[Review] Indigo Xtreme vs. AS5, MX-2, IC Diamond, Shin-Etsu X23-7783D

[Gimmpy224]

Awesome Review as always Vapor! Was a great read and also answered quite a few questions I has wondered about over the years but was too lazy to look into ^_^.

TY again sir!

Ill be lurking around like my usual self to see when you and Skinee post more .

Lookin forward for more!

[SergeantNathan]

I've known for years that heatspreaders' MAIN function was to protect the core from chipping, rather than, well, spreading the heat. It wasn't until 2 years ago I realized how LITTLE the heat-spreading function mattered, when the drop/line methods of TIM application became more heavily suggested. I initially rebelled, thinking the old tired'n'true method of spreading an even coat was the best way, and that a drop in the center was only for time-constrained, corner-cutting, sloppy, factory-applications. Then I saw the positive performance results and dug for the reasons WHY. It then became obvious that it was all about creating the most efficient path from the core to the heatsink... the heatspreader and TIM were just less-efficient materials in the way. As long as the TIM covered the area right above the core(s), that's all the TIM you needed... getting the tiny little extra thermal capacity off the heatspreader's perimeter was not worth the risk of making the TIM layer over the core(s) thicker than necessary.

Okay, this is where I need community input... why do some applications still suggest spreading? Take Liquid Pro (and now, to a lesser extent, Liquid Ultra) for example. If it's so hard to spread, why not just do the dot-in-the-middle approach? Has anyone tried it? My theory is that because of its liquid nature, it very easily molds into the thinest, most optimum interface under pressure, so the sole downside I could imagine of spreading, which is a too-thick layer of TIM, is minimized, with the benefit of that little bit of extra thermal capacity of the heatspreader is now bonded to the heatsink. If my theory is correct, then more runny pastes in general, as long as both heatspreader and heatsink are well finished (if not well finished, thicker pastes would be needed), are best applied with the spread method.

That also makes me ponder how well IC7 would perform if thinned with 99% isopropyl alcohol, applied in a very thin layer, mounted, and then given a chance to have the alcohol burn off. Every IC7 application I've seen, despite mounting pressure, seems to leave a too-thick layer.

And a side question... anyone know how much "spare area" is on a die? I've chipped the corners on many die's in my years, and only one I can remember dying as a result.

[Panther_Seraphi]

Well i killed a pretty special 8600GS by chipping the core on the corner. You could barely see the damage but it would only boot 1/10 times if you were lucky and the pictures were scrambled

It ran 740 on the core from the stock 540 with no volt mods being done.

Wish i still had it to play with.

[SergeantNathan]

I played with some 99% isopropanol and some IC7 (which, according to the label, "Contains 92% diamond by weight"). The tubes are the original revision (I bought them within a few months of release), which supposedly is thicker than more recent stuff... combined with the fact they've just been sitting in my bin'o'TIM for awhile, this paste is about as thick as it gets. It might even make do as a high performance thermal pad for thicker gap.

The paste, with a little bit of mechanical mixing, thins well in alcohol. I didn't test it's performance, but thinned out sufficiently, it spread really well under pressure (I used a flat piece of glass on top of a heatsink and watched it spread)

I tried a test on a rougher heatsink, first with just a water drop, then with some thinned AS Ceramique (which was harder to mix, BTW), and sure enough, the liquids followed the grooves, leading to an uneven application. Furthermore, after the alcohol burns off, the paste would probably decrease in volume enough in the valleys that performance would be compromised. That's the theory, anyway.

So, assuming a well finished AND flat surface (on both interfaces), I think thinning pastes may be the way to go for maximizing performance. Due to the very quick burn-off time, it's not practical for TIM companies to use alcohol as a carrier substance, so we'd have to thin it during the actual application time, and even then, work quickly. Due to IC7's low percentage of carrier substances, I think it makes an excellent test subject
for "thinning".

I use the word thinning carefully though, as it's all about temporarily suspending the best performing materials (be it diamond or metal or some ceramic or whatever) so that they flow into all the tiny crevices, rather than dissolve any existing carrier substances (which may be a side benefit, but it does complicate mixing, as shown by the more difficult to mix Ceramique). ie Temporiarily replacing the existing carrier substances, rather than improve them. (Maybe super fine diamond powder in pure alcohol is best then?)

So, anyone care to test any of these theories? I don't have the equipment to do it myself.

