Thermal Paste Roundup 2015 – 47 products tested with air-cooling and liquid nitrogen (LN2)

Test circumstances

There are several different ways to apply the thermal paste on the CPU or GPU. The layer itself should be as thin as possible but thick enough to fill all gaps between the CPU and the cooler. Pastes with a very low viscosity can be applied using method #1 because the pressure of the cooler will spread it between both surfaces. The same principle applies for methods #2-4. However, from my experience some pastes are too thick and won’t properly spread between the surfaces. That’s why I always use method #5. I use an applicator which is often included in thermal pastes to spread it over the CPU. 

Using method #5 I can be sure that each paste will be tested under the same conditions and the whole surface is covered.

Spreading Test of different methods:


I’m using a Phanteks PH-TC14PE air-cooler to have the same mounting pressure and temperatures on the test bench as using a normal computer setup. For air-cooling each paste will be applied and measured 3 times to make sure there are no measurement- or mounting-errors. If one out of the 3 measurements is way out of line a 4th measurement will be done. Afterwards the arithmetic average will be calculated.

To see how consistent the results are I did 10 tests with Thermal Grizzly Kryonaut. I always applied the same amount of paste and spread the paste as you can see on picture #5. Of course there is always a small difference in the measurement but you can see that the results are very consistent.

Test #12345678910
Delta Temperature7.887.917.927.887.877.927.927.897.857.87

160 Watt continuous load simulate an overclocked Intel Core i7 Haswell CPU on high load.


  1. Introduction and Basics
  2. Test requirements
  3. Improvements compared to the last test
  4. Test circumstances and consistency
  5. Gramm vs. Volume
  6. Air-Cooling results and buy recommendation
  7. Extreme Cooling Test
  8. Extreme Cooling Test (results#1)
  9. Extreme Cooling Test (results#2)
  10. Extreme Cooling Test (results#3)
  11. Product Details
  12. Conclusion

2,748,131 total views, 0 views today


  • Boris

    I’m interested to know how you managed to get such abysmal results with the “EK Waterblocks EK TIM” aka Indigo Xtreme, those numbers just don’t seem right, how did you go about applying it to 2 copper plates?

    • Yeah

      Seems 100% right to me. Shit is garbage

      • der8auer

        Honestly I tried with 3 different pads and none of them spread perfectly over the whole surface. In addition it was almost impossible to remove it without residues from the copper. I even had to use sand-paper to remove the rest. So honestly considering the price I can’t recommend it.
        And on a side note: I paid for all the products myself. Considering that I spent over 60€ for the EK TIM I was really disappointed.

        • Jon

          I disagree, I’ve had Indigo spread evenly every time I’ve used it. It works just fine if you do it properly. However, the only places that sell it are Performance PCs and Frozen CPU, both businesses that are in disrepair and screw up your orders quite frequently.

    • Jakusonfire

      The EK TIM is not GC extreme. They still supply GC with some blocks but their branded paste is different. They have stated it is about equivalent to mx4 I think

      • Jakusonfire

        Or maybe I misunderstood. Which EK-TIM is it? EK-TIM Indigo Extreme or EK-TIM Ectotherm paste?

        Indigo is very fussy about how it is mounted, used and cooled back down from phase change especially. Its easy to ruin it by cooling too fast. In my experience using it, it has outperformed everything I tried. I just don’t use it because I change hardware too often and don’t care about 2C

  • Xtreme Addict

    Really nice round up! 🙂

  • Achill3uS

    jeez, hell of a work here 🙂 nice roundup Roman!

  • VSG

    What a tremendous job, Roman! I just saw this after Massman alerted this article on HWBot but I am definitely going to share this around now. Just about everything matches my own findings on water too so it’s not just air and LN2!

  • Chili

    Is that another spot for an 8-pin under the 6-pin with extra pads for 8-pin? This would mean the reference could be converted to dual 8-pin or even 6+8+8?

  • Mac

    Great work! But why is CL copper so bad? It is almost dead last. Other comparisons have it close to CLU.

    • der8auer

      I measured and applied the paste 3 times there. I also did a check with Liquid Nitrogen and the paste was just incredibly bad for me. Where did you spot a good result with this paste? Couldn’t find a review of it.

