 range per gpu.
run two, all four gpus drop to around 320 ish.
run all three and i'm at around 295, as evidenced by the following image.

http://s1.postimage.org/6odho6o3a/Screen_Shot_2011_08_18_at_11_34_51_AM.png

is this... normal?
i only have a dual core in the system, am running windows 7 with 4gb of ram.
...

all of the cards are overclocked to 800, with a 300 memory clock rate.
using guiminer with phoenix, and the newest phatk kernel.

...
in my single card rigs i overclock my 5970s to a safe 820/300,
but the third card in the 3 card mining rig has problems when i do that, so i've found 800 to be a stable number.

still.
am i wrong in expecting that each gpu should produce 350 m/h? or is this typical?

any help would be greatly appreciated.
let me know if you need more information.
thanks.

[1714071795]

1714071795

[1714071795]

1714071795

[hdfxst]

what power supply are you running?

[Bert]

What are the temperatures like ? Any cooling issues ?
Very high temperatures can force the GPU's to underperform to avoid physical damage.

EDIT: Just checked through your previous posts for more background information. And since you already have 4 Ghash/sec of mining equipment running you know all about good cooling. You can ignore my questions.

[rannici]

it's a 1200 watt coolermaster.

i have the fans set to run at a constant 80%, because the air conditioning in the closet they are hosted is not great.
but at that setting it seemingly keeps the cards at a consistent 79 degrees C.

would cooling cause a constant slowdown in hash?
it's been my experience that the videocards simply stop functioning if they are too hot or are overclocked beyond their capability.

this is a consistent slowdown on each core of about 50 mh/s when all six are running.

i'm debating taking one (or two) out permanently and building another computer, as that would be a gain of at least 200 mh/s.

[ArtForz]

Yes, you should be seeing next to no hashrate drop with multiple GPUs (< 1%).

5970s go into thermal throttling when core hits 100°C or the VRMs hit 125°C.
But that shows up as random drops in hashrate, not a consistent slowdown.

PSU issues tend to lead to lockups and BSODs, not cause reduced hashrate.

Random shot in the dark, driver/sdk. Maybe try different/older catalyst versions?

[jrtubs]

it sounds like u dont have enough power to run all your cards, i know with my rigs 5870 pulls about 230 watts, not sure what your card would be running but lets just say 200watts a card, 6 cards is 1200watts, the over head on a power supply be about 100 so 1300watts to max it totally out. its better to run 5 cards and over clock it then trying to get 6 cards running on one rig.

[MacT]

 I have a similar problem in just a 2-card setup.  Individually they will mine at 100%. used together they will only use 70% and 33% of each gpu.   Haven't figured it out yet but definitely is not a power issue in my case.  I'll post here if it gets fixed and may help with your problem.

[jrtubs]

I have a similar problem in just a 2-card setup.  Individually they will mine at 100%. used together they will only use 70% and 33% of each gpu.   Haven't figured it out yet but definitely is not a power issue in my case.  I'll post here if it gets fixed and may help with your problem.

I would double check your settings sounds like each of your miner might be trying to run on the same video card instead of [0-0] on both miner vs [0-0] and [0-1] i had that happen to me a few times just had to double check make sure the settings are right.

[trentzb]

Do you have a kill-o-watt meter or know what you are pulling at the wall? I don't think it is a power issue that is dropping your rate however 1200W for 3 5970s seem a little tight. Don't 5970s do about 380-400W stock?

[MacT]

I would double check your settings sounds like each of your miner might be trying to run on the same video card instead of [0-0] on both miner vs [0-0] and [0-1] i had that happen to me a few times just had to double check make sure the settings are right.

 Thanks- I dont want to hijack this too much, hehe.  In my case, yes I have made sure that one card is running 0-0 cypress and the other is set 0-1 juniper. Also I set them to different cpu affinity- but NO LUCK.

[rannici]

in my case also, they are also all set to different gpus.