Sidenote: it's hard to work with out of the tube, but slightly thinned, IC7 DOES make an excellent metal polish (shined up a few heatsinks really nice)

[Vapor]

I really like the thinned ICD idea....a bit of a results spoiler but it does poorly with great contact due to a high bondline thickness. I'm going to try 99% isopropyl and ICD mixed (what ratio did you use? any more details on how you mixed it?) and maybe release it as a bonus result between rounds (round 1 is nearing completion and is at its max of 7 TIMs, round 2 is slated to have a max of 7 TIMs as well).

I also figure thinning it would make it cheaper--part of the reason it requires such a large bead in application is because it is so thick. By thinning it, not only are you diluting it (such that 1g of paste is no longer 1g of ICD), but you'll probably need to use an even smaller bead because it will spread better. At ~$4/gram and large quantities required per application, it's not exactly an inexpensive paste.

I have figured out how to minimize its polishing characteristics (I remove it with ArctiClean and dabbing rather than wiping it off dry like I did last time), so I'm not as concerned about altering the surface as I was initially.

[SergeantNathan]

Ratio was varied. Both times I used around a rice-size of ICD, but the first time I just dripped a drop of alcohol from the cap, which created somewhere between a very thin paste and a liquid suspension. The second mix I just dipped the tip of the very small flathead (that was also used to mix) into the alcohol, and mixed the little dribble on the flathead in with the TIM. This resulted still in a definite paste, albeit a very thin one, akin to, say Turniq TX-2. But due to the alcohol's rapid evaporation, it seemed to be safe to err on the more side: within 1 minute, most of the alcohol was gone, and the thin paste became something similar to G-751, and within two minutes, I was close to the original consistency. Due to such a short time frame, the mixing should probably be done right on the heatspreader, with the heatsink ready to mount immediately.


Also, terminology note: we should all refer to IC Diamond shorthand as "ICD", like Vapor does, rather than "IC7". I called it IC7 simply because others did, but the "7" on the label merely means "7 carat"... and they also have a 24 carat tube... so, yah. Knowing is half the battle'n'all...

[skinnee]

Knowing is half the battle'n'all...

Yo Joe!


^ sorry, couldn't resist the flashback to my childhood.

[SergeantNathan]

I was assembling a system for a client today, so I took the opportunity to play with my ICD-thinning idea. The CPU was an AthlonX4-640, lightly OC'd to 3.3GHz. I used the stock 1055T heatsink that I had lying around as it's better than the stock 640 heatsink. I plugged the fan in with a molex-to-3-pin adapter to ensure the fan stayed constantly at 100%. First was a quicky application of NT-H1 (since it supposedly has minimal cure time, and I had little time to do this little test). Stressing it then letting it cure overnight, I took the max load temps. When I dismounted the heatsink, I found the bondline to be a bit thick, even though it spread enough to cover the core sufficiently.

I spent a little more time applying the ICD. I used around the same amount of material as I did with the NT-H1 (about the size of a grain of rice), but dipped a tiny flathead in the 99% isopropanol and mixed in the droplet with the ICD. It was a bit thinner than the NT-H1 at that point... I didn't want it much thinner, as the finish on the heatsink was very rough. I quickly mounted the heatsink before the alcohol burned off, applying extra pressure to ensure a good mating. The results were impressive. A 6 degrees Celsius drop (lol... 6 degrees of separation...). I doubt that was merely due to the better performance of ICD alone... the application probably made most of that difference. And this was with hardly any curing time. Dismounting the heatsink showed a thinner bondline and a healther spread than I had with the quicky application of NT-H1.

To prep the system for deployment, I went back to the stock heatsink. After cleaning off the stock TIM, I found the base to be much better finished than the 1055T heatsink. So, this time, I mixed in two droplets of alcohol to a drop of ICD (it took a good 15 seconds or so of mixing to get an even consistency) and quickly mounted the heatsink, as I could see the alcohol quickly burning off. I applied some extra pressure and some torquing back and forth to really mate the surfaces well before locking it down. I was impressed that with this optimum application of ICD, this significantly inferior heatsink performed only 3 degrees worse than the 1055T heatsink with a quicky application of NT-H1 (plus the 640 heatsink's fan was a tad bit quieter).

