2x ASUS 7950 low voltage problem

[hsk81]

Hi everybody,

I'm relatively new to mining and have the following setup:

Graphic cards: 2x ASUS HD7950 DirectCU II TOP GDDR5 3GB 384Bit AMD Radeon DX11.1;
MB: ASUS M5A99FX PRO R2.0 AMD 990FX/SB950 DDR3 2133MHZ(O.C.);
CPU: AMD FX X4 4100 SOKET AM3+ 3.6GHZ 12MB CACHE 32NM;
RAM: CORSAIR 8GB VENGEANCE DDR3 1600MHZ CL10 (single module);
PSU: GX 80PLUS 750W (bronze, 85% efficiency);

I don't use any risers; if a test each device separately I get a performace of about 580kH/s @ 1.090V (did not tweak extremely long, might get higher numbers); this is fine. But when I try both cards together then one of then cards voltage drops to 0.800V with corresponding engine clock down to 300Mhz; the other cards is fine but my overall performance is lower than 580kH/s.

I tried increasing the SB (Southbridge) voltage in the BIOS but that did not help at. Tried to combine a Sapphire 7950 with an Asus 7950 and got the same problem: One of the cards has systematically significant lower voltage (and therefore lower engine clock). I'm quite stuck .. I suspect the motherboard, but I feel with all its tweakable parameters a litte overwhelmed .. 

[toolbag]

Are you sure it's not just a faulty/undersized power supply? Many power supplies can't put out anywhere near their rated wattage. I just got into mining myself with a single 7870 and I bought a 630W power supply just to be sure I had enough overhead to handle the card.

[gurvy]

second vote for the power supply. they can draw alot

[hsk81]

@toolbag and @gurvy: Thank you very much for the answers .. is there maybe a way to determine by other means if the PSU is overrated? I guess I need to get a 1200W gold and then retest. Will report my findings.

[Spotter]

I'm running 2 GPUs on 700w PSU. One 7970 and second 7950. I get 1300Khash with GUIMiner scrypt (Litecoin).

There could be cooling issue as I had. After about 5 minutes upper GPU swhiched off...
I installed additional vent to blow between GPUs and turned voltages down. GPU temp dropped 18-20c

[hsk81]

@Spotter: I don't think it's a cooling issue, since my GPUs do not draw enough power/voltage from the very start (they don't even heat up in the first place to create a cooling problem).

[Spotter]

You probably right.....
But anyway you dont need 1200w to run 2x 7950. I suggest 750-850.
700 is pushed to the last. Before I turned voltages down I measured 760w power draw from plug!?. I cant imagine how the 700 withstands but it does.

[toolbag]

I once diagnosed a power supply issue just by measuring the 5v and 12v lines on a hard drive power connector while the system was running using a run-of-the-mill multimeter. Turns out having only 4.6v or so on the 5v line was enough to cause random lock-ups and crashes. Replaced the power supply and all was well.

[hsk81]

Nope, the PSU was not the cause I replaced my 750W with an (oversized) 1000W PSU Bronze XIGMATEK and have still the exact same problem:

Good GPU:

Code:

GPU 0: 270.1 / 284.3 Kh/s | A:3  R:0  HW:0  U:1.74/m  I:20
76.0 C  F: 36% (2125 RPM)  E: 900 MHz  M: 1250 Mhz  V: 1.090V  A: 99% P: 20%
Last initialised: [2013-05-01 17:58:27]
Intensity: 20
Thread 0: 234.8 Kh/s Enabled ALIVE

Evil GPU:

Code:

GPU 1: 325.3 / 253.8 Kh/s | A:4  R:0  HW:0  U:2.32/m  I:20
36.0 C  F: 0% (1192 RPM)  E: 300 MHz  M: 150 Mhz  V: 0.850V  A: 0% P: 20%
Last initialised: [2013-05-01 17:58:27]
Intensity: 20
Thread 1: 297.2 Kh/s Enabled ALIVE

I'm using the following CG miner command:

Code:

cd /opt/cgminer && rm -f *.bin && ./cgminer --scrypt --auto-fan \
    --url $URL:$PORT --userpass $LTCUSER:$LTCPASS \
    --thread-concurrency $THREAD_CONCURRENCY -I $INTENSITY \
    --gpu-engine $GPU_ENGINE,$GPU_ENGINE \
    --gpu-memclock $GPU_MEMCLOCK,$GPU_MEMCLOCK \
    --gpu-powertune $GPU_POWERTUNE,$GPU_POWERTUNE \
    --gpu-vddc $GPU_VDDC,$GPU_VDDC

with the following parameters:

Code:

INTENSITY=20 ; ## -I $INTENSITY
THREAD_CONCURRENCY=16384 ; ## --thread-concurrency $THREAD_CONCURRENCY
GPU_ENGINE=900 ; ## --gpu-engine $GPU_ENGINE
GPU_MEMCLOCK=1250 ; ## --gpu-memclock $GPU_MEMCLOCK
GPU_POWERTUNE=20 ; ## --gpu-powertune $GPU_POWERTUNE
GPU_VDDC=1.090 ; ## --gpu-vddc $GPU_VDDC

and of course:

Code:

export DISPLAY=:0 ;
export GPU_MAX_ALLOC_PERCENT=100 ;
export GPU_USE_SYNC_OBJECTS=1 ;

If I run each card separately then I get 580kH/s and combined a only get 524kH/s .. well at least I know now that is must be something else. But what? 

