8 mounts Swiftech XT vs 8 mounts HK 3.0 all copper

[rge]

I am leaving in a couple hours, quick stopover then on to Hawaii until end of this month….so unfortunately wont be able to respond to anyone for few days at least, unless you ask in next 2 hours….though may or may not log in from vacation in few days if any down time and reasonably sober...was going to wait til got back...but why not...

(After mounting HK 3.0 about 20+ times testing mx3 vs mx2 and then different amounts tim...before finally realizing there is very little difference between 2... I got to the point I can keep most of my mounts within 1C of each other, majority of time within .5C, and usually you know while doing it whether you did a good mount or a questionable one)

8 mounts of Swiftech XT versus 8 mounts of Heatkiller 3.0 all copper version

TESTING METHOD

--TIM MX2, measured between 6-7mm pea size for all mounts (that size gives me best/most reproducible mounts), and nice thin coverage across entire IHS.

--Same mounting pressure for all Swiftech XT and same mounting pressure for all HK mounts , ie Swiftech all the way down, HK 8 turns each (calipers to verify, I used slightly tighter than suggested specs but my cpu lapped, approx 2 turns from max all way down). Always 1 turn at time in criss crossing pattern to get even mount. Note my 8th mount of Swiftech was with HK mounting system (modified) so I could vary pressure and recheck, since using lapped cpu.

--All tested on Lapped i950, OC 4.4ghz, 1.34 cpuz vcore at load,

--NOTE...since using GB board, one edge of HK and Swiftech block I flattened the edge so both mount vertically without touching caps. Both, however, I had mounted before flattening edge with same results…just gets annoying pushing caps out way each time, and affects the ability to get good repeatable mounts.

--Used prime ffts large, same fft on each run, monitored heat dissipated with everest all between 178 and 181W, 10 minute warmup (by 6-8mins water temp is already maxed) and used Realtemp to then record next 3 or 5 mins (half of each though results same with either) at 1 second intervals so all 4 cores of data to then average. There is no difference between 3 and 5 or even 30 mins of data, will be within .2 to .3C within same mount...as long as keep ambients from changing during measured run...hence smaller runs are easier to keep very accurate.

--Same fluke thermocouples to measure exact same 1 location in water temp before pump, and thermocouples placed in front rads to measure intakes.

--I am in a 25 x 23 x 12 feet high ceiling room, with computer expelling all hot air in a channel of styrofoam so can not recirculate until travels long distance. A/C off during run and on between runs, otherwise a/c if click on during test will drop ambients quicker than water temp can respond, screwing up delta and measurements. Point is to keep ambient within ~.2C during entire 3-5 minute measurement run, and hence water temp within .1C during run, otherwise start run over.

--All temps corrected to 24.7C ambient, ambient ranged from 24.3 to 25.2, most were between 24.5 and 24.9C.

--All tests done with my 1 large loop, 2x360 rad equivalent (pa120.3 +mcr 220 + mcr 120), 355,pump xspc top, fans all maxed (on controller) GT 1700 rpms pushing, yates UV 1600 pulling, gtx 295 in loop (which btw raises air to water delta only 1C vs it being out of loop...since they downclock at idle)

pic of Swiftech apogee XT installed (flow is pump-mcr120 front-pa120.3 top-cpu (middle is inlet)-gpu-mcr220 rad-reserv-pump)
http://img397.imageshack.us/img397/2897/swiftechxtpst.jpg

pic of Heatkiller 3.0 all copper installed (flow exact same as Swiftech XT)
http://img162.imageshack.us/img162/1978/hkmntspst.jpg


Results:

--If averaging all 4 cores, HK 3.0 is 0.8C lower core temps than Swiftech XT ON MY CPU.

(averaged best 5 of 8 mounts, does not change outcome if use all 8)

Swiftech XT
71.3, 71.6, (72.4 excluded), 71.5, (71.9 excluded), 71.2, (72.1 excluded), 71.3 (included this, though with modded heatkiller mount to increase pressure since had lapped cpu)

HK 3.0 all copper
70.7, 70.8, 70.4, (71.3 excluded), 70.6, (70.8 excluded), (71.1 excluded), 70.5

--However, unlike HK 3.0 versus earlier GTZ, where there was clear 3C lower temps on all cores, this was interesting:
--If just average core 0 and core 2 (ignore core 1 and 3) Swiftech XT was ~0.5C lower ON MY CPU on nearly all mounts

--If just average core 1 and core 3
Heatkiller was ~2C better on each of those 2 cores on nearly all mounts

--Interface differences seemed to exceed overall block differences.

