Grinchy: well, strictly speaking - less then 16 months, "At one point this project was completely restarted because the test system motherboard failed, which rendered months of progress wasted", but of course, it was great work nevertheless. Pitty it's results seems a bit worthless to me as actual means to objectively choose/buy best performance TIM pastes because of imho a bit flawed testing methodology allowing too big margin of error very possibly retossing TIM performance placement order.
To me manufacturer advised curing times looks like moot, as by simple peak on skinneelab result charts showing temp drop by cure-in time speed of curing AS5 (with advised 200hours) already drops to almost nothing after 8 hours, similar to IC Diamond to which 2hours curing time is recommended and speed of curing of Shin-Etsu X23-7783D (no curing time or special application suggested) isn't that much different either with settling down in arround 5 hours. So to sum it up - it's not sufficient for just one hour of curing even if manufacturer hasn't specified, and overkill even if he specified insane time like 200hours. Extrapolating seen cure speed drop for AS5 even longer i doubt to find total improvement after 12h being more then 1-2C after 200h. It's also easy to see why Shin-Etsu MicroSi G751 outperformed X23-7783D if you used for both cure time of 1hour, as even by
manufacturer's data X23 should outperform G751 after solvent evaporation (curing?) so yet again wrong comparison/testing because of wrong cure time?
In application methods pics to me it looked as too much paste used, but that should matter a bit less if you were consistent with that ammount for all pastes, though i suspect bigger chance of possible +/- error (in light of one test mount):
I reread testing methodology once again, but somehow never noticed anything being done to eliminate mounting error (eg. like Vapor's 5 mounts minimum per paste with dropping best and worst results and averaging rest) though i maybe wrongly misunderstood "our tests were conducted between five and nine times after complete power down thermal cycles" was not just retesting of same mount to compensate possible ambient temp fluctuations and imperfections of motherboard's temp sensors, but actually complete remounts like Vapor did? It's very easy to botch some mount to drop performance by 1-4 degrees, how one can trust results where top half of 40 paste results are within 1.05 degrees if there is only one mount per paste?
Judging by other tests i kind of expect Coolaboratory's Liquid Metal Pro to perform like Indigo Xtreme. I can understand if someone makes some uber wonder paste somehow beating it in performance .. but seeing in this roundup 23 pastes beating it while in tests performed by others it's superrior to all of those that beat it makes very hard to beleave in any credibility of results of this TIM roundup in general.
Hmm, you claim your results are consistent to those of Vapor's? In what way?
Vapor's o/c&l/c a64:
IX -.7 | X23-7783D baseline | AC MX-2 +.67 | AS5 +3.48 | CM IceFusion +6.86
Vapor's o/c&l/c i7 testbed:
IX -1 | X23-7783D baseline | AC MX-3 +1.1 | IC Diamond +1.2 | AC MX-2 +1.3 | AS5 +1.7 | AS Ceramique +1.8
Yours(?)80 TIM paste roundup:
AS5 -.20 | X23-7783D baseline | AC MX-2 +.65 | IC Diamond +.65 | CL Liquid Metal Pro +.8 (which i expected to perform on par with Indigo Xtreme and i've seen outperforming all the other pastes in other reviews) | AS Ceramique +.8 |
P.S.
Few suggestions for future tests for imho more credible results for your next roundup:
*) imho it's worth to use more heat dissipating cpus (probably these days such would be heavily overclocked and overvolted i7 with HT on+whatever that "heats up" cpu most, which probably is LinX?) to more clearly differentiate paste performance (should ease lessening error impact on results imho). Used in roundup core 2 Quad with relatively mild overclock just cannot generate heat load of hot owen i7 can be with heat dissipation in some cases even 250-300W
*) More mounts per each TIM to eliminate mounting error (for Vapor difference between average and best was ~ .8. +/- 1.6 degree is big enough difference to totally reorder performance order especially of top TIM pastes differing so little between each other
*) dispite absence of manufacturer cure-in recomendations or specified short cure time rise minimum cure time for any paste to 6-7 hours with charting temp change by curing to see if extra cure time is needed (check if TIM "settled down" - stopped increasing effectiveness more then 0.1% during hour)
You shouldn't get all my posts as strictly bashing/ranting, i just wish for tests to be actually usuable for making objective buying decisions because of proper testing methodology used for trusty data instead of just wasted time/work on getting results that are to great extent left to game of luck/big margin of errors. After all, in most cases no one of actual user community is unable to redo tests themselves because of lack of time, inability of getting(/buying) all testable samples to test on single test-bed, so everybody is in dire need of trusty 3rd party tests seeing love of manufacturers to game with posted specs (eg. in fan case with noise/performance numbers).
EDIT
*) it won't impact results in any way, but imho any test/roundup that takes that long preferably should be released not "when ready" but as soon as possible, with added extra results of next tested objects later on, so that community can start using work results much sooner, not waiting for completion of giant testing in more then a year. Of course it's preferable to use in first batches some baseline known pastes from previous tests, so that one can estimate not yet tested pastes in comparison to those + add newest/most interesting pastes to first batches, as their performance data might be more interesting to community then proven old legacy products.
Reply With Quote
Originally Posted by Martinm210
Bookmarks