  • Pingback: recomendation for thermal compound

  • Pingback: recomendation for thermal compound - Page 2

  • I’ll agree as to the quality and comments on the reign of of GC-Extreme; it’s my “go to” TIM for GPU Waterblocks as it tends to stay pliable longer giving me more time to apply to all the VRM and memory chips on both sides of the thermal pad. But for CPU water blocks and air coolers, I don’t see a TIM roundup as being complete w/o Shin Etsu being in the mix.

    It’s always competed well with Gelid DC Extreme and is a bargain at $4.99 on newegg.

  • Those Phobya temps seem a bit on a biased side. Other tests I’ve seen always had Liquid Ultra, with Liquid Pro immediately following behind, at the very top.


  • Pingback: gelid gc extreme vs arctic silver 5

  • b

    Great test!
    I can add that from my experience MX-4 has better resuts after warming it up for some time. I put on paste, check temperatures and then let computer run with full load for some time (sometimes with worse cooling conditions, just to warm it up more). After that I let it cool down and the next time I check, temperatures are lower by 5-7°C at minimum load and a bit less at full. Also it takes much longer to get to the performance limit of entire cooling solution.
    I wonder how this pre-test heating would affect your tested deltas.

  • M2V

    I really miss control test in the table, one that uses no thermal paste at all! It would be really interesting to see how does bare contact compare with lower end solutions.

  • gqneon

    Above all else, freaking awesome job and thank you for taking so much time and personal expense to publish this. I’m in the middle of trying to maximum overclock a CPU I moved to a new board and have been swapping tims. I had liquid pro on it in old build delidded, phenomenal product. Moved to new board with AS5, 10 degree temp increase but partially due to upsetting the IHS from the core removing and installing. Went back to liquid metal pro when I couldn’t get my old stable OC settings to prime95 stable and saw 7-10 degree drop in core temp again. Stuff is amazing.

    Going to swap tim on 3 gpus when the third waterblock gets here from AS5. I’m terrified of putting liquid pro on the though because of all the caps around the core. Seriously considering kryonaut being nonconductive or maybe even more gelid extreme , but that stuff is like gorilla snot , hard to apply and hard to remove.

    Do you recommend liquid metal on gpus? And would it be safe to put nonductive paste on the little caps etc near the core if I do to prevent errant liquid metal from running if it were to occur?

    Thanks again for your hard work. It really debunks a lot of marketing BS!

  • VDelVec

    I dunno. My experience with Indigo Xtreme always had it perform BETTER than any of the Liquid Metal products. Also, you should try their new stuff, Indigo Xtreme XS, it can be found here:

    • VDelVec

      In fact, I have some extra Indigo Xtreme XS lying around. I can send it your way instead of you having to pay $21 USD for it.

  • Jesse

    I understand, It’s very difficult to get consistent results in this, but you need to know method 5 causes bubbles, which are very bad for temps, always use method 1 if you want cooler results.

  • b

    der8auer could you post a video showing how you spread paste with the applicator you mentioned on page 4? I wonder what it look like and where can i get it from. Recently i have had issues spreading paste on PlayStation’s IHS, i used credit card and wasn’t happy with result.

  • Rosemary

    Dear der8auer,

    Great round-up. I would like to know your stance on Tuniq TX-4 air-cooled.


  • Pingback: Anonymous

  • Wu Nee

    @der8auer Could you try the ultimate in retro heat sink compounds? Back in the 1990s and early 2000s, the recommended and most popular grease was the only thing easily accessible in the US. It is Radio Shack part no. 276-1372, with an Archer brand name on it. I still have some!! I have seen it on eBay for sale as new old stock. I am dying to know how it performs, as it was surely made before anyone used a CPU heat sink. It was made I’m sure for power transistors and the like. Try ebay auction # 272031525869 … there is also a new package, but I am not sure if the formula was changed. The original is greasy white. I’m sure I’ve used this same tube a dozen times or more over the years starting with Pentium P54C.

    • Zdenek

      I JUST applied it on my 2 GPUs in my laptop (before I even found this article). It was an emergency and I found an unopened tube I bought in Radio Shack years ago. The compound on my GPUs was completely dry and one of them was running 99 C on idle. Now they’re running at 74 C and 62 C which is more acceptable but I ordered some Thermal Grizzly Kryonaut in hope I can get those bad boys below 55 C.