[Apothem]

Are you running them linked in Crossfire?

Also I've found that sometimes it can just be the GPU that does this. You might just have a bad GPU if everything else is running fine.

Also stock clock settings on the 7950 should be netting you around 550 khash on average, depending on the card of course.

Also the boost clock can get in the way of this as well, as it has a tendency to throttle your GPU usage. I'm still looking for a solution to this myself.

[bitcoinminerz]

Hope you manage to fix it buddy.

[hsk81]

@Apothem: Nope no crossfire, did try it though without any observable performance gains or losses. I also tried to plug in the thinner Sapphire 7950 into PCIe slot #1 and the larger Asus 7950 into slot #2 (instead of slot #3) but no gains here either:

Good GPU: (Sapphire)

Code:

GPU 0: 294.3 / 274.6 Kh/s | A:5  R:1  HW:0  U:2.14/m  I:20
76.0 C  F: 85% (4016 RPM)  E: 900 MHz  M: 1250 Mhz  [b]V: 1.169V[/b]  A: 99% P: 20%
Last initialised: [2013-05-01 22:14:20]
Intensity: 20
Thread 0: 276.9 Kh/s Enabled ALIVE

Evil GPU: (Asus)

Code:

GPU 1: 253.5 / 274.6 Kh/s | A:2  R:1  HW:0  U:1.43/m  I:20
37.0 C  F: 8% (1230 RPM)  E: 300 MHz  M: 150 Mhz  [b]V: 0.850V[/b]  A: 0% P: 20%
Last initialised: [2013-05-01 22:14:20]
Intensity: 20
Thread 1: 230.1 Kh/s Enabled ALIVE

An interesting observation is that if I plug my two Asus 7950 into the PCIe slots #1 and #3 the first Asus 7950 has a voltage of V: 1.090V while my Sapphire here uses V: 1.169V. The conclusion here is that my problem does not seem to be caused by an upper voltage limit, since V: 1.169V + V: 0.850V  (Sapphire+Asus) is larger than V: 1.090V + V: 0.850V (Asus+Asus). Observing this eliminates (almost) all my suspicions that my motherboard might be the root of all evil (was very unlikely anyway, since its brand new and was not the cheapest available).

And since all cards deliver individually excellent hashing rates (595Kh/s for both Asus and 630Kh/s for Sapphire), I started to think that software might be the reason why I can't increase my performance with the number of graphic cards linearly.  Another option might be that I need those powered risers (PCIe extension cables) to solve my voltage issues; since I don't have those cables available right now I'll continue to pray to the ASIC god to increase my hashing power through a miracle .. damn. 

[hsk81]

Heureka! So the issue was software drivers: following the instructions at http://www.reddit.com/r/litecoinmining/comments/1c4n7i/guide_getting_started_with_n_x_radeon_7950_on/ and updating my AMD drivers (Step #8) did the trick .. but now I've another problem: Overheating  

Hot GPU:

Code:

GPU 0: 593.7 / 466.6 Kh/s | A:14  R:0  HW:0  U:2.61/m  I:20
87.0 C  F: 85% (3592 RPM)  E: 1100 MHz  M: 1500 Mhz  V: 1.090V  A: 99% P: 20%
Last initialised: [2013-05-02 13:58:33]
Intensity: 20
Thread 0: 597.5 Kh/s Enabled ALIVE

Cold GPU:

Code:

GPU 1: 596.8 / 590.7 Kh/s | A:27  R:0  HW:0  U:5.04/m  I:20
72.0 C  F: 37% (2138 RPM)  E: 1100 MHz  M: 1500 Mhz  V: 1.090V  A: 99% P: 20%
Last initialised: [2013-05-02 13:58:33]
Intensity: 20
Thread 1: 595.9 Kh/s Enabled ALIVE

Although the upper RPM is crazy the heat conduction does not seem to be as great as the lower GPU; I opened my case to increase air circulation but does not seem to help that much .. I think I'll replace my GPU 0 which is a fat ASUS with a thinner Sapphire to have more space for air circulation, or just rip everything out of the case and put them into an open laundry basket or something like that.

[z3r0]

Make sure you have a dummy plug in the the 2nd gpu bro, or plug in an extra screen if u have.. reboot n try again.. hope that helps

[hsk81]

@z3r0: Hi, thanks for the tip; do you mean this would lower my GPU temperature? Where can I get such dummy plugs? I've seen some people also attach a resistance to a dummy but don't know what that actually does ..