Example of Swiftech XT in all mounts gave cpu temps in ~ following ratio
70.5, 74.5, 70.5, 70.2, ie core 0,2,3 ~same, core 1 was 3-5C higher

HK 3.0 always for each mount gave ~ ratio
71, 72.5, 71, 68, ie core

Very end of typical Swiftech XT run, note typical ratios of core temps seen in all its mounts.

10/11/2009 10:40:49 4416.07 71 75 70 71 38 100
10/11/2009 10:40:50 4416.07 71 75 71 71 38 100
10/11/2009 10:40:51 4416.07 70 74 71 70 38 100
10/11/2009 10:40:52 4416.07 71 75 71 71 38 100
10/11/2009 10:40:53 4416.07 70 74 70 70 38 100
10/11/2009 10:40:54 4416.07 71 75 71 70 38 100
10/11/2009 10:40:55 4416.07 70 74 70 70 38 100
10/11/2009 10:40:56 4416.07 71 75 71 70 38 100
10/11/2009 10:40:57 4416.07 70 74 70 71 38 100
10/11/2009 10:40:58 4416.07 71 75 71 71 38 100
10/11/2009 10:40:59 4416.07 71 75 71 70 38 100
10/11/2009 10:41:00 4416.07 70 74 70 70 38 100
amb 24.7 71.5 avg 4 cores 29.5 water
70.47 74.60 70.57 70.29 avg each core




Tail end of typical HK mount, with typical core ratios see in all of its mounts (the 38C is gpu temp, ignore it)

10/11/2009 23:06:43 4416.08 71 73 72 68 38 100
10/11/2009 23:06:44 4416.08 71 72 72 67 38 100
10/11/2009 23:06:45 4416.08 71 73 72 67 38 100
10/11/2009 23:06:46 4416.08 71 73 72 67 38 100
10/11/2009 23:06:47 4416.08 71 73 72 68 38 100
10/11/2009 23:06:48 4416.08 71 73 72 68 38 100
10/11/2009 23:06:49 4416.08 71 72 71 67 38 100
10/11/2009 23:06:50 4416.08 71 73 72 68 38 100
10/11/2009 23:06:51 4416.08 71 72 70 68 38 100
10/11/2009 23:06:52 4416.08 71 73 71 67 38 100
10/11/2009 23:06:53 4416.08 72 73 72 67 38 100
10/11/2009 23:06:54 4416.08 71 72 72 68 38 100
10/11/2009 23:06:55 4416.08 71 73 71 68 38 100
10/11/2009 23:06:56 4416.08 70 72 72 67 38 100
10/11/2009 23:06:57 4416.08 71 73 71 68 38 100
10/11/2009 23:06:58 4416.08 71 72 72 68 38 100
10/11/2009 23:06:59 4416.08 71 73 72 68 38 100
10/11/2009 23:07:00 4416.08 71 73 71 68 38 100
amb 24.8 70.7 avg 4 cores 29.8 water
70.97 72.54 71.59 67.79 avg each core





Impression:
1) My cpu is no question better with HK by 0.8C, but I would have to test it on several cpus before I would call it one way or the other, because interface differences, ie relative core differences, exceeded difference between 2 blocks on my cpu.

2) I would not be surprised if reviews will be mixed, especially if flow is optimized for one or other, especially given the interface ??bow?? differences between blocks hence some cores cooler others hotter, or I could be entirely wrong...have to wait and see.

3) Also dont know if lapping affected Swiftech results, need to test a non-lapped one, though I did use the different mount and walked it down very tight watching temps, got a good mount, but not better than Swiftech mounting.

3) If retesting the same mount over and over, ie loading, let cool, then again, get 0.2 to 0.3 max variability, usually less, so accuracy is only about .2 to .3 at best just accounting for measurement accuracy alone.

4) Without accurately measuring intake ambients (and if possible water as well for control/double check), and controlling the intake ambients while you test, you wont be able to tell difference in temps between two blocks, unless you have way better eyes than me.

5) Using the heatkiller before and after lapping, I got the exact same temp ratios between cores BEFORE AND AFTER LAPPING, not to mention the exact same temps. But again, my particular IHS was very flat to begin with, but interface differences ?bowing diff are interesting.

6) The Swiftech mount system was much faster each time, and way easier on my thumbs. Since I have gotten used to using the wrench and counting turns on HK, dont mind it quite as much, but for unscrewing, just wish they had larger thumbscrews...ends of my finger/thumb are raw.