  • Zdenek

    Holly molly this Kryonaut stuff is good!!!

  • Raymond C

    An HP2560P laptop becomes very hot and the HP forum suggested cleaning out the dust and applying thermal paste to the CPU could help dissipate the heat. I checked the thermal grizzly options on AMZN and is available in 4, 8, or 11 gram packages. @author: what volume did you purchase? I would like to understand what is ‘enough’ for an HP laptop CPU. Thank you

    • Dan

      4 grams will be good for many coatings, if applied properly – that is, not too much. Most Intel CPUs only need about as much as a grain of rice – possibly a TINY bit longer. Larger AMD CPUs can use an amount about the size of a pea.

      Clean the old TIM off properly. I use AcrtiClean 1 & 2, it works much better than alcohol. You can use a paper towel to clean the heatsink and CPU, however, finish it off with a coffee filter – they leave no lint or fibers.

  • Hi, tried Akasa 455 on my x520 and got about a 7c drop over whatever was there (anisotropic grey goo), this was after I’d reapplied new grey goo the last time it went twitchy on me.

    Pretty sure the problems with DVDs crashing were overheating not a flaky drive, checked the same disk on my PC and it worked fine.
    I’ve used this in Peltier rigs before and it works well, the large area of contact means the hot side chip never gets much above 80c which is vital when making 3 stage or more systems.

  • BearSpray

    How could I get in contact with you to learn about how to replicate this experiment so I could test different coolers and such? And would it be possible to make the heat variable?

  • Pingback: Anonymous

  • stonebreaker

    I personally wash my hands well and spread it with my pinky.Figure it has the least oil.I went thru the trouble to spread with plastic one time and had the worst result ever ona 5460.Took the tuniq tower off gave it one more chance with the clean pinky and boom.

  • Dan

    Thank you for taking the time to put this review together.

    I have an old tube of GELID GC-Extreme. I put it on my Intel i5 3570K after cleaning off the older Prolimatech I applied 4 years ago [can’t remember which one, might have been PK-1]. Temperatures dropped a few degrees. I run my old 3570K between 4.0GHz to 4.4GHz on air [it will go higher, but I do not like temps much over 60 degrees C, I do not want to destroy anything from heat].

    Running Microsoft Flight Simulator X with add-ons, temps vary from about 55 degrees C to 64 degrees C [depending upon the core – as reported by CPUID HWMonitor and MSI Afterburner, they both provide the same temperature readouts] with the CPU running at 4.4GHz.

    Note: MS FSX is very CPU dependent and gives a CPU nearly a 100% workout. – Also, store TIM with the tip pointing down as the liquid has a tendency to migrate to the tip if the tube is on it’s side, and when applied you get a spot of “oily” material instead of TIM.

    I am considering cleaning the old heatsink fins with soap and water in an effort to remove 4 years of oxidation, I use compressed air to clean with regularly.

  • phil

    I’m not seeing any results anywhere, in FF or Chromium, with or without adblockers, with or without flash etc.

  • smoothgol

    Why were the Extrem Cooling Test results taken down?

  • borgrom

    Do you measure the thermal conductivity of these materials and compare what the manufacturer says? Some of them that have been touted at like 12 only test at like 2 W/mK using ASTM methods like hot disk. Do you have any comment on this? Your tests presumably require knowledge of the exact thickness of the materials, correct—if you are going to solve something like Fourier’s law?

  • Pingback: Arctic Cooling Mx-4 Radio Shack – Adidass News

  • Pingback: Arctic Silver Thermal Paste Radio Shack – Adidass News

  • Reiman Santos

    I have used the TX-4 in the past and was awesome, can you include it? Thanks in advance

  • David Campbell

    Can you test the Cooler Master MasterGel Maker Nano? –

    I’ve seen mixed reviews, most putting it on par with Gelid Extreme and Kryonaut, but some have it as low as Noctua and MX-4, whereas others crown it as the new best thermal compound. I’d love to see your testing on it.

    If possible, could you also include the other two in the MasterGel series, MasterGel and MasterGel Pro –

    Thanks 🙂

  • Stephen

    This has been a fantastic work to date. I have come here many times and realized I’ve never thanked you for all your work and time and documentation. This is tremendously helpful and I’ve been referring to your paste roundups for years without realizing it. I’m going to try some Kryonaut tomorrow on some GPU’s only because for some reason I’m hesitant to use CoolLabs on these new Titan X Pascals, even though I’ve used it on my last few 980 tis.

