mersenneforum.org Titan's Best Choice
 Register FAQ Search Today's Posts Mark Forums Read

 2016-03-25, 10:53 #1 Brain     Dec 2009 Peine, Germany 331 Posts Titan's Best Choice I'd like to exchange experiences with the (classic) Nvidia GTX Titan (GK110): Hardware settings only. I'm trying to get an overview of the best stable setting for running CUDALucas, mainly for the memory clock. When I started running my first Titan there were incorrect residues with 3000 MHz mem clock. Now I'm running on 2600 MHz and it's fine. DP enabled, of course. I like to step up the clock a bit more but I'd like to benefit from your experiences: What is your stable Titan (X/Z) setting? Mine: Code: 2x Classic Nvidia GTX Titan (GK110) Memory Clock: 2600 MHz Power Target: 100% --> levels out to Chip Clock D0: 823 MHz @87°C (air) Chip Clock D1: 836 MHz @84°C (air) If we also want to share sw settings like "Threads" that's okay for me. I haven't noticed any big performance differences when running CL 2.05.1 on Windows 10 64bit. Happy Easter, Sebastian
 2016-03-25, 15:13 #2 airsquirrels     "David" Jul 2015 Ohio 11×47 Posts I've always been really surprised when people here post about needing to make all these down clocking mods to get stability on their Titans. Mine are all running stock clocks I have: 5 Titans on water (~40-50*C) - very stable always matching residues 3 Titan blacks on water - (50* C) - very stable always matching residues 2 Titan Zs (EVGA SC) on air (75-85* temps) - one mismatch out of hundreds of tests, running both cores with separate instances of CUDALucas. I had three more normal Titans on Air that ran without issue and produced matching residues, however two of them exploded the MOSFETs or inductors on the power phase that supplies the memory, so no more regular Titans on Air doing LL. They seem to be fine on TF work as it thrashes a lot less memory.
 2016-03-25, 17:50 #3 ATH Einyen     Dec 2003 Denmark 1100111100012 Posts My Titan Black is overclocked by the manufacturer: http://www.evga.com/Products/Specs/G...4-911e445a6b9f GPU Base clock: 967 Mhz Boost: 1072 Mhz vs Base: 889Mhz Boost: 980 Mhz for the base Titan Black settings: http://www.geforce.com/hardware/desk...specifications I'm not sure about the memory clock, the EVGA link says "7000Mhz Effective" which sounds like a load of .... GPU-Z says the memory is only running 1750Mhz. It is very stable not a bad test yet and I have done a lot of double checking with it, but I have not tried to overclock the memory. It is strange that the geforce.com link does not mention the base memory clock it just says 7.0 Gbps which is a transfer rate and not a clock.
2016-05-31, 15:02   #5
tServo

"Marv"
May 2009
near the Tannhäuser Gate

11×71 Posts

Quote:
 Originally Posted by Brain I'd like to exchange experiences with the (classic) Nvidia GTX Titan (GK110): Hardware settings only.
IMHO, the key to running these boards if using air-cooling is the back plate that cools the memory chips on the back of the board. If you are using water cooling, you may be cooling the board so much that the back plate is not needed. However, I noticed that Evga's Titan Hydro copper boards ( which were factory over clocked ) not only were water cooled but also had a backplate.
I run some Fermi boards in summer and I can do that without destroying the boards by manually cranking the fans between 70 -> 95 percent ( they are in the basement so noise is not an issue ). With my Titans, I have noticed that cranking the fans allows the boards to run faster because they are not performance capped by heat issues.
Just my 2 cents. YMMV.

2016-05-31, 15:43   #6
diep

Sep 2006
The Netherlands

11000011102 Posts

Quote:
 Originally Posted by tServo IMHO, the key to running these boards if using air-cooling is the back plate that cools the memory chips on the back of the board. If you are using water cooling, you may be cooling the board so much that the back plate is not needed. However, I noticed that Evga's Titan Hydro copper boards ( which were factory over clocked ) not only were water cooled but also had a backplate. I run some Fermi boards in summer and I can do that without destroying the boards by manually cranking the fans between 70 -> 95 percent ( they are in the basement so noise is not an issue ). With my Titans, I have noticed that cranking the fans allows the boards to run faster because they are not performance capped by heat issues. Just my 2 cents. YMMV.
In factory where they produce the GPU's, that probably is TSMC, the gpu's produced in the middle in general clock way higher without problems than the gpu's at the outside.

