Article Thumbnail

Ivy Bridge CPU TIM Paste Replacement

Written on September 6, 2012 by Matt Bach
Share:

Introduction

Even before launch of the Intel Ivy Bridge CPUs in April 2012, it was discovered that the CPUs were running a bit hotter than expected. It was expected that they would run at lower temperatures than their Sandy Bridge predecessor, but this was found to not be the case. One of the earliest reports of this issue came from Overclockers.net, where it was claimed that the heat issue was caused by Intel using TIM paste instead of fluxless solder between the CPU die and the Integrated Heat Spreader (IHS). 

The TIM paste was proven to be the culprit when the Japanese site PC Watch published an article showing how they removed the IHS and replaced the TIM paste with both high quality thermal paste and Liquid Pro. They reported a load temperature drop of 8-11 °C at stock clock speeds, and an amazing 15-20°C drop in load temperatures when overclocked to 4.6GHz.

Since then, there have been numerous forum threads across the internet of end users replacing the TIM paste themselves and the results they found (example one and two). The results aren't quite as impressive as PC Watch's and vary widely between a drop in load temperatures of 2 °C all the way up to a drop of 10 °C. Some users also used this as a way to increase the amount of overclock they were able to achieve, and while results are scarcer than temperature results, the trend appears to be a 100-200 MHz bump to their overclock.

Ivy Bridge has been out for a while now, so we were interested in seeing if anything has changed since April. So, like PC Watch, we decided to open up an Intel i7-3770K and replace the TIM paste ourselves with some Gelid GC-EXTREME thermal compound. In case you are thinking of doing this yourself, keep in mind that this is something that can easily cause damage to the CPU and will completely void your warranty with Intel.

Replacing the TIM Paste

Using a razor blade, we removed the IHS to get to the CPU die. After replacing the paste, we applied Permatex RTV Silicone Gasket Maker to the edges of the IHS and clamped it down onto the main CPU. While removing the IHS completely would result in better temperatures, it would make the CPU lower than heatsinks are designed for and would require a lot of extra work to get the heatsink securely seated to the bare CPU die. The CPU die is also fairly fragile and the IHS works as a crush plate, preventing the heatsink from damaging the CPU die, so we wanted to leave it in place for that reason as well.

The actual work was fairly straight-forward, and since a full guide is available in the article we linked, we will simply leave you with the pictures showing various stages of our progress.

Before modification IHS removed TIM paste and sealant cleaned

 

New thermal paste applied IHS replaced and clamped down


It is worth pointing out that we actually had to perform this process multiple times on multiple CPUs. The first time was simply to get a feel for the process and see how challenging it was. The second was intended to be the CPU on which we performed our extensive testing on, but after replacing the TIM paste we found that the fourth memory channel was no longer functional. Since we did not want a damaged CPU skewing our results, we had to mod a third CPU for our testing. This just goes to show that even when professionals are performing this modification, there is a decent chance of damage occurring to the CPU.

Test Setup

While we wanted to see how much better CPU temperatures are with the TIM paste replaced and what it would mean for overclocking, we also wanted to see how it would affect system noise. Lower temperatures should result in lower fan RPMs, but the question we want to answer is how much of a difference it will make.

At Puget Systems, we have a very heavy focus on quiet systems with our popular Serenity line of desktops. We normally outsource our noise certifications for our Serenity line of systems to partners including Silent PC Review since the systems are so quiet that we can not accurately measure them in-house, but in the interest of time we decided to instead simply record the CPU fan RPMs to get a general idea of whether or not replacing the TIM paste will result in lower system noise. If the results are significant enough, our planned followup is to work with our partners to get accurate full system noise comparisons.

To perform this testing, we used a two system configurations centered around our Deluge A2 (performance) and Serenity (quiet) line of systems.


