|
|
Testing PC cooling using die simulators vs. real CPUs
|
Date Posted: Dec 14 2005
|
Author: pHaestus
|
Index:
|
|
Posting Type: Article
|
Category: FAQ's, Editorials, Q&A's
|
Page: 4 of 4
|
Article Rank: from 2 Readers
Must Log In to Rank This Article
|
Forum Discussion Link
|
|
|
Testing PC cooling using die simulators vs. real CPUs By: pHaestus
|
The transition to CPUs with IHS has presented some questions for the traditional copper die simulator that I’m not certain have been fully addressed. Are old test results on smaller-sized dies relevant to today’s CPUs? Does the IHS spread heat so effectively that the large low-heat density die simulators are appropriate or not? Does the physical dimensions and shape of an IHS, the extra thermal interface between the CPU die and the underside of the spreader, and the extra force necessary to make them all transfer heat to a cooler make for a situation that is just too difficult for us to reliably estimate with a copper core? What about dual core CPUs?
It seems that three options are available to users of die simulators: attempt to simulate the effects of the IHS in their test bed, ignore this change in processor altogether, or try to estimate how results from their existing test bed relate to performance on new CPUs with heat spreaders.
Ignoring the issue won’t make it go away. It’s as simple as that. Redesign of die simulators to account for the effects of an IHS seems like the most likely solution, but there are some concerns when doing this. It seems intuitive to just use a real IHS on top of the existing die simulator. However, there are some physical problems with just stacking things up like that which could affect thermal transfer if the spreader isn’t properly aligned and is allowed to float. If the IHS is permanently affixed to the die simulator, then it may possibly become an artificially rigid obstacle to thin baseplate waterblocks that rely on flexing to mate well with the CPU surface. It is theorized that the CPU socket/motherboard area in a real PC flexes with these thin baseplate blocks enough to get good contact in the middle of the processor but in simulators that this may not be possible. I note that this is only a theory and has not been experimentally proven.
If the modern CPU becomes too difficult to simulate with a copper die, then directly using the CPU as a heat source becomes much more attractive. I have noted that it is possible to groove an IHS top and affix a thermocouple. Both AMD and Intel recommend this “CPU Case� temperature measurement for cooler testing. Both companies presumably know the thermal properties of their products well enough to make such a recommendation.
What it comes down to is that we are perhaps moving out of a period where die simulator advantages far outweighed their drawbacks into a time where the complexity of CPUs as heat sources may preclude simulating them with simple copper dies. In other words, to do good quality die simulator testing we will also have to do good CPU-based testing if we hope to make the high resolution simulator results relevant to real world CPU cooling. For this reason, the third option in my opinion is the only viable choice for testers to embark upon.
A note: this would require web reviewers to (gasp) interact with one another more strongly than we have in the past. The fact of the matter is that everyone is measuring performance of cooling devices on some heat source. It should be possible to correlate data from my Socket-A CPU testing, my (upcoming) Socket 939 testing, and all the different die simulator tests that are being done. The general idea would be to find a way to tie the results from different simulators back to observed performance on actual CPUs.
Choosing a die simulator vs. a CPU for YOUR test bench
A new reviewer might like some practical guidance on how to proceed with cooler testing in the current state of technology. And all readers might benefit from seeing some of the underlying weaknesses of different methods as well. I see three options for testing depending on what is most important to you:
• Testing with a die simulator. If you seek to produce results in the “C/W� format then this is your best option. If you are doing serious research and design, then having tight control of all the variables of your heat source can allow you to produce high quality data. This requires extensive instrumentation for data collection as well as a clear strategy for maintenance (surface flatness of the die). The tradeoffs you make for high resolution data are cost (of course!) and constant maintenance, troubleshooting, and quality control. And you are also conceding that mounting issues, secondary cooling paths, and secondary heat sources are not as important as that “C/W� number that you’re collecting.
• Testing with a stock CPU with intact IHS. If this is your preference then the only good course of action is to use an Intel 775-based CPU and motherboard and have a machinist groove the IHS so that you can fit a tiny type-t thermocouple over the core. This allows you to get good temperature resolution/accuracy, a real CPU as your heat source, and low risk of damage from repeated cooler remounts. The Intel TIM inside the CPU is an indium solder that will not degrade or come loose as far as anyone knows so it is stable. The socket 775 design does not rely on pins and so damage from repeated mountings is minimal. The major tradeoff you are making is that the temperature monitoring is taking place at the top of the IHS instead of in the CPU and so temperature compression is going to make the differences between coolers less pronounced.
• Testing with a real CPU and internal diode readings. If this is your preference then the only brand of CPUs available to you is AMD. This technique requires desoldering the diode pins and soldering in wires in their place connected to a quality diode reader. This technique also in my opinion requires removal of the IHS. The reasons are twofold: 1) the TIM joint inside the CPU will probably change with remounts and thermal cycling and 2) the CPU will rip out of the socket and bend pins/rip out your wiring if the IHS is left on the chip. The advantage of doing this sort of testing is that you have good accuracy and resolution for the actual internal CPU temperatures (and so less compression). The tradeoff is that your test results are being tailored for the hardcore fringe who will perform surgery on their CPUs.
For my part, I am interested in comparing cooler performance on an AMD 3000+ with IHS vs. a bare die. My feeling is that the AMD IHS presents a nightmare in terms of long-term TIM stability (between CPU die and IHS) and in CPU life expectancy on my test bench (bent pins). I suspect that most of the people who water cool for performance are willing to pop the IHS off their AMD64 too if it means extra cooling. I am more than happy to share the results from my test system with all other reviewers so that we can much more easily make cross-platform and cross-testbench comparisons. This will not be a fast and easy process, but I think it is the best shot for us to improve the usability of our test results to the general cooling community.
|
|
|
|
Random Forum Pic |
|
From Thread: Can the MCW6002A w/step base fit the K8 socket |
| ProCooling Poll: |
So why the hell not? |
I agree!
|
67%
|
What?
|
17%
|
Hell NO!
|
0%
|
Worst Poll Ever.
|
17%
|
Total Votes:18Please Login to Vote!
|
|