[hsk81]

So as a conclusion: It is possible to run 2x ASUS 7950 without low voltage problems by upgrading to the newest drivers (for Ubuntu). I solved the overheating problem by underclocking the hotter GPU:

GPU 0: 75.0 C

Code:

GPU 0: 442.0 / 434.6 Kh/s | A:103  R:0  HW:0  U:3.04/m  I:20
75.0 C  F: 87% (3582 RPM)  E: 725 MHz  M: 850 Mhz  V: 1.090V  A: 99% P: 20%
Last initialised: [2013-05-03 16:02:34]
Intensity: 20
Thread 0: 443.0 Kh/s Enabled ALIVE

GPU 1: 72.0 C

Code:

GPU 1: 599.7 / 576.8 Kh/s | A:89  R:12  HW:0  U:2.62/m  I:20
72.0 C  F: 36% (2092 RPM)  E: 1100 MHz  M: 1375 Mhz  V: 1.090V  A: 99% P: 20%
Last initialised: [2013-05-03 16:02:34]
Intensity: 20
Thread 1: 599.4 Kh/s Enabled ALIVE

The overall performance I get is slightly more than 1Mh/s:

Code:

cgminer version 3.1.0 - Started: [2013-05-03 16:02:29]
--------------------------------------------------------------------------------
 (5s):888.3K (avg):1.009Mh/s | A:205  R:12  HW:0  U:5.8/m  WU:886.2/m
 ST: 2  SS: 0  NB: 11  LW: 191  GF: 0  RF: 0
 Connected to ltcmine.ru diff 128 with stratum as user hsk81_0
 Block: d5e66ac92a1de975...  Diff:31.6M  Started: [16:35:18]  Best share: 127K
--------------------------------------------------------------------------------
 [P]ool management [G]PU management [S]ettings [D]isplay options [Q]uit
 GPU 0:  75.0C 3584RPM | 396.6K/434.2Kh/s | A:107 R: 0 HW:0 U:3.02/m I:20
 GPU 1:  72.0C 1987RPM | 499.6K/577.1Kh/s | A: 99 R:12 HW:0 U:2.79/m I:20
--------------------------------------------------------------------------------

I use the following command line:

Code:

#!/bin/bash

export DISPLAY=:0 ;
export GPU_MAX_ALLOC_PERCENT=100 ;
export GPU_USE_SYNC_OBJECTS=1 ;

LTCPOOL=${1-"http://my-pool.tld:port"}
LTCUSER=${2-"my-username"}
LTCPASS=${3-"my-password"}

INTENSITY=20 ; ## -I $INTENSITY
THREAD_CONCURRENCY=16384 ; ## --thread-concurrency $THREAD_CONCURRENCY
GPU_ENGINE=725,1100 ; ## --gpu-engine $GPU_ENGINE
GPU_MEMCLOCK=850,1375 ; ## --gpu-memclock $GPU_MEMCLOCK
GPU_POWERTUNE=20 ; ## --gpu-powertune $GPU_POWERTUNE

cd /opt/cgminer && rm -f *.bin && ./cgminer --scrypt --auto-fan \
    --url $LTCPOOL --userpass $LTCUSER:$LTCPASS \
    --thread-concurrency $THREAD_CONCURRENCY -I $INTENSITY \
    --gpu-engine $GPU_ENGINE --gpu-memclock $GPU_MEMCLOCK \
    --gpu-powertune $GPU_POWERTUNE

If somebody has better performance with both ASUS 7950 running at lower or equal 75 C degrees, then please share it. Thanks. 

[hashrateproducts]

One issue I've come across is with higher temps the electricical resistance increases in the GPU PCBs and the rig draws more power at the wall to compensate for this resistance to meet the voltage requirements. When one of my 3x7950 rigs was running 75C+ it was drawing just over 1000w at the wall (mind you it should only draw 200w per card).

With an open air case and riser cables you should be able to bring those temps down considerably. Also get a kill-a-watt! You might be able to borrow one from the library if you dont want to buy one, most cities have energy conservation plans and lend them out.

Not only will you prolong the life of your cards, save energy, but can also clock them higher to get over ~630 khs. I have mine clocked at 1100 core and 1600 mem running at 59C with 60% fan speed.

[kanta]

Best check to see if your are overdrawing your power supply's rated capacity is to take a power measurement at the wall. Once you go above stock clocks on either your cpu or gpu the current draw rises quickly. The manufacturer of you power supply is also a pretty big factor.  Cheap PSUs are often inefficient

[kevinmyers80]

controls the thermal paste,  I had given problems

[pythons]

With an open air case and riser cables you should be able to bring those temps down considerably. Also get a kill-a-watt! You might be able to borrow one from the library if you dont want to buy one, most cities have energy conservation plans and lend them out.
 

yes, kilowatts are great, everyone should get one (they are cheap)