So if you are lucky and have 1 from the center then that will have less heat issues.

Also some months later on usually there is process improvements - mostly software related. Typically newer batches of tesla's for example eat less power.

There can be many reasons therefore why one or another suffers more or less from heat related issues - yet one thing is for sure - running mfaktc hits nearby the worst case path of those GPU's - maxing out its power usage and therefore heat.

If you would however manage to cool it to nearby 20C, then power usage of the GPU will drop another 10% or so compared to what it burns when it is 50C.

So real good watercooling might be worth investigating IMHO

Last fiddled with by diep on 2016-05-31 at 15:44

 2016-05-31, 22:18 #7 henryzz Just call me Henry     "David" Sep 2007 Liverpool (GMT/BST) 25·11·17 Posts With water cooling is the cooling based upon room temperature or is there some sort of refrigeration going on?
 2016-06-01, 06:05 #8 LaurV Romulan Interpreter     "name field" Jun 2011 Thailand 2×17×293 Posts Subscribing to what tServo says. Also, you may need some small fans around in the box. The aeration for cases designed for air cooling is not good for water cooling. When I switched to water cooling, in the beginning, I had heat issues with other components in the box. I mean, I had everything on water, mobo, memories, cpu, gpu, so the air in the box wasn't moving much. When you have a huge air-cooler for the CPU, that is "accidentally" cooling many other small things in your box, because it is moving a lot of air around. Things which don't really "need" cooling, like small ICs or mosfets there, so they don't have heatsinks, but when you move to water, they start getting hot. First trick was to turn the case with the back side, up. Hot air going up between the cards, and getting out on top. Then installing small/quiet fans in the case here and there... Also, be careful with aluminum parts, they have a lot of galvanic effects with other things normally used in a cooling installation, which are usually copper, nickel, steel, etc. Don't mix metals, if you go aluminum, then use everything from it! (radiators including).
2016-06-01, 13:06   #9
tServo

"Marv"
May 2009
near the Tannhäuser Gate

78110 Posts

Quote:
 Originally Posted by henryzz With water cooling is the cooling based upon room temperature or is there some sort of refrigeration going on?
99% of the time, it's room temperature. I have occasionally seen some phase change cooling devices but the companies rarely last. Too expensive to purchase and run.

I believe Asetek has offered some tec ( peltier ) assisted devices in the past. They were placed between the cpu or gnu and the heat-exchange plate of the water cooler.

2016-06-01, 22:54   #10
TObject

Feb 2012

1100101012 Posts

Quote:
 Originally Posted by henryzz With water cooling is the cooling based upon room temperature or is there some sort of refrigeration going on?
Once you go below room temperature you have to take into account condensation that enters the picture.

2016-06-02, 00:05   #11
chalsall
If I May

"Chris Halsall"
Sep 2002

101000110101102 Posts

Quote:
 Originally Posted by TObject Once you go below room temperature you have to take into account condensation that enters the picture.
Or, more specifically, the dew point.

Here in Bimshire our rum and ice and coconut water drinks sweat water as condensation.

So do our servers, if not properly managed....

 Similar Threads Thread Thread Starter Forum Replies Last Post ATH Hardware 15 2017-05-27 22:38 ixfd64 GPU Computing 10 2017-05-17 15:19 ixfd64 GPU Computing 20 2015-04-28 00:27 ATH GPU Computing 295 2013-05-12 21:35 Manpowre GPU Computing 27 2013-05-12 10:00

All times are UTC. The time now is 12:12.

Tue May 24 12:12:46 UTC 2022 up 40 days, 10:14, 0 users, load averages: 1.53, 1.84, 1.79