In the Deluge A2 configuration, we will be running the system at stock CPU clock speed, as well as at a 4.5 GHz overclock, which is the highest stable overclock we were able to achieve with this hardware. On the Serenity system, we will not be doing any overclocking, but will test with both a discrete video card as well as with the onboard video from the CPU.

While we did record temperature and fan speeds from both the CPU and the GPU, the GPU results did not vary by any significant amount so we will only be focusing on the CPU results. CPU temperatures were recorded with CoreTemp with the results presented as the average temperature of the four cores. CPU fan RPMs were recorded using SpeedFan. We let the system idle for at least 45 minutes before recording the idle temperatures, and to put the system under 100% load we ran a combination of both Prime95 and Furmark.

Our margin of error for our temperature measurements is 1 °C and for fan speeds is 50 RPM.

Temperature Results

In the Deluge A2 configuration, we see a much smaller drop in temperatures than we expected. While we expected the idle temperatures to stay the same, the 2.5 °C drop in load temperature at the stock speed and the 2.25 °C drop in load temperature when overclocked is much less than we thought we would see given what we have read online.

We are using the Asus QFan fan throttling feature that is on this motherboard, so it may be that these underwhelming results are more a matter of the motherboard realizing that it does not have to ramp up the fans as much to keep the CPU at a good temperature. So while the temperature is not significantly lower, perhaps the CPU fan speed is significantly lower. The answer to that question, however, will have to wait until the next section.

The results from our Serenity configuration are more what we expected to see with a 7.5 °C drop in load temperatures with the Asus GTX 670 installed. Curiously, the load temperatures only dropped by 3.5 °C when we used onboard video. While the ~2 °C drop we saw in the Deluge A2 configuration is underwhelming, these results are much more promising.

Fan Speed Results

We postulated that the mediocre temperature results we saw in the Deluge A2 testing was due to the Asus QFan ramping, but unfortunately that does not appear to be the case. At both stock and overclocked clock speeds, the CPU fan speeds were all within our margin of error, and thus are effectively unchanged by replacing the TIM paste.

Once again, our results in the Serenity configuration are much more interesting than the Deluge A2 configuration. While the fan RPMs when using onboard video are essentially unchanged, we did see a 80 RPM drop in the CPU fan speed when using the Asus GTX 670. Very precise measuring devices in an anechoic chamber might be able to discern a noise difference, but unfortunately in the real world a difference of 80 RPM is so minor that it would be almost impossible to detect with the human ear.

Overclocking Results

Before we get into our overclocking results, we need to make one very important statement: When we perform an overclock - any overclock - stability is always our foremost concern. Getting an extreme overclock may be good for bragging rights, but if it causes your system to be even slightly unstable, that's not a trade-off we believe our customers want us to make. For this reason, our overclocks are often not as high as what some may post to the enthusiast forums or from our competitors who put sheer clock speed ahead of stability.

With that said, we were able to achieve an additional 200MHz on our overclock by raising the CPU voltage an additional .045V, resulting in a final overclock of 4.7GHz. We were not at all temperature limited before the TIM paste replacement, so this result indicates to us that replacing the TIM paste is doing more than just reducing the temperature of the CPU cores. There is a lot going on when overclocking a CPU, and it appears that replacing the TIM paste assists in ways other than simply reducing the CPU temperature.

As we stated, the CPU core temperature was not the limiting factor in our overclocking, which becomes obvious when you look at our temperature results. Both of the above results are with the TIM paste replaced, and we saw an 8.25 °C increase in CPU temperatures as a result of the extra .045V going to the CPU.

A roughly 8 °C increase in CPU temperatures is quite a bit, and it shows in our fan speed results. In order to keep the CPU cool, the CPU fan had to ramp up almost 400 RPM. This is a significant amount, and results in a noticeable noise increase.

Conclusion

Temperature: Our temperature results were very inconsistent. The Deluge A2 configuration at both stock and OC clocks saw about a 2 °C drop in load temperatures; while the Serenity with the Asus GTX 670 saw an impressive 7.5 °C drop in load temperatures. The Serenity configuration with onboard video was in the middle with a 3.5 °C drop in load temperatures after replacing the TIM paste.

