Overclock.net › Forums › Cooling › Other Cooling Discussions › Why TIM isn't as important as you think it is.
New Posts  All Forums:Forum Nav:

Why TIM isn't as important as you think it is. - Page 3

post #21 of 27
Let me start by stating that I'm not a benchmarking guru at all but it strikes me that the methodology of TIM comparison is all wrong.

All the benchmarks seem to quote idle and max loaded temps.... while I can see some value in quoting the idle temps I think its quite possible that the loaded temps are of little or no value when comparing TIM performance.

The problem being at max load it may well be that the cooling solution rather than the TIM is the cooling limiting step or bottleneck which ever you description you prefer.....in which case the choice of TIM would have no relevance to the temperature provided it was able to pass sufficient heat to saturate the cooling solution.

It has frequently struck me that there seems to be little difference in the max loaded temperature when comparing TIM's with as widely varied thermal conductivities as 4 w/m/c to 80 w/m/c.

This could well be an explanation of this.

I believe that to compare TIM's a lesser heat load should be used...say 50%....one that would not saturate the cooling solution and so would be a comparison of the heat transfer to the cooling solution via the TIM rather than testing the cooling solution itself.
Edited by technogiant - 9/2/13 at 1:35am
post #22 of 27
Thread Starter 
Quote:
Originally Posted by technogiant View Post

Let me start by stating that I'm not a benchmarking guru at all but it strikes me that the methodology of TIM comparison is all wrong.

All the benchmarks seem to quote idle and max loaded temps.... while I can see some value in quoting the idle temps I think its quite possible that the loaded temps are of little or no value when comparing TIM performance.

The problem being at max load it may well be that the cooling solution rather than the TIM is the cooling limiting step or bottleneck which ever you description you prefer.....in which case the choice of TIM would have no relevance to the temperature provided it was able to pass sufficient heat to saturate the cooling solution.

It has frequently struck me that there seems to be little difference in the max loaded temperature when comparing TIM's with as widely varied thermal conductivities as 4 w/m/c to 80 w/m/c.

This could well be an explanation of this.

I believe that to compare TIM's a lesser heat load should be used...say 50%....one that would not saturate the cooling solution and so would be a comparison of the heat transfer to the cooling solution via the TIM rather than testing the cooling solution itself.

I understand your argument, but if you read the document again, you will see that there was no mention of idle temperatures, or etc. . Because I was aiming for scientific accuracy, I was limited to one variable - the choice of TIM - and one dependent - the performance of the processor. For the purposes of accuracy, the ambient temperatures were held at a constant 20°C, and the load temperatures were also held constant. Because the only time the load temperatures would be at a constant is the point of TJMax (67.7° in this case), I took advantage of Intel's thermal throttling technology to keep it there.

Granted, you may find this a test of the cooling solution, but as I already explained, all temperatures were held constant. I believed that using a more effective (temperature-wise) TIM would increase the "overhead" performance, so to speak, allowing the proccessor to have more computed before hitting a thermal throttling cycle.
Project - ARC
(90 photos)
Project - ARC
(16 items)
 
 
CPUMotherboardGraphicsRAM
Intel Core i7 3930K RAMPAGE IV GENE AMD Radeon HD 7900 Series (HD7970-DC2T-3GD5) G.Skill 
RAMRAMRAMHard Drive
G. Skill G. Skill G. Skill Vertex 4 
CoolingOSMonitorMonitor
Custom Liquid Windows 8 ASUS VS278Q-P Black ASUS VS278Q-P Black 
MonitorMonitorPowerCase
ASUS VS278Q-P Black LG Flatron IPS234W (secondary monitor) Seasonic Platinum 1000w Fractal Design Arc Midi 
  hide details  
Reply
Project - ARC
(90 photos)
Project - ARC
(16 items)
 
