## Miners dogpiling on one device

1

I've built a mining rig running Ubuntu 11.10 with 3 Radeon 6850 GPUs.

When I run ./poclbm.py without any arguments, it returns the list of available devices as...

[0] Barts
[1] Barts
[2] Barts
[3] AMD Sempron(tm) 145 Processor


giving me the (possibly erroneous) idea that I seem to have the drivers setup correctly.

When I start up one miner and pass the argument to use device 0, the hash rate is reported at slightly over 200 Mhash/sec. However, when I start up another pointed at device 1 in a separate term window, the hash rate for both is reported at around 100 Mhash/sec. And when I fire up the third miner and point it at device 2, then all three miners report their rate at around 60-70 Mhash/sec.

This suggests to me that the miners are not honoring the device argument that I'm passing and all dogpiling on one of the devices (probably the freckle faced red-headed one).

This happens with both m0mchil's poclbm and with PhoenixMiner. Below are the arugments I'm passing to each...

./poclbm.py http://username.worker:pw@host.com:port -d 1 -v

./phoenix.py -u http://username.worker:pw@host.com:port -k phatk2 DEVICE=2 VECTORS BFI_INT


Any ideas what I might be doing, or have done, wrong?

1

You could check the temperature on the devices, the one active should be running much hotter than the others.

If they're all running hot the bottleneck is elsewhere (CPU or memory/bandwidth). If none of them are running hot you've bought one hell of a CPU to get 200Mhash.

Else it's advisable to check whether or not you can use the non-default videocard at all.

I'm afraid I lack the videocards to test this myself :)

Excellent idea. It hadn't even occurred to me that there might be a bottleneck elsewhere. However, I left only one miner running overnight, so this morning I popped the case open and checked how warm each GPU was for comparison, and one was clearly warmer. Then I fired up the other two miners and went to take a shower. Upon return, I checked the GPU's again and the conditions were the same... the one that was warm before was just as warm, and the other two were still quite cool. So, I don't think this is it. – Jon Garvin – 2012-01-12T16:32:45.233

Did you try the --device=2 syntax? – Lodewijk – 2012-01-12T17:32:32.587

Try using guiminer, https://bitcointalk.org/index.php?topic=3878.0 . And maybe just any other version of Linux (different install, different drivers, who knows)

– Lodewijk – 2012-01-12T17:39:33.637

I take back my earlier comment. After more research, I discovered the command aticonfig --odgt --adapter=all which reports the temperatures on each GPU. After firing up two miners, instructed to use the first and third GPUs, and letting those run for several minutes, that command reports that the temp on the 3 GPUs are 68c, 27c and 62c. That tells me you were right about a bottleneck elsewhere. I'm going to accept this answer so you get credit and post a new question looking for help hunting down the bottleneck. I'll post a link to it here shortly. – Jon Garvin – 2012-01-13T01:41:17.573

– Jon Garvin – 2012-01-13T02:23:04.590