    Thanks again!

  • Pingback: Arctic Cooling Mx-4 Newegg | Newegg Shell Shockers

  • jon

    I have an NH-D14 air cooler and an I5 4690k, would I be able to use a metal thermal paste even though the fin component on my radiator is aluminium?

  • A Physicist


    While the results show a very good repeatability thanks to a good precision (, meaning the experiment gives the correct rankings between the various pastes, I’d say they don’t represent completely accurately the actual performance of the pastes.

    The problem lies in the rather significant distance between the interface and the 2 thermal probes. If one neglects the heat dissipation through the sides and that it’s just a 1-D conduction problem, Fourier’s law in 1-D is simply : q = -P/A = -k*dT/dx

    where q is the power density, P the dissipated power (P has a positive value ; the minus sign is to show it’s a loss.)
    A the surface area through which the heat flows, k the local thermal conductivity
    dT/dx the temperature gradient along the x-axis

    If one assumes the interface is quite flat, and that there’s the same heat flow across the 3 thermal resistances, one can rewrite the equation above as : q = – k1*dT1/dx1 = -k2*dT2/dx2 = -k1*dT3/dx3

    where k1 and k2 are the thermal conductivities of copper and the thermal compound respectively
    dT1 the temperature drop between the first thermal probe and surface of the copper block it’s embedded in
    dx1 the distance between that probe and the copper block’s outer surface, which touches the thermal compound
    dT2 the temperature drop across the thermal paste layer
    dx2 the average thickness of the thermal paste
    dT3 and dx3 are like dT1 and dx1 but for the other copper block and thermal probe

    The results in the review actually measure the temperature difference between the 2 probes, dT, so it is : dT = dT1 + dT2 + dT3 = (P/A)*(dx1/k1 + dx2/k2 + dx3/k1)

    The thermal performance of the thermal compound is dT2 = (P/A)*dx2/k2, so to get that, one should substract the thermal resistances of the 2 copper layers, which gives : dT2 = dT – (dT1 + dT3) = dT – (P/A)*(dx1 + dx3)/k1

    I don’t know what are the exact values of dx1 and dx3, but based on the test protocols, it’s around 1 cm, and given that copper has a thermal conductivity of around 400 W/m K in this temperature range, that would mean : dT1 + dT3 = (P/A)*(dx1 + dx3)/k1 ~= (160/0.03^2)*0.01/400 = 4.44K

    So all the reported temperature deltas should be decreased by that value, making the top contenders much better.

    Note there’s a caveat about the quick computations done here ; I neglected the side losses, and I’m not completely sure that A remains constant in all 3 thermal resistances, affecting q and the dT values locally. I also modelled the thermal paste layer as being completely flat, neglecting the areas where the 2 copper blocks touch each other directly.

    If a specialist in heat transferts can pitch in, that would be greatly appreciated.

    • Another Physicist

      You must not be an experimental physicist. In a controlled environment, this setup works just fine for the following reasons:

      a) The extra terms in the heat dissipation equations you’ve derived are present and identical in every single run. This means that because we are comparing delta values to an ambient, those terms would subtract out anyway.

      b) The setup is controlled enough where the only variables present are paste type, application consistency, and ambient temperatures. In his methodology, he states that he’s done enough testing to rule out #2 as a major cause of error. He’s also presenting the data in relative deltas. There are plenty of other literature on the subject that says that thermal conductivity is *roughly* linear with ambient. So as long as the room is relatively controlled and the ambient is properly subtracted over time, everything is fine.

      You even acknowledged yourself: “So all the reported temperature deltas should be decreased by that value…”. Translating the entire graph by a constant factor is irrelevant. The takeaways of this test was not to determine absolute temperature deltas for real world applications (CPU’s and GPU’s do not produce heat evenly, their loads fluctuate with time, etc) but to gauge relative performance between the pastes. One should understand that choosing Kryonaut over Artic Silver 5 is not promising you a particular temperature, it’s promising you a 1.5C improvement. This 1.5C improvement would be exactly the same even if both data points appropriately accounted for the other constant heat dissipations in the setup

      • X

        I could be wrong, but the air charts comparison page, I believe the temps are actually in kelvin, not Celsius. Which would mean there is almost zero practical difference between most pastes.