This wasn't meant as a comprehensive test, just a dabble in curiosity. One hypothesis I have about a downside of this method is the drying out of the paste. This in-and-of itself may not be bad, but if the heatsink shifts at all, there is no liquid carrier left to adjust to the shift, and the mating would probably be seriously compromised. I guess a quick way to test this theory is to torque a heatsink with a Indigo Extreme application enough to "break the seal" and see how much performance suffers. This issue is why you're told never to re-use paste, and why I always use "wetter" pastes on videocard heatsinks (since they tend to be much easier to shift around accidently). Anyone care to test this out?



***Ambient was kept darn well near spot-on 23 degrees Celsius throughout. I only posted relative temps since the CPU temp sensor wasn't calibrated worth crap, listing idle temps well below ambient***


***I should also note something I did with each ICD application that may or may not have improved its results. I put a little dot of ICD on the heatsink, and then poured a dollop of isopropanol on it and then used a lint-free cosmetic pad and lightly polished the heatsink, letting a super thin layer of TIM to remain rather than completely cleaning it off. I then used the same, still-moist-and-traced-with-ICD pad to do the same on the heatspreader***

[wez]

***I should also note something I did with each ICD application that may or may not have improved its results. I put a little dot of ICD on the heatsink, and then poured a dollop of isopropanol on it and then used a lint-free cosmetic pad and lightly polished the heatsink, letting a super thin layer of TIM to remain rather than completely cleaning it off. I then used the same, still-moist-and-traced-with-ICD pad to do the same on the heatspreader***

I usually do this with AS5, but leaving out the isopropanol
Not to recommend for highly bowed surfaces, but works a charm for lapped ones.

[Tom DeLonge]

I'd also like to see the difference without TIM. Maybe there's some compound that's even worst than direct contact heatsink to heatspreader

[Phatboy69]

Hey Ive got 7 sets of indigo xtreme to play with for a 990X and 4 GTX580s all going under Aquacomputer blocks (the cpu will be a silver HF).. and I was wondering..
1. Has anyone actually tried indigo with a GPU successfully?
2. If the CPU temp is maxing out around 90c is indigo going to keep reflowing under those conditions and end up being useless???

Cheers,
Phat

[paulbagz]

Hey Ive got 7 sets of indigo xtreme to play with for a 990X and 4 GTX580s all going under Aquacomputer blocks (the cpu will be a silver HF).. and I was wondering..
1. Has anyone actually tried indigo with a GPU successfully?
2. If the CPU temp is maxing out around 90c is indigo going to keep reflowing under those conditions and end up being useless???

Cheers,
Phat

1. I asked about this a month or so ago and nobody could give me answers, mainly because the IX doesn't match the size of the GPU.

-PB

[Phatboy69]

Yeah but its a pretty close fit, it only needs to cover the area where the die is under the heat spreader anyway. I figure its worth a try anyway! Was just hoping someone else had been brave enough to try it and have some tips..

[jalyst]

^
I thought of that too, but most suggested IX was too specialized for CPU, & that I was best going for something else.
I would love to know the results, compared the best 'more conventional' options.

[Vapor]

I wouldn't even try it. I'm not confident enough in any GPU's clock and voltage throttling to ensure the silicon won't die or be damaged during reflow. On the other end of the spectrum, if their overheat protection measures are overzealous, then reflow won't ever occur--you need the CPU/GPU to continuously run at its thermal limit, not just have the system shut down.

IX will not reflow unless the IHS and the base of the block reach ~75C. With a new ETI between the IHS and the block, your GPU temps may reach 90+ degrees with the pump on but unless the base of the block is really heating up as well (which requires the pump to be off), reflow has no chance of occurring. Can a GPU throttle itself at its thermal limit or will it just shut down the system or just fry itself?

Reflow actually has nothing to do with the silicon temperature (except that the easiest way to increase block/IHS temps so much is to just run the chip). It has everything to do with the temperatures of the base of the cooler and the chip surface (IHS).

There's other issues as well: ensuring the card is perfectly horizontal, most GPU blocks don't have spring-based mounting mechanism so you'd probably have to finalize tightening the block during the liquid phase of the reflow process, and running the most common blocks (full cover) at high temperatures could also have a negative effect on the other components (VRMs would be the biggest concern). Additionally, the plastic frame of the ETI needs to extend past the edges of the IHS in all directions--that's the only protection from having a bead of liquid metal being squeezed out onto your PCB and solidifying there (possibly on an SMD component, likely killing the card).