To be fair to Intel, the temperatures on Ivy Bridge CPUs are great even without the fluxless solder, so while some of these results are significant, they are unlikely to greatly affect the majority of PC users. 

Fan Speeds: The only test that showed a decrease in CPU fan speeds beyond our margin of error was the Serenity configuration with the Asus GTX 670 which saw an 80 RPM drop in CPU fan speeds. This is very minor, however, and for all practical purposes is small enough to not matter.

Overclocking: We were able to achieve a 200MHz (.2GHz) higher overclock with the TIM paste replaced, but this also resulted in much higher CPU temperatures and fan speeds. So while replacing the TIM paste may allow a user to overclock further, the CPU cooling needs to be able to handle the additional heat that is produced.

So now the big question: is replacing the TIM paste worth it? Considering all of our results and the fact that even as careful as we were we ended up damaging one CPU, we would say that in most cases it is simply not worth the risk. While we did see load temperatures drop across the board, the CPU temperatures were already well within acceptable limits prior to replacing the TIM paste. If replacing the paste also resulted in significantly lower fan speeds we would be tempted to say that it is worth it from a system noise standpoint, but it didn't.

What is interesting is that the limited testing we performed on the first CPU we modded (which again was more to experience the process rather than do in-depth testing) showed a 10 °C drop in load temperatures at stock clock speeds after replacing the TIM paste. In comparison we saw only a 2 °C drop we saw in our extended testing on our third modded CPU.

What we believe happened is that the first CPU had a particularly bad TIM paste application, while the third was relatively very good. No matter how automated the manufacturing process, there are always variances from plant to plant and even from day to day. The first CPU we tested likely had a relatively bad TIM paste application, while the CPU we used for testing had relatively good paste application. This is supported by two facts. First, our pre-TIM testing showed that the first CPU we modded ran about 7 °C hotter than the CPU that we performed our extended testing on. Second: after replacing the TIM paste, both CPUs had temperatures within 1 °C of each other. So in other words, replacing the TIM paste will only cause a significant difference if the TIM paste was badly applied in the first place.

At the start of this project, we were exploring the possibility of offering TIM paste replacement as an option for our customers. After seeing how inconsistent the results are and how large the risk is, we have decided against doing so at this time. Still, we know that there are many out there that love to tinker with their system for the possibility of lower temperatures. So let us know in the comments below; is replacing the TIM paste something you would ever do on your own CPU, or do you feel that it is simply not worth the effort and risk?

Tags: CPU, Cooling, Performance, Temperature, Noise
Foonly

So you replaced original thermal paste with another (arguably slightly better) thermal paste, and you are surprised the results didn't blow your pants off? Who would have expected that? :-)

Please, try again with a good metallic TIM, like Coollaboratory Liquid Pro. The thermal conductivity of this stuff is an order of magnitude better than any paste. If you're going to risk damaging your CPU, you'd want to apply the best, no?

Posted on 2012-09-07 12:01:50

The thing is that others (including PC Watch and the two forum posts we linked) saw up to a 10C decrease in temperatures by using high quality thermal paste similar to the paste we used. Certainly using something like Liquid Pro would improve the results (PC Watch did use Liquid Pro actually, and they saw an additional 3-5C drop) , but we wanted to take things in the same direction as the enthusiasts who are reporting their results on the web.