 
CPUMotherboardGraphicsRAM
Intel Core i7 3930K RAMPAGE IV GENE AMD Radeon HD 7900 Series (HD7970-DC2T-3GD5) G.Skill 
RAMRAMRAMHard Drive
G. Skill G. Skill G. Skill Vertex 4 
CoolingOSMonitorMonitor
Custom Liquid Windows 8 ASUS VS278Q-P Black ASUS VS278Q-P Black 
MonitorMonitorPowerCase
ASUS VS278Q-P Black LG Flatron IPS234W (secondary monitor) Seasonic Platinum 1000w Fractal Design Arc Midi 
  hide details  
Reply
post #23 of 27
Quote:
Originally Posted by barracks510 View Post

I understand your argument, but if you read the document again, you will see that there was no mention of idle temperatures, or etc. . Because I was aiming for scientific accuracy, I was limited to one variable - the choice of TIM - and one dependent - the performance of the processor. For the purposes of accuracy, the ambient temperatures were held at a constant 20°C, and the load temperatures were also held constant. Because the only time the load temperatures would be at a constant is the point of TJMax (67.7° in this case), I took advantage of Intel's thermal throttling technology to keep it there.

Granted, you may find this a test of the cooling solution, but as I already explained, all temperatures were held constant. I believed that using a more effective (temperature-wise) TIM would increase the "overhead" performance, so to speak, allowing the proccessor to have more computed before hitting a thermal throttling cycle.

Hi Barracks, my comments weren't directed at your research but more a general comment and this seemed a relevant thread to air those views.....but having said that I've re read your pdf more thoroughly.....I'm not 100% sure I've fully understood but if the performance results you have published are the resultant performance following thermal throttling of the cpu then they are indeed related to temperature albeit in a slightly more convoluted manner.......so in essence I think my previous post would apply to your results also....if the thermal capabilities of the heat sink are not large in comparison to the heat transfer capabilities of the TIM then the differences in the TIM may be masked.

It's possible that by artificially lowering the heat load by imposing the 55c throttling limit that that you have started to see the differences in the Tim performance and possibly why the ceramic was starting to push ahead.....but even at this lowered heat level I think the thermal performance of the heat sink assembly may well be affecting results.....I think for reliable TIM testing a very powerful cooling solution would be required so that it's effects on results would be negligible.

If I may be so bold as to offer critique I believe one aspect of your testing that may have introduced error was the use of alcohol which I presume was ethanol as the carrier for the various conductive particles.

Firstly this may well be evaporating, phase changing if you will, and so altering temperatures.

Secondly this effect may not be equal in all three particle suspensions, ceramic and aluminium (actually the outer coating of your aluminium particles will be invariably aluminium oxide due to it's reactivity with atmospheric oxygen) being very polar or highly charged in their chemical nature would tend to hold the alcohol more strongly due to hydrogen bonding with the alcohol where as diamond being non polar would tend to release it more readily.

Thirdly for the very same reasons heat exchange within the TIM between the particles and carrier alcohol would be favored in the case of ceramic and aluminium because of their stronger interaction with the alcohol while the lesser interaction of the alcohol with the diamond would hinder internal heat transfer in that case.

All interesting research though and a definite step in the right direction for a better means of testing TIM in my opinion, simply restricting the heat load seems to have shown some differentiation in the results......now what about trying it with an uber powerful cooling method?
post #24 of 27
Thread Starter 
Quote:
Originally Posted by technogiant View Post


Hi Barracks, my comments weren't directed at your research but more a general comment and this seemed a relevant thread to air those views.....but having said that I've re read your pdf more thoroughly.....I'm not 100% sure I've fully understood but if the performance results you have published are the resultant performance following thermal throttling of the cpu then they are indeed related to temperature albeit in a slightly more convoluted manner.......so in essence I think my previous post would apply to your results also....if the thermal capabilities of the heat sink are not large in comparison to the heat transfer capabilities of the TIM then the differences in the TIM may be masked.