Then the biggest deterrence, in my book, is "why bother?" Largely because of the typically limited range of voltage adjustment, GPUs don't need every last degree of improvement you can find because they're not thermally limited when you switch to quality WC; they're typically voltage limited.

[Phatboy69]

Thanks Vapour I appreciate the input and agree with your last point about being voltage limited, but I think I'll still give it a go though just to see if a few degrees difference gets me a higher OC.

I will base installation on the instructions for a waterblock installation on CPU but obviously on a GPU.
The waterblock installation method only takes 20secs to complete on CPU so I guess the GPU waterblock will take quite a bit longer due to its size.
I've got a complete spare system lying around, so I'll load up some software and do each card one at a time; Mounted in conventional case so the GPU will be horizontal for reflow.

Installed software will be;
Afterburner and GPU-Z (to underclock the card and monitor GPU/VRMs while reflowing)
EVGA OC Scanner (the patched one allows you to set a critical stop temp and maximum load for the card)
Ill preconfigure all the software with a standard GPU first and do some simulated runs so I get comfortable with the process.

The Process;
I'll finally have a use for that useless iphone spirit level app! Ill use it to check the GPU is level when installed.
I'll start at the lowest clocks/volts/load and work my way up till I hit the magic 90c and keep it there until the temp drops down and comes back up again. (indicating the reflow has completed)
I have a fully self contained TT bigwater kit I can also use just for the reflow to cool things down and check the results before I take the GPU out.
The patched EVA OC scanner will be perfect for this because you can set a critical stop temperature and also a workload percentage for the GPU.
Ive got a few spare in case I have to try a couple to get it right and if it doesnt work after 2-3 attempts I'll just fall back to MX4.

[Vapor]

I hope you consider that card as a luxury and not a necessity....

[dejanh]

I'm curious how this turns out though I fear it will not end well. I certainly do not have the cojones nor money to try it.

[jalyst]

@Phatboy69

So how did you go??
I have some older gear i might try this one soon.

[Phatboy69]

@Phatboy69

So how did you go??
I have some older gear i might try this one soon.

Hi I still havent got the blocks yet, expecting them to arrive late this week or early next.

I have this new gadget for my laptop that will allow me to do the GPU reflow pretty simply too!
Im about to pull the fans off a GTX580 and check the die size compared to the i7 to see if the indigo extreme will fit!
I'll take a couple pics.

Vaio Z128 Laptop +External GTX580 = BIG Grin!


Pulled the fans off a GTX580 tonight to check the die size against the i7 and see if the indigo extreme will fit.

Looks pretty good to me! The heat spreader on the 580 is slightly bigger which is safer than if it's too small! Also the actual GPU die is only about 60% of the size of the heat spreader so the hottest area will get full coverage.

I'll know soon if its a fail or not... I expect my blocks this week and I can now use my laptop with the new external PCIe slot to reflow the indigo extreme on the GPU.

580 with heatspreader removed. (not mine!)




The candidates...

[Captain_Harlock]

Why not consider something like Coollaboratory Liquid MetalPad instead. Many people have used them on video cards, and things like the xbox 360 with great results. You can find them all over the internet, several uses and fairly cheap compared to IX.Not quite as good as IX, but we are talking a minor performance hit. Something to consider.

[paulbagz]

Phat, one thing to consider in the short term future (if the performance doesn't work out) is that IX will obviously come out in a S2011 size application which should be more then big enough for 580 GPU dies

-PB

[CyberDruid]

Great review Vapor. I used Indigo Extreme on a client build last year and I was very impressed. I did three remounts to check on how it flowed. Amazing stuff...it was so thin I just stared in disbelief. I could read the text off the IHS perfectly from the removed IX film. The client provided IX for the GFX waterblock installation but I used ICD instead...my Spidey Sense was tingling...it seemed like a huge risk to load up the GPU with no active cooling.

Like you said in the OP the only thing difficult about IX is that it's a different procedure...but it's a well developed system. Aside from the cost I would prefer it over any other TIM for CPU mounts.

[chaos_being]

Yo Joe!


^ sorry, couldn't resist the flashback to my childhood.

Phatboy, good luck with that IX and your GPU's. From what I've seen that stuff is great, but the reflow process...kind of scary.

I just ran out of MX-2 on my most recent build, I've been away from PC building for a while, looks like I may have to try some new TIM next time!

[jalyst]

<SNIP>

Thank for sharing this mate, I look forward to your findings!!!