2015/03/27 16:00:25
hallowen
Got some Fairly good scores for a "Vanilla" NVIDIA Titan X using my "GPU ONLY" water cooling block.
 



 
 
2015/04/07 13:43:06
hallowen
Although this Titan X has been working pretty well testing/benching, It's Not enough for me to want to spend the extra cash on three more ATM (even if I could find more available) not to mention the loss I'm bound to incur when I get ready to sell just this one! LOL!
 
Think I'll wait and see what else is offered up later on before I decide whether I'll be going back to a 4-Way SLI setup again.
2015/04/07 13:49:06
EVGATech_DaveB
Hmm, maybe in the interim, you can put it on phase and get it colder?  I would love to see one at subzero really being pushed!
2015/04/07 14:59:49
hallowen
From the GPU Phase Change tests I did using my EVGA GTX 480 in a few posts back, I found that even though the initial temps on the GPU at idle on the desktop was -14C, Upon starting the Graphics tests on Fire Strike the Temps would immediately climb to around +22C and would not hold sub-zero temps any better than Chilled water cooling, probably due to the higher heat output (250W plus) that the GPU put out compared to the lower heat output (140W plus) of the CPU under Phase Change.
 
Alas, As I expected, It becomes all too obvious why DICE/LN2 Cooling is the only way to "Maintain" very low temp sub-zero cooling under High Load Bench Testing including all the preparation needed to the components, But even then if you do use this type of cooling, you will STILL also have to do the "Major Hardware Mods" if you wish to come close to what the Highest GPU OC's are reaching.
 
I have been losing enough $$$$ as it is every time I sell components to update my Rigs, But at Least I am not Voiding my Warranties by physically altering the Graphics Cards to the point where I can't Sell them, So it's still CPU=Phase Change and GPU=Chilled Water for my builds if I want to get SOME monetary returns on my investments. 
 
 
 
 
2015/04/07 17:10:48
EVGATech_DaveB
Hmm, have you tried cascade, or building one, as it sounded like you had enough SS's to be able to do it, that may well outperform your chiller.  It just sounds like an interesting idea.
 
Isn't the ss temps you got largely due to the wattage of the SS you had on it versus the wattage of the GPU?  In that case wouldn't a SS of more power be able to net a better temp?  Just a thought.... but maybe you already hit a wall with power and whatnot as well, as I know you are chewing up a TON of power already.
 
I don't know just thinking out loud here, but those types of things were always interesting and love to see stuff built and tested.  Like the other half has said "with your love of building things that have no practical application, you missed your true calling as an engineer"..... who knows, she may be right.
2015/04/08 14:20:33
hallowen
As you may have seen in previous posts over the years, I have built a few LARGE Cascade Phase Change Units (for CPU's only) that did manage to get below -100C, But under Heavy Overclocked Loads still had a hard time maintaining half of that temp, and that was years ago with Intel Single-Core Socket 478 Processors!
 
It just comes down to looking at the practical end of building such an enormous Cooling Device and asking yourself "Can I operate this Cooling System 24/7 and is This Really going to be Worth all the expense/trouble?"
 
Back in the days, I've seen a few Multiple-Cascade Phase Change Units reach around -150C, But they could never maintain temps of -150C to -170C (like LN2 does by just keep adding more under heavy benching loads to maintain the desired temp), Not to mention the expense/size and power requirements it takes to do so.
 
  
 
2015/04/08 19:05:19
awalleyeguy
 But a 480 must put out a lot more heat than a Titan X?
2015/04/09 06:33:53
hallowen
awalleyeguy
 But a 480 must put out a lot more heat than a Titan X?


 
True, From what I've seen a GTX 480 uses around 450W compared to a Titan X which is supposedly rated at 250W, But my ASETEK Vapochill Phase Change I used in my tests is only rated for a 180W Load and can't maintain sub-zero temps over that rating.
 
I will agree that the Vapochill would fair better with the lower heat rating of a Titan X and that would result in some additional Overclocking overhead, But it still would require a higher heat load Phase Change to maintain temps.
By contrast, Even my Custom 300W load rated Phase Change only adequately Cools my 140W rated 5960X CPU running at 4.8GHz and temps DO rise to the upper +20C during Firestrike CPU Tests when the CPU is under Full Load.
 
So Yes, It would allow higher Overclocks on a Titan X, But still could not reach/maintain the Extreme sub-zero temps of LN2 or even DICE for that matter that if you so desired would be required to reach and HOLD the 2000+ MHz that has been achieved by K|ngp|n and a few others.
 
Since I've always wanted my Systems to be able to not only run Benches, But also Games, Web surfing, etc...24/7 without the constant "maintenance" of watching/adding/adjusting sub-zero cooling liquid temps/levels, I have been using only Phase Change and Chilled Water Cooling Solutions on my Rigs that I can leave almost unattended.
 
 
2015/04/10 17:58:27
hallowen
OK Guys, I went ahead and put my ASETEK Vapochill on my Titan X just to show how the temps change during a Fire Strike Bench using Phase Change Cooling.
Notice that the GPU temp before the Bench is -26C on Precision X, But Just as I expected, At the end of the Graphics Tests, the Temps were running between +3C to +9C after each Test.
 
While I did increase my Score on this First Test with the GPU on Phase Change to a New High, I still need to Test some more tomorrow to determine how much higher I can push the GPU using the Stock Vbios and only 0.8930 Volts further than 1502MHz GPU/2005MHz Memory
 

 
 
2015/04/11 12:59:13
hallowen
Got a higher score (#20 in Fire Strike HOF) by enabling K-Boost in Precision X, But Notice that the GPU Temp before the bench is now only -11C due to using a higher GPU voltage of 1.230v at 1514MHz GPU/ 2005MHz Memory which is starting to reduce the effective cooling under bench loads while running Fire Strike graphic tests.
 
Possibly could to change to a modded Vbios for a higher Power Target/additional voltage and will continue testing with the stock Vbios for further improvements.
 
As I have stated before and these results show, If you want to keep up in the Top 10 in Benching whether using one Card or Four in SLI, You'll need LN2 Cooling to accomplish those Ultra High GPU and/or CPU Overclocks.
 
 

Use My Existing Forum Account

Use My Social Media Account