It's possible that by artificially lowering the heat load by imposing the 55c throttling limit that that you have started to see the differences in the Tim performance and possibly why the ceramic was starting to push ahead.....but even at this lowered heat level I think the thermal performance of the heat sink assembly may well be affecting results.....I think for reliable TIM testing a very powerful cooling solution would be required so that it's effects on results would be negligible.

If I may be so bold as to offer critique I believe one aspect of your testing that may have introduced error was the use of alcohol which I presume was ethanol as the carrier for the various conductive particles.

Firstly this may well be evaporating, phase changing if you will, and so altering temperatures.

Secondly this effect may not be equal in all three particle suspensions, ceramic and aluminium (actually the outer coating of your aluminium particles will be invariably aluminium oxide due to it's reactivity with atmospheric oxygen) being very polar or highly charged in their chemical nature would tend to hold the alcohol more strongly due to hydrogen bonding with the alcohol where as diamond being non polar would tend to release it more readily.

Thirdly for the very same reasons heat exchange within the TIM between the particles and carrier alcohol would be favored in the case of ceramic and aluminium because of their stronger interaction with the alcohol while the lesser interaction of the alcohol with the diamond would hinder internal heat transfer in that case.

All interesting research though and a definite step in the right direction for a better means of testing TIM in my opinion, simply restricting the heat load seems to have shown some differentiation in the results......now what about trying it with an uber powerful cooling method?

 

Wow, I have not thought of the choice of carrier to be affecting the properties, but it seems like a great possibility now. I would be willing to try sub-ambient cooling, but I lack the equipment and the knowledge to maintain the integrity of the experiment. Perhaps, you could try an experiment yourself, and publish the results.

 

I really appreciate your comments as they do provide much more insight toward the results of the project.

Project - ARC
(90 photos)
Project - ARC
(16 items)
 
 
CPUMotherboardGraphicsRAM
Intel Core i7 3930K RAMPAGE IV GENE AMD Radeon HD 7900 Series (HD7970-DC2T-3GD5) G.Skill 
RAMRAMRAMHard Drive
G. Skill G. Skill G. Skill Vertex 4 
CoolingOSMonitorMonitor
Custom Liquid Windows 8 ASUS VS278Q-P Black ASUS VS278Q-P Black 
MonitorMonitorPowerCase
ASUS VS278Q-P Black LG Flatron IPS234W (secondary monitor) Seasonic Platinum 1000w Fractal Design Arc Midi 
  hide details  
Reply
Project - ARC
(90 photos)
Project - ARC
(16 items)
 
 
CPUMotherboardGraphicsRAM
Intel Core i7 3930K RAMPAGE IV GENE AMD Radeon HD 7900 Series (HD7970-DC2T-3GD5) G.Skill 
RAMRAMRAMHard Drive
G. Skill G. Skill G. Skill Vertex 4 
CoolingOSMonitorMonitor
Custom Liquid Windows 8 ASUS VS278Q-P Black ASUS VS278Q-P Black 
MonitorMonitorPowerCase
ASUS VS278Q-P Black LG Flatron IPS234W (secondary monitor) Seasonic Platinum 1000w Fractal Design Arc Midi 
  hide details  
Reply
post #25 of 27
Well I'm currently making a subambient build which is in the project phase atm.....while it would be quite difficult for me to do such extensive testing as that would require repeatedly opening the chamber which would need to be resealed on each occasion it's something I'll look towards trying if I'm able to.
post #26 of 27
Quote:
Originally Posted by barracks510 View Post

Wow, I have not thought of the choice of carrier to be affecting the properties, but it seems like a great possibility now. I would be willing to try sub-ambient cooling, but I lack the equipment and the knowledge to maintain the integrity of the experiment. Perhaps, you could try an experiment yourself, and publish the results.

I really appreciate your comments as they do provide much more insight toward the results of the project.


