TIM comparison on LN2

[Rauf]

With the last years development with cracking skylake, introduction of kryonaut, the “huge blob methodâ€, kingpin blue paste, full pot pascal, die shrinking etc. TIM quality has never been more important than now. So, I decided to do a few tests with the most popular brands, and also to share some of my experiences with different TIMs and setups. One important note is that this information is presented as is and the testing is obviously not done in a proper scientific way with a large number of tests of each paste etc. I don’t have enough time for that… But I have done my best to do a proper comparison. However, please note that what I present here are not facts. It is quite common for overclockers to test something once and then think of that as the one and only truth. But the truth is that there are so many parameters that it is almost impossible to find any universal fact about overclocking. What works for you might not be relevant for another chip, pot, mounting method etc. So once again I would like to stress that what I present here are my “findingsâ€; They are not facts, but hopefully my results and experiences will at the very least provide some useful information.

 

The contenders

The contenders are of course: T. Grizzly’s Kryonaut, Kingpin’s blue paste and Gelid’s GC Extreme. And to keep thing transparent, I have bought each myself, I’m not sponsored or involved in any way with either product.

 

To crack or not to crack

When skylake was launched the issue was that the old mounting methods resulted in TIM cracking because of the small die and really low temperatures. However, there are actually two separate phenomena which are often mixed up; not all cracks nowadays are actually cracks! We now also have the TIM “explosion†phenomena, which happens mostly with high power, low temperature GPUs (pascal), but can also happen on CPUs in very high-power benchmarks such as XTU. I’m not sure exactly how or why the TIM explosion happens, but power seems to be the main ingredient to make it happen. For example, during the GOC 2016 qualifier with GTX 1060, it was very clear that the best GPUs had the most problems with blowing the TIM, while some of the lower clocking ones could run full pot. The difference here obviously being the heat flux of a GPU running at 2850MHz vs 2600-2700MHz. TIM explosion is easy to spot when dismounting as you will see a clear path on one side of the GPU where the TIM has exploded away from the core. With the “blob†method we can prevent cracking, but unfortunately the explosion is still not possible to prevent by any method I know of.

 

Mounting

For the sake of providing as much information about the results I have gotten, an maybe to help anyone still having problems with cracking, I’ll describe my mounting method here. The way of mounting that works well for me is to very carefully clean die, ihs (if CPU) and pot with pure acetone. I remove all glue from ihs, but keep most of the glue on CPU pcb. I only remove the protruding edges of the glue to ensure that the ihs doesn’t sit on one of those little edges. I have not found any improvement from removing the glue completely from pcb. Then, if using kryonaut or gelid I just apply a big blob of TIM on the die and then carefully place the ihs, using the ihs to slowly let the TIM spread. Keep the ihs levelled to spread the TIM evenly. If using kingpin I spread a layer of TIM on the die, not too thick and not too thin, then place the ihs on the die. I loosen all screws for the socket hold down when mounting and carefully tighten it after mounting ihs to make sure the ihs is always levelled with the die. After the ihs has been properly mounted there is not much more to say, the rest of the mounting is not very important and hard to do wrong. One thing to note though might be that I always mount my pot very tight. I don’t actually think it makes a difference, but if you run into problems you can always try to mount tighter.

 

My setup

For these tests I have used my golden 7700K on ASUS Z270 Apex MB. Normal ihs, not lapped. Pot used was a kingpin venom. To replicate real world benching I was benching mem at 4133+ 12-11 and cache at 6300+.

 

Testing and results

I decided to test a few different multithreaded benchmarks to see the difference in maximum frequency, depending on how high the load is and what type of load it is. The goal for each test was to find the maximum frequency each test could run with the different TIMs. Actual score is obviously not important. Also, Vcore was not fixed as the different pastes might like different voltages. The tests I decided to use was Cinebench R15, R11.5, XTU and Aquamark (4T enabled, two core affinity in benchmark). These tests should be enough to give an idea of TIM performance. For simplicity, all 8T tests were done in the same OS: Win 10 64 Creator. Aquamark was tested in Win 8.1 64. Each TIM was tested twice on LN2.

 

Enough talk, now for the results:

As you can see, kingpin’s blue paste worked best for me in all tested benchmarks. In the cinebenches the difference was quite significant.

 

Other findings regarding TIM on LN2

Here I’ll just write down some of the experiences I’ve had with the different TIMs. No extensive testing have been done regarding this, so some of it might be inaccurate and just random differences or bad luck. But I think it might be worth sharing anyway, if nothing else it’s something to consider and exchange experiences about.

 

During the tests for this TIM comparison I found that Gelid would degrade a bit after 1-2 hours of benching. It was not a proper crack, but benching frequencies would drop around 100MHz. I normally don’t use Gelid, so I’m not sure if this was a coincident.

With kryonaut I never have issues with cracks, but during the first test round I had a TIM “explosion†while running XTU. The second time worked well. It was also the first and only time I had TIM explosion on CPU. But on the other hand I rarely run XTU, these tests were actually the only runs I have done with this CPU.

I have not been using kingpin paste for very long, but have managed maybe 20 sessions with it. And so far I have not had any cracks or explosions. However I have only benched 1080 Ti at around -140, not lower. From what I have heard from others TIM explosion will happen if you go lower in temp.

 

While benching for GOC 2016 with GTX 1060 I had lots of problems with TIM explosions with both Kryonaut and Gelid. During my tests here I found that Gelid would explode at a slightly lower temperature than kryonaut (-175 vs -165), but kryonaut had far better performance: there was a very significant difference in maximum frequency. Also, kryonaut had the advantage that after the first TIM explosion, it would withstand TIM explosion slightly better after a quick reheat of the pot to positive temps. Gelid would not work at all after a TIM explosion. Sadly, kingpin paste was not available at this time, so I could not test it with the GTX 1060.

 

While benching at computex this year I was using a 6950X BW-E cpu (soldered ihs) with cold bug at around -100C. Mostly by chance I ended up using all three different TIMs and Gelid and Kryonaut was performing well, as expected. But to my surprise kingpin paste was way worse, like almost 100MHz lower in Timespy CPU test. Maybe kingpin does not perform that well in the -100 temperature range, or maybe it was just some random issue. I have not rebenched this to verify.

 

So in conclusion, next time I’m buying TIM I will order kingpin It is a very good TIM for full pot CPU benching. It is working very well for GPUs that can go very cold as well, but I have not done any real comparison here, maybe in the future… But I will also keep a tube of kryonaut for the high CB CPUs like BW-E and Sky-X.

[GeorgeStorm]

Was already planning on grabbing some KPC Blue next so glad this supports that idea

[flanker]

good article man, thanks

[TAGG]

Somehow this makes me want to test my cheapo chinese HY710 on LN2... most of my SS scores are done with that paste, because I don't feel like wasting Kryonaut...

I don't know how much i loose, but these scores don't seem too bad to me:

TAGG`s HWBOT Prime score: 4321.32 pps with a Core 2 Duo E8400

TAGG`s Cinebench - R15 score: 782 cb with a Core i5 2500K

TAGG`s HWBOT Prime score: 4309.41 pps with a Core 2 Duo E7400

All done on crappy 10 cent/ gramm Thermalpaste