tesla's

bdub · Post by **bdub** » Wed Mar 26, 2014 7:59 am

so I have a project using tyan 7055 mobo, with two 2670v2 (ivy bridge ep 10 core cpu's), and two tesla k20c accelerators (or whatever they're called). the latest version of [email protected] client (7.4.4, I think) does recognize the cpus and the tesla cards... but it seems to be running only 30 of the 40 available hyperthreaded cores, and the tesla cards don't seem to be doing much. I figured they'd pound thru work units like butter.

guess I need to let it sit for a day and see what estimates and completion numbers are... I have this thing thru the weekend at least, so you'll see a major jump in my output I think.

cheers!

Post by **evasive** » Wed Mar 26, 2014 8:10 am

it looks like you'll need to force the client to each of them with commandline options. Does this help?
https://foldingforum.org/viewtopic.php?f=52&t=5519

bdub · Post by **bdub** » Wed Mar 26, 2014 9:56 am

thanks evasive, well, the new client seems to be using the tesla's. i don't think any command line is needed, except maybe for some optimization. they aren't outputting as much as the cpu's as far as percentage of project done... but I haven't checked too closely on how big each project is.
everything in the box, which includes 10 hdd's, an ssd, an areca raid card, and the aforementioned components use up 600 - 630 watts according to the meter they are working thru, when folding... I figured it would be more to 900w with those teslas in there. we have dual 1200w psu's, so no problems with power.
I'm wondering if there's some extra driver I need. I got what NVidia has for tesla and winserv08r2, but maybe some PhysX or other weird driver needed?

it's all gotta be working okay now, as I def feel warm air coming out the back of the tesla's. they also report as producing an estimate of around 25000 points a day right now. the cpu says it's doing only 7k. the cpu estimate sounds off to me though, should be higher than that.

Post by **Karlsweldt** » Wed Mar 26, 2014 11:29 am

Are you using the preferred screen saver? That uses the least amount of rendering data, and provides up front progress plus the WU type.
I use it on my system, and have set the 'time out' to idle as 2 minutes.
Only when the video is at "idle", does the GPU do its mission.
When my GPU gets going, a lot of warm air does exit the GPU exhaust. But not excessive.

My [email protected] setup: Avg. 4500 PPD.
Mobo: Intel D975XBX2KR I975X LGA775.
CPU: Intel Core 2 Duo 2.4 GHz, (SL9S8) 1066 FSB speed.
CPU cooler: Dynatron K987.
Memory: G Skill 2 GB, GSK F2-8500CL5D-2GBPK 1066 Timing 5-5-5-15.
Video: Nvidia GTX280 PCI-E x 16, 1 GB. [240 CUDA cores!]
PSU: Antec 650 watt (EA-650).
Hard drive: Seagate ES ST500NM0011 500GB 7200 RPM SATA.
OS: Windows 7 Home Premium.

bdub · Post by **bdub** » Wed Mar 26, 2014 12:18 pm

I never use screensavers. if the monitor isn't on, it's black.... but with this new client, I see that the web interface actually says it's "sleeping".
in any case, it looks to me that the teslas are a success with this client. I'm pretty sure they each will output 25k per day, but i'll report back tomorrow to see what actual output was.
the cpu's are another matter. I can't get all 40 virtual cores to run full speed. the client seems to be holding it to 30 cores. I'm sure there's a work around somewhere, but I'm just gonna let it run as is thru the weekend.

what I need is a temp monitoring program for the teslas, and have it also report workload. haven't found anything like that. gpu-z, cpu-z, openhardwaremonitor, hwmonitor, all no joy.

https://fbcdn-sphotos-f-a.akamaihd.net/ ... 7094_n.jpg

Post by **Karlsweldt** » Thu Mar 27, 2014 6:56 am

The least video 'overhead' is with a blank screen. Should be a choice for that. If the video feed only "turns off" after a set time, the GPU may still be rendering video needlessly. Only in an 'idle' state, does the GPU do its folding.
As to [email protected] monitors, some are worth having.. others are botch jobs.
You may want to try this one from Softpedia.
http://fahtracker.com/ and http://fahmon.net/ also offer a monitor program.
Couldn't find an active link for the older "Electron Microscope" [email protected] monitor.

bdub · Post by **bdub** » Fri Mar 28, 2014 6:02 am

so after running for a while, it looks like the one of the tesla's says it is outputting 38k points a day, the other one 25k, and the cpu is around 7k. odd the difference between the two teslas.
the teslas are not hardware bridged, as I do not know if sli is recommended for these things, and they are also in slots that connect to a different cpu bus. I don't know if they should be on the same cpu bus or not. seems it would be more advantageous to be separated, not only each using their own cpu resource, but by being able to have some space physically between them for cooling intake.
I have several other machines folding currently, but this chart will maybe give a slight glimpse into output over the next few days. this machine with the teslas in it does seem to be taking on bigger projects by default I think...
http://folding.extremeoverclocking.com/ ... s=&u=42254

and I'm not looking for [email protected] monitors... I'm looking for a TESLA hardware monitor. something that shows the tesla gpu temp and current workload percentage in a graph or numerically, such like hw monitor, aida64, speedfan, etc does. none of the ones I've tried report on the tesla's any more than their model and physical chip layout.
the motherboard uses an onboard chip for video to a monitor, aspeed, which does not really take any resouces away from anything, I don't think. maybe a small bit of ram, which this board has in spades (32gigs).

Post by **evasive** » Fri Mar 28, 2014 6:18 am

C:\Program Files\NVIDIA Corporation\NVSMI

nvidia-smi.exe

info on usage for getting info:
http://choorucode.com/2011/08/12/cuda-o ... e-desktop/

bdub · Post by **bdub** » Fri Mar 28, 2014 7:19 am

hmmm, I see that. I'm wondering if I can do something ill to these k20c's, as I'm looking through this webpage for something that supports k-series teslas...

http://www.nvidia.com/object/software-f ... ducts.html

then I found this...
http://www.migenius.com/products/nvidia ... benchmarks
looks promising for k20c to go into tcc mode, yeah?

AWE-YEAH! very good, thanks evasive. I can tell using the utility in there how much wattage the card is using, what is the temp of each card. I can also thell that the driver seems to already be in "tcc" mode... I think.
sweet, thank you.
I just shut off one of the tesla folding instances and it shows a change when I run the tool a second time.

bdub · Post by **bdub** » Fri Mar 28, 2014 7:55 am

even at 90-99% load, they only use half of their rated wattage envelope. about 115w out of 225w. they sit at about 55-60 centigrade at full load, and the gpu fans are running about half speed.
when running folding and crystalmark on the raid array, I top the system out at 660 watts. still way below the 1200 watt psu's rating.