If you are thinking of using Liquid Metal, be aware that there are a few things to keep in mind. First, it is conductive, so you have to be very careful where it gets applied. One drop in the wrong spot and you no longer have a working CPU. Also, it is primarily created from the metal Gallium which can eat away at many types of metal. Aluminum is the biggest metal to avoid, but I've heard reports that it can have an effect on copper as well; it just takes longer. I'm not a metallurgist, so I can't say exactly what happens, but you definitely have to be careful what you are applying liquid metal to. Here's a cool video showing the effect of Gallium on an aluminum heatsink: http://youtu.be/JHHI2Lk79cY

Posted on 2012-09-07 18:59:12
rge

Actually your results fell in line with most others that tested at stock or near stock voltage, since at low vcore, the dissipated power is very low, and replacing tim will only net a few C. You tested at stock vcore (which is around 1.05v or 1.1v max), and then you say you added .045 vcore which is around 1.1v to 1.15v. Who in their right mind would delid a cpu to only run 4.5ghz and 1.15 vcore, when power dissipated, hence temps are not going to be high in that range. That makes as much sense as delidding to run stock.

Assuming accurate testing (ie measuring intake ambient temps, running prime only, and averaging core temps over a collected period while logging ambient temps) if thermal resistance is decreased by improving tim, as power dissipated increases, a corresponding increasing improvement in temps will occur.

You tested at low end near stock 1.1 or 1.15v and 4.5ghz. But you failed to test at high end. You did the control, but never did the actual test.

Most 3770K's with HT on, require ~1.3v for 4.6 to 4.8 prime stable for 12 hrs (without whea errors), and temps get up in 80's with ambient of 22-24C, and in 1.4v range and 4.9ghz temps are up near 100. Where one would expect to see 10+ reduction in temps is when power dissipated is much higher than you tested, in 1.35v range and 4.7-4.8ghz and higher.

Why waste 3 cpus, to only do the control part of the experiment at low end, then fail to actually test at high dissipated power end, where EVERYONE ELSE SAW GAINS.

Posted on 2012-09-11 03:20:14

I think this is an excellent example of how the focus of our articles, which are primarily from the viewpoint of a system integrator, are different that most sites on the web that would do this sort of project. Where most look at how their testing could theoretical impact a product (in this case the raw CPU voltage) we almost solely look at the practical applications. In our testing, we did not get any higher of a stable overclock when we applied a higher voltage (up to our standard cut-off point of 80c on the hottest CPU core), so we had no reason to set the vCore higher than we did. Stability is always our highest priority, even above performance, and while many advanced users are comfortable running a CPU above 80c, that limit has served us very well over the years to ensure that our customers never have any CPU thermal issues.

So, yes, we agree that if we applied a higher voltage to the CPU then the difference in temperature would likely be greater. But we were concerned with the practical effect that replacing the TIM paste has on overclocking (in our case, about .2 GHz) rather than the thermal difference at voltages that we as a system integrator would be comfortable running.

Posted on 2012-09-11 04:03:28
rge

The point is your following conclusion failed to have appropriate disclaimer, hence is over generalized and inaccurate, because you only tested at/near stock vcore and with either a cherry OCing sample or you were not prime stable... " So in other words, replacing the TIM paste will only cause a
significant difference if the TIM paste was badly applied in the first
place."

So let me correct your statement:
For cpus running at/near stock voltage, Replacing tim1 paste will only result in a significant temperature drop if tim1 paste was poorly applied. If overclocking in higher range of 4.6 ghz to 4.9 ghz, and your cpu requires vcore for prime stable near 1.3v range, other testers have found a much higher drop in temps, ie 10+C range (which is expected based on physics of power increasing by square voltage). Though we dont sell systems with such high overclocks, so we did not test in that range and hence can not comment on temp drop in that range.

And if you found a cherry cpu that overclocks to 4.7 with .08v over stock and it is prime stable without whea errors (errors prime doesnt see, only viewable in event viewer/warnings), then congrats, but that isnt the norm. You will see a chart with some of better prime 12+ hrs stable cpus at oc.net, and 4.7 with .08v over stock would be a rarity.

Posted on 2012-09-11 17:39:04
Digi

It reminds me of those old athlon xp cpu's that had no heat spreader.

Posted on 2013-09-22 03:44:22