...
thanks for the advice on the psu.
this wiki also confirms that advice, with their triple 6990 setup:
https://en.bitcoin.it/wiki/Mining_rig

...
you're right, i'm probably pushing my psu beyond the stated watt limitation.
i've read that it's possible to do this with good brands, but it will significantly decrease the life of the psu.
.. though i'm not sure that that would cause the decrease in hashing power described here,
but i will be buying a 1500 watt psu in the near future.

[jrtubs]

in my case also, they are also all set to different gpus.

...
thanks for the advice on the psu.
this wiki also confirms that advice, with their triple 6990 setup:
https://en.bitcoin.it/wiki/Mining_rig

...
you're right, i'm probably pushing my psu beyond the stated watt limitation.
i've read that it's possible to do this with good brands, but it will significantly decrease the life of the psu.
.. though i'm not sure that that would cause the decrease in hashing power described here,
but i will be buying a 1500 watt psu in the near future.

would be cheaper to use 2 power supplies then to buy a 1500watt those are really pricey.

[ArtForz]

3 5970s at 800/300 are well < 900W, any half-decent 1.2kW should be able to handle that fine.

[rannici]

i never really understood what this meant..

http://www.geeks3d.com/20110622/vga-hack-test-dummy-vga-dongle-win7-three-radeon-hd-5770-multi-gpu-opencl-single-monitor/

... would doing this solve the problem?

i read there is a way to do this by editing windows HKEY registry settings...

[Bert]

With Catalyst 11.7 and later under windows this is no longer required. And it was required to allow the OS to be able to use the cards, but that is not your problem since you can see all 6 GPU's.

[max in montreal]

I have a similar problem. 3X 5970 and windows 7 sees all gpu's. When i mine with them individually all of them work. I can get 5 GPU's to mine but not 6... My COPU is always at 100%, but thats the ccc cpu bug i think...

[Bobcat]

hey. i'll probably post this again to mining, once my account stops being noobified, and if i don't get the answer here,

but. i've just built my first 6 gpu rig, running with 3 radeon 5970s.

but, if i run one card, i get what i expect to be normal hash rates in the 350 range per gpu.
run two, all four gpus drop to around 320 ish.
run all three and i'm at around 295, as evidenced by the following image.

http://s1.postimage.org/6odho6o3a/Screen_Shot_2011_08_18_at_11_34_51_AM.png

is this... normal?
i only have a dual core in the system, am running windows 7 with 4gb of ram.
...

all of the cards are overclocked to 800, with a 300 memory clock rate.
using guiminer with phoenix, and the newest phatk kernel.

...
in my single card rigs i overclock my 5970s to a safe 820/300,
but the third card in the 3 card mining rig has problems when i do that, so i've found 800 to be a stable number.

still.
am i wrong in expecting that each gpu should produce 350 m/h? or is this typical?

any help would be greatly appreciated.
let me know if you need more information.
thanks.

This is might be your solution, I am getting the same issue when run 4x6970 (issue might be with flag setting on the GPU)
I used GUIMiner 701

GPU1  -v -w128 -f4
GPU2  -v -w128 -f4
GPU3  -v -w128 -f4
GPU4  -v -w128 -f4

When i used these setting above for all my 4 cards I drop approx 50 mhash per card, i discovered that 2 of my card (GPU1, and GPU4 flag setting screw up the mhash, u can tell when the mouse movement is lagging)

So I endup take off the the flag setting on one of the card like below or vice of versa, in the event if your system crash or hang due to over clock and the machine reboot then you might need reverse the flag setting meaning remove -f4 on GP4 and add -f4 it back to GPU1 or vice of versa. Also ensure your Display Properties have extented all your video card.
since u have 6 gpu you have to trace down which cards are conflict then work it from there. also u can oc core voltage to 931 and pwr control set to 20%it would gain 20-50 mhash more per card and of course have to increase the fan too, but be aware it  will used more power. Good luck
GPU1  -v -w128
GPU2  -v -w128 -f4
GPU3  -v -w128 -f4
GPU4  -v -w128 -f4

or

GPU1  -v -w128 -f4
GPU2  -v -w128 -f4
GPU3  -v -w128 -f4
GPU4  -v -w128