Reconfiguring for winter.

[VirtualLarry]

I currently have two operational ATX rigs. They have H81M mobos, G3258 CPUs, and each one has a 7950.

I have two of these boards NIB:

http://www.newegg.com/Product/Product.aspx?Item=N82E16813130779&ignorebbr=1

These Z97 boards have two PCI-E x16 slots (second one x4 electrically).

Was planning on moving my two 7950s into one rig, and I just ordered today, a pair of XFX 270X cards.

So that would give me two Z97 rigs, each with two PCI-E x16 slots. One with 7950s, and one with 270Xs.

I also recently picked up two Antec Edge (SeaSonic) 750W Gold PSUs (with LED fan ). So those should handle the need for dual dual 6-pin PCI-E power cables (4x).

At that point, the only thing that I would need to upgrade, or bypass, is my 550VA / 330W UPSes on each machine. With two GPUs in each boxen, they will draw too much under full GPU load.

Unfortunately, this leaves open the question of being able to run any CPU projects, if I'm running two GPU projects on my poor little G3258 dual-cores. Can you just leave one CPU core open, to feed two GPUs, or does it need two cores, one for each?

Also, anyone want to suggest projects to run? MW@H runs great on the R9 280/X/7950/7970 cards, because they are strong in DP. I don't know what the 270X is good at yet.

Edit: And in my other room, I've got my Thuban rig, a 1045T with 16GB of DDR3, and three PCI-E slots (though I may not be able to really use the last slot). I also have a pair of R7 260X cards that I can put in there for crunching.

Edit: Wasn't intended to be a brag thread, thought that it might be interesting to someone. I know most of you have much bigger fleets than mine.

 

[kowalabearhugs]

Very nice crunching setup.

Can you just leave one CPU core open, to feed two GPUs, or does it need two cores, one for each?

It depends on the project. You might be able to get away with that on the 270xs, but I'd imagine the 7950s would require a full core.

Primegrid GFN World Record and Short require DP and run very well on the 79xx. I run 1 WU at a time on my 7970 and it's stays at 90-99% utilization. Those cards are much more efficient at GFN work than PPS Sieve.

Einstein is another that requires DP. The contents of my config file are as follows and, again, utilization % stays in the 90s. If anything you could change the BRP4G and BRP5 to 0.33 instead of running 4 tasks simultaneously.

<app_config>
<app>
<name>einsteinbinary_BRP4</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>
<app>
<name>einsteinbinary_BRP4G</name>
<gpu_versions>
<gpu_usage>0.25</gpu_usage>
<cpu_usage>0.25</cpu_usage>
</gpu_versions>
</app>
<app>
<name>einsteinbinary_BRP5</name>
<gpu_versions>
<gpu_usage>0.25</gpu_usage>
<cpu_usage>0.25</cpu_usage>
</gpu_versions>
</app>
</app_config>

The POEM app was revised last Winter/Spring(?) and is now tuned for better performance on GCN 1.0+(7xxx and up) and it doesnt require DP. I haven't tinkered as much with that application, but it does produce a high PPD.

Collatz also recently released a new app that is much more efficient. Here is info for configuring : http://boinc.thesonntags.com/collatz/forum_thread.php?id=1273

I've been using this file on my 7970, but there is still a bit of wiggle room for faster processing.

verbose=1
threads=7
kernels_per_reduction=48
sieve_size=30
lut_size=16
reduce_cpu=0
sleep=1

 

[VirtualLarry]

Thanks! I haven't done the switch-over yet, going to do that in the next two weeks. Currently test-crunching on my Q9300 again, with a R7 250X. (Also tested it with Skyrim.)

MW@H WUs on the 250X show estimated time as 4:30. On my 7950, they were like 0:37. BIG difference!

 

[wiseguyin]

This is one of the things i am building my first rig for. Once I have it all up and running (x99 based), all i gotta do is just wait for my nobel for finding the cure for cancer. Or alien life. Or something