7) hopefully I got everything posted up ok, sorry if little disorganized and not graphs, could have easily done some if had more time...and my 2 hours is now down to less than 1..

[ricflairw000]

sweet thanks!!!
i was wondering how it would stack up

[Boulard83]

very nice ! THX alot !

[Daddyjaxx]

The aesthaetics of the Heatkiller is just better than Swiftech. 2-3c normally means little to me. Price is number 1, availabilty, and then looks. If I could get an XT locally at MC and be messing with it in 15 minutes, I'd get it over the HK. But the HK looks so much better.

[LtBlue14]

Thanks very much for the rigorous testing. It is much appreciated.

70C temps with an equivalent of two triple rads and the best CPU blocks surprises me, given that the 295 was idle. Shouldn't one expect lower?

[Conumdrum]

The aesthaetics of the Heatkiller is just better than Swiftech. 2-3c normally means little to me. Price is number 1, availabilty, and then looks. If I could get an XT locally at MC and be messing with it in 15 minutes, I'd get it over the HK. But the HK looks so much better.

Hmm
For me it's temps, then price, and looks are wayy last. It's just a computer, not a dancer at a strip club.

[Otter]

Thanks for the data. :)

[jediman]

i would guess lapping negatively affects the swiftech. from what I have read, the swiftech mounting system is very precise but not tweakable for the case when the cpu is slightly smaller than stock causing insufficient pressure.

[hokiealumnus]

They reportedly compensated for lapping with this block and it can be cranked down a little bit tighter than with the GTZ mounting kit.

[Conumdrum]

Nice to see some validation. Pretty close to the data Gabe did considering the variances.
Yea, it's not perfect data, but you can see the similarities. Gabe only had a HK3.0, the copper top one is better and it was used in the above test.

........................................
From XS, cut n paste

Intel® Core™ i7 920 (Bloomfield D0)
(@4095 Mhz - 1.392 V Actual)
Water-Block............................................. ........Delta T (*)
Apogee™ GTZ............................................... ...40.82 °C
Apogee™ XT (factory inlet configuration)...............37.88 °C
Apogee™ XT (alternate inlet configuration) (**)......37.99 °C
Reference competitive data (Heatkiller rev3.0).......39.39 °C

[Gilgamesh]

*cough* i call Gabe's numbers BS, but he wouldn't be a good marketer if he was honest :)

[rge]

Well after settling back from vacation....I played around with the XT block a little more. I have mounted the HK vertical and 180 vertical (upside down) and get same temps either way. I prefer with writing upside down since outlet is then on top and prevents trapped air, but temps are same and ratio of core temps same.

After getting the weird different ratio of core temps with Swiftech XT, ie slight decrease on two cores and few C higher on other 2 cores, I rotated the block 180 degrees and got yet a third pattern of core temps. Which means something to do with block itself. Took it apart, and pretty sure I know why half of the block cools different then the other half. It is not the outside surface, but the inside surface that is clearly different between two halfs. Under a low powered microscope, looks like top half milling went well, bottom well...its different...some areas partially cut, best pic I could get, but wish I could get one of top and bottom half via scope. Both the pattern and depth of cuts are different between the two.

http://img41.imageshack.us/img41/2259/pinsswxt.jpg

Here is blow up of section of top half block, then blow up of bottom half:
http://img195.imageshack.us/img195/5281/swxt2.jpg

[H36]

Hey rge, I've got two theories on your weird core temperature profiles.

1) the socket latching mechanism (specifically the T-25 screws at the top of the socket) are preventing the block from seating properly on a lapped IHS. I've detailed it more here: http://www.ocforums.com/showpost.php?p=6273385&postcount=4 This theory doesn't fit with the third profile, however.

2) The backplate might be defective, I've received two so far that have the following issue: the bottom of any number of the PEM nuts is too 'tight' and prevents the thumbscrews from tightening down all the way via hand tightening. Via screwdriver tightening, it half-strips the screws (http://img21.imageshack.us/img21/8577/halfstripped.jpg) pretty nastily :( One backplate was directly from Swiftech and the other came with an Apogee GT LGA1366 adapter I bought recently.

For #1, removing the socket assembly would be the surest way to test, albeit a bit of a pain.

For #2, if you see the thumbscrews not screwing down all the way to the board, that's a sure indication. Unfortunately, screwing down all the way with a screwdriver will partially strip the screws. Rotating the backplate 180 degrees and seeing if the core temperature profile flips is another way I guess.

[Spawn-Inc]

ya i'm a bit disappointed but also understand swiftech's answer for the milling.

they accept that as a fully working block and passes quality control just fine. but it's not exