Articles: Cooling

Bookmark and Share

Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 ]

Testing Preparations and Intel Core i7 Cooling Tricks

Before we moved on to the actual testing of the cooling solutions performance we decided to study the technical specifications and physical parameters of a very important component of a CPU that may affect the cooling efficiency quite significantly. I am talking about the processor heat-spreader. First let’s talk about its technical specs. Our Intel Core i7-920 processor was made in Costa-Rica and marked as “SLBCH”:

Its nominal frequency is 2667MHz and core voltage – 1.25V. Other specs are as follows:

We tested the evenness of the heat-spreader using a metal ruler and it seemed OK:


You can notice a slight bump in the CPU cross-section, which showed in the thermal compound imprint on the ideally even polished vase of the Scythe Mugen 2 cooler:

Of course, an imperfect imprint like that is totally unacceptable for our today’s massive test session that will determine the fate of quite a few cooling solutions. That is why we decided to void the warranty and even out the heat-spreader.

After a few minutes of work on a glass with fine sanding paper we could clearly see the unevenness:

After another 15 minutes and one more even finer sanding surface, we managed to eliminate all the major bumps:

Fixing the CPU heat-spreader surface as described above immediately lowered the temperature of the hottest processor core by 4°C! Impressive, isn’t it? However, when we removed the CPU cooler, the thermal paste imprint didn’t impress us at all:

Another 15 minutes of work on the glass surface with the finest sanding paper we could find improved the CPU temperature by another 1-2°C and generated a more or less acceptable imprint:

We decided to skip the polishing of the heat-spreader to mirror-shine. And the reason for that is explained very well on the scheme from the official Intel web-site:

Look at the actual die size beneath the heat-spreader: it is only 14.3 x 18.8 mm, which makes ( according to Intel’s official specs). So, the die size makes only a little over 20% of the heat-spreader. Therefore I believe that the main heat flow will be going from the CPU to the cooler in the spot right above the actual die. That is why the main goal is goal is to ensure a perfect contact in the center of the heat-spreader. That is exactly why Thermalright coolers are so efficient even despite their prominent base plates. Of course, if you even out the base on these coolers you will improve the CPU temperature a little, because some heat is still distributing over the CPU heat-spreader.

And one more thing that we should dwell on in this part of our today’s article. Look at the CPU core under the heat-spreader:

It is evident that the die is of rectangular shape and its longer side goes along the processor heat-spreader. In other words, when you install the CPU into the board, the die will be positioned along the longer side of the board. It is logical to assume that the more heatpipes will fall into the die zone, the more efficiently the cooling will be. So, the coolers should be installed onto Core i7 processors with their heatpipes going across the CPU (parallel to PCI-E slots on the board). We paid special attention to this particular matter in our today’s review on purpose. So, if we analyze everything we know, we will see that the majority of coolers that could be installed with their heatpipes across the processor did, in fact, prove more efficient than with the heatpipes along the CPU. The only exception was Thermalright IFX-14, however, it could be its uneven base that contributed to the final result. Until it is evened out, we can’t really make any conclusions about the preferable cooler orientation on the CPU.

In conclusion to this part of our article I would like to offer a small diagram with the Intel Core i7 thermal readings under workload created by different applications:

Just like with Intel Core 2 Quad/Duo, the highest processor load is created by Linpack algorithm. OCCT program may have an error (or may be using a different Linpack version), because it heats the CPU less than LinX.

Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 ]


Comments currently: 30
Discussion started: 03/25/09 05:15:06 AM
Latest comment: 06/15/16 10:27:46 AM

View comments

Add your Comment