You could always get a basic direct die phase change cooler. You can buy one directly off FrozenCPU, or you can have one made by one of the various artisans on OCN.

You could get a half inch thick copper block, and mill the sides down like a pyramid to ensure equal heat transfer, as well as make mounting it easier. This would be the simplest way to test at sub-ambient/subzero temps. Since the heatsink would be copper on copper, and be far below ambient, it wouldn't favor any material and wouldn't be dependent on the cooler.

Just my 2cents.gif on how I would approach sub-ambient testing of TIMs in a way that allows easy mounting/dismounting/reseating etc.
Edited by ZytheEKS - 9/2/13 at 9:57pm
The Laboratory
(11 items)
 
  
CPUMotherboardGraphicsRAM
AMD FX-8350 Vishera 8-Core 4.0GHz Asus Crosshair V Formula-Z Radeon R9 290X Generic 1333MHz R.A.M. 
Hard DriveOptical DriveCoolingOS
Wester Digital 5400RPM HP CH20L "The Russian Winter" Windows 8 
KeyboardPowerCase
Steel Series Merc Fatal1ty 1000Watt PSU DIY Rig 
  hide details  
Reply
The Laboratory
(11 items)
 
  
CPUMotherboardGraphicsRAM
AMD FX-8350 Vishera 8-Core 4.0GHz Asus Crosshair V Formula-Z Radeon R9 290X Generic 1333MHz R.A.M. 
Hard DriveOptical DriveCoolingOS
Wester Digital 5400RPM HP CH20L "The Russian Winter" Windows 8 
KeyboardPowerCase
Steel Series Merc Fatal1ty 1000Watt PSU DIY Rig 
  hide details  
Reply
post #27 of 27
Back to the paper, it's an interesting read.
For the conclusion, have you taken the type heatsink used into consideration?
For example, it's been a while, but when I was doing TIM research I read that metal pastes worked better with certain kinds of heatsinks than others due to conductivity and other factors.
Since your TIMs were only applied to one heatsink, and testing was done on an open bench rather than in a case, you might find significantly different results under other conventional circumstances.
Personally, it also woud've been nice to see more contemporary hardware since CPU architecture and heat disbursal would've been quite different with an Ivy Bridge CPU, for example (since it uses paste rather than solder).

All things considered though, it sounds like you put a lot of work into this.
Thanks for sharing.
Code: 777
(17 items)
 
  
CPUMotherboardGraphicsRAM
3960X Rampage IV Extreme GTX Titan G.Skill Ripjaw Z Series 64GB (8x8) DDR3 2,400MZ  
Hard DriveHard DriveHard DriveCooling
Samsung 840 Pro Samsung 830 WD VelociRaptor x4 RAID 10 H80i 
OSMonitorKeyboardPower
Windows 8 Pro (Media Center Edition) BenQ XL2420T x3 Max Nighthawk X8 Rosewill Lightning 1300 
CaseMouseMouse PadAudio
COSMOS II Logitech Performance MX Razer Vespula Logitech X540 
Audio
ASUS Xonar DSX 
  hide details  
Reply
Code: 777
(17 items)
 
  
CPUMotherboardGraphicsRAM
3960X Rampage IV Extreme GTX Titan G.Skill Ripjaw Z Series 64GB (8x8) DDR3 2,400MZ  
Hard DriveHard DriveHard DriveCooling
Samsung 840 Pro Samsung 830 WD VelociRaptor x4 RAID 10 H80i 
OSMonitorKeyboardPower
Windows 8 Pro (Media Center Edition) BenQ XL2420T x3 Max Nighthawk X8 Rosewill Lightning 1300 
CaseMouseMouse PadAudio
COSMOS II Logitech Performance MX Razer Vespula Logitech X540 
Audio
ASUS Xonar DSX 
  hide details  
Reply
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Other Cooling Discussions
Overclock.net › Forums › Cooling › Other Cooling Discussions › Why TIM isn't as important as you think it is.