        • Andrew

          Kelvin and Celcius are the same temperature scale, they just have 0 at different values. A delta of 2 degrees Celcius is exactly the same as 2 degrees Kelvin.

      • A Physicist

        It’s indeed a setup that’s fine for a controlled environment, but it does have some flaws :

        a) The importance of extra temperature deltas caused by the 2 layers of copper is indeed something that can be substracted, but it isn’t pointed out in the review, and this somewhat lowers the relative performance of liquid metal compounds compared to regular ones, as the temperature deltas aren’t directly proportional to the thermal resistivity of the paste itself.

        b) Agreed.

        One thing I did learn later on though is that you can’t say the Kryonaut will yield you a 1.5°C gain over the Arctic Silver 5. Because real-life uses are heavily dependent on thermal spreading resistance :

        And this is magnified even more when you delid and apply liquid metal compared to regular pastes between the die and the IHS, as the ratio of their surfaces is significantly different from 1 generally. So yes, the roundup helps see what are the relative performances, but the differences are very likely to be much greater than reported here.

        The air cooling results for example pit various pastes with a 160W load, and you see a mere 1.23°C difference between the Thermal Grizzly Kryonaut and the Coollaboratory Liquid Ultra. It happens I did use both of these pastes on a stock i7-4790K, rated at 88W, and I could already see a difference of 3-4°C (limited resolution caused by Intel Extreme Tuning Utility) between these 2 pastes when applying the pastes just on the IHS-waterblock interface. Whether theoretically with a simple application of Fourier’s law or with der8auer’s results, you’d simply get nowhere near this temperature drop for a realistic thickness of thermal paste, and I can’t stress enough the importance of thermal spreading resistance for real-life uses.

  • Benik3

    After I read, that you used always method #5 – spread paste over the whole CPU, I totally lost interest about this topic.
    It’s the worst method it will really not make always the same conditions thanks to bubbles!
    Too bad, otherwise it could be really nice test 🙁

    • der8auer

      Sorry but this “bubble-myth” is probably one of the worst myths in this industry. Common coolers have a mounting pressure of 200-600 N. Do you really think that there are any air bubbles left if you have 600 N (60 kg) on a surface of 3×3 cm? I know this video on youtube where someone puts a plate on the thermal paste but by hand you can maybe generate 10 % of the real mounting pressure.

      • Iketh

        Method 5 is indeed fine as long as the cpu and heatsink have been mildly lapped. Heatspreaders straight out of the box sink in the center, so method 5 can leave a large area not in contact if the layer is applied too thin. When you remove the heatsink by pulling straight up off the cpu, it is easy to see whether the heatsink is making contact with all of the paste in the middle.

        The problem is exacerbated by thicker TIMs.

  • Pingback: Ferrofluid and Hubsinks for HOT Direct Drive Hubmotors | ELECTRICBIKE.COM

  • I would have liked to see inclusion of really cheap thermal pastes bought directly from China via eBay or AliExpress. Obviously they won’t be applicable solutions for everyone but my laptop’s CPU is only 35 W and it isn’t overclocked.

  • Matheus

    Hi, der8auer,

    What a great job you did here. I’m particularly interested in such results for a total different application than CPUs, could you please tell me what temperature did you achieve in the probes? (As I can only find the deltas listed).

    Many thanks!

  • Pingback: Bagaimana cara memakai Thermal Paste pada CPU dan GPU yang baik | Senam Jari Blog's

  • Luis

    You guys never tested paste from Cryorig CP5, CP7, and cp15

  • Pingback: The Best Thermal Paste in 2017: The 7 Best Thermal Compounds Reviewed

  • Pingback: I performed brain surgery on my CPU to lower its temperature 15 degrees | Yar Info

  • Fallout season was working after using the same motherboard in all rounds.

  • Hi do you know that you can increase your conversion ratio
    couple of times and earn additional cash every day.
    There is incredible landing pages tool. It’s very easy even for noobs,
    if you are interested simply search in gooogle: pandatsor’s tools

Leave a Reply

Your email address will not be published. Required fields are marked *