Problem with 4 GPU

I put 4 R4770 to one box and they were joined by CF. Each card is working correctly in other boxes. I see 4 GPU in CCC and in Boinc.

When I give "<count>2</count>" two cards work quickly and correctly:
http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=10579276

<core_client_version>6.10.21</core_client_version>
<![CDATA[
<stderr_txt>
Running Milkyway@home ATI GPU application version 0.20b (Win64, CAL 1.4) by Gipsel
instructed by BOINC client to use device 0
instructed by BOINC client to use device 1
CPU: AMD Athlon(tm) II X4 620 Processor (4 cores/threads) 2.60017 GHz (259ms)

CAL Runtime: 1.4.467
Found 4 CAL devices

Device 0: ATI Radeon HD4700/4800 (RV740/RV770) 512 MB local RAM (remote 1855 MB cached + 1855 MB uncached)
GPU core clock: 750 MHz, memory clock: 800 MHz
640 shader units organized in 8 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads
supporting double precision

Device 1: ATI Radeon HD4700/4800 (RV740/RV770) 512 MB local RAM (remote 1855 MB cached + 1855 MB uncached)
GPU core clock: 750 MHz, memory clock: 800 MHz
640 shader units organized in 8 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads
supporting double precision

Device 2: ATI Radeon HD4700/4800 (RV740/RV770) 512 MB local RAM (remote 1855 MB cached + 1855 MB uncached)
GPU core clock: 750 MHz, memory clock: 800 MHz
640 shader units organized in 8 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads
supporting double precision

Device 3: ATI Radeon HD4700/4800 (RV740/RV770) 512 MB local RAM (remote 1855 MB cached + 1855 MB uncached)
GPU core clock: 750 MHz, memory clock: 800 MHz
640 shader units organized in 8 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads
supporting double precision

Starting WU on GPU 1

main integral, 640 iterations
predicted runtime per iteration is 350 ms (33.3333 ms are allowed), dividing each iteration in 11 parts
borders of the domains at 0 152 296 440 584 728 872 1024 1168 1312 1456 1600
Calculated about 3.28897e+013 floatingpoint ops on GPU, 2.47165e+008 on FPU. Approximate GPU time 232.576 seconds.

probability calculation (stars)
Calculated about 3.34818e+009 floatingpoint ops on FPU.

WU completed.
CPU time: 15.5533 seconds,  GPU time: 232.576 seconds,  wall clock time: 234.103 seconds,  CPU frequency: 2.60021 GHz

</stderr_txt>
]]>

Unfortunately, when I give "<count>1</count>" everything works terribly slow. CPU load at 100%, GPU load very low.
http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=10583886

<core_client_version>6.10.21</core_client_version>
<![CDATA[
<stderr_txt>
Running Milkyway@home ATI GPU application version 0.20b (Win64, CAL 1.4) by Gipsel
instructed by BOINC client to use device 2
CPU: AMD Athlon(tm) II X4 620 Processor (4 cores/threads) 2.60018 GHz (325ms)

CAL Runtime: 1.4.467
Found 4 CAL devices

Device 0: ATI Radeon HD4700/4800 (RV740/RV770) 512 MB local RAM (remote 1855 MB cached + 1855 MB uncached)
GPU core clock: 750 MHz, memory clock: 800 MHz
640 shader units organized in 8 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads
supporting double precision

Device 1: ATI Radeon HD4700/4800 (RV740/RV770) 512 MB local RAM (remote 1855 MB cached + 1855 MB uncached)
GPU core clock: 750 MHz, memory clock: 800 MHz
640 shader units organized in 8 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads
supporting double precision

Device 2: ATI Radeon HD4700/4800 (RV740/RV770) 512 MB local RAM (remote 1855 MB cached + 1855 MB uncached)
GPU core clock: 750 MHz, memory clock: 800 MHz
640 shader units organized in 8 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads
supporting double precision

Device 3: ATI Radeon HD4700/4800 (RV740/RV770) 512 MB local RAM (remote 1855 MB cached + 1855 MB uncached)
GPU core clock: 750 MHz, memory clock: 800 MHz
640 shader units organized in 8 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads
supporting double precision

Starting WU on GPU 2

main integral, 640 iterations
predicted runtime per iteration is 350 ms (33.3333 ms are allowed), dividing each iteration in 11 parts
borders of the domains at 0 152 296 440 584 728 872 1024 1168 1312 1456 1600
Calculated about 3.28897e+013 floatingpoint ops on GPU, 2.47165e+008 on FPU. Approximate GPU time 3768.49 seconds.

probability calculation (stars)
Calculated about 3.34818e+009 floatingpoint ops on FPU.

WU completed.
CPU time: 2936.19 seconds,  GPU time: 3768.49 seconds,  wall clock time: 3770.08 seconds,  CPU frequency: 2.60021 GHz

</stderr_txt>
]]>

<cmdline>b-1</cmdline> or <cmdline></cmdline> - it does not matter.

Could you help solve this problem?

I didn't think you should join them by CF if you want all GPUs to crunch wu's.

3 work properly. 1+2+3, 1+2+4, etc. 4 does not work.

3 work properly. 1+2+3, 1+2+4, etc. 4 does not work.

Have you tried taking them out of the cross fire mode and putting dummy plugs into the 3 cards that don't support the monitor?

This is how they are created. BTW, I think I read that you can use 50-150 OHM resistors, but mine are 68 OHM from Radio Shack. I think a card of 5 resistors was less than $2 USD.

Dummy Plug

The Pin Assignment can be found here:

VGA Pin Assignment

OK, thx!

I plugged 4 monitors and 4 GPUs working well!

Dummy VGA is a good solution.

OK, thx!

I plugged 4 monitors and 4 GPUs working well!

Dummy VGA is a good solution.

You're welcome and good luck!

You need something plugged into a display port on the 2nd card. If your monitor has additional inputs you can run one from that, even if you don't actually use it. They also sell "dummy plugs" that fool the card into thinking something is plugged into it. Until a port is active, the second card is "disabled" and will not crunch.

---

You need something plugged into a display port on the 2nd card. If your monitor has additional inputs you can run one from that, even if you don't actually use it. They also sell "dummy plugs" that fool the card into thinking something is plugged into it. Until a port is active, the second card is "disabled" and will not crunch.

Thanks

I just plugged a second monitor, there is no picture.
CCC in the ATI card is still disabled ....

Now you need to go to your display properties and extend your desktop onto that monitor, which will "activate" the card. So long as there is something attached to it, you only have to go through this routine once.

---

Now you need to go to your display properties and extend your desktop onto that monitor, which will "activate" the card. So long as there is something attached to it, you only have to go through this routine once.

>which will "activate" the card

THAT'S IT!

Huhuuu
and because it is so beautiful, I plug one more into the Board;-)
Hope it goes well!

Look this:

http://www.overclock.net/overclock-net-folding-home-team/384733-30-second-dummy-plug.html


THX Mr. JerWA

I had already run 2 cards and 2 monitors with 4 GPU on.
With 3 graphics adapter and 3 monitors to run again only 3 GPU !?!??!?!?!
Crosfire is disabled on all 3 cards.
Now what is wrong again?

HI Relclag,

Due to a poorly documented Windows limit, I have NEVER been able to get more than FIVE GPU's running, no matter what I do.

In order to get five running involves a bit of fiddling switching things on and off, with multiple reboots. Once everything is going the system will remember the settings.

Oddly enough, in order to get my four ATI GPU's working properly, I had to link and enable crossfire. My five Nvidia GPUS, on the other hand, are not linked, but SLI is enabled to have them working properly. Both setups are on i7 machines. running Boinc 6.10.18.

---

HI Relclag,

Due to a poorly documented Windows limit, I have NEVER been able to get more than FIVE GPU's running, no matter what I do.

In order to get five running involves a bit of fiddling switching things on and off, with multiple reboots. Once everything is going the system will remember the settings.

Oddly enough, in order to get my four ATI GPU's working properly, I had to link and enable crossfire. My five Nvidia GPUS, on the other hand, are not linked, but SLI is enabled to have them working properly. Both setups are on i7 machines. running Boinc 6.10.18.

Thanks David, for your important information!
For so much tinkering, I've no time ;-)
I take one out, and give an Nvidia into the PC, to make another project.

Nice greetings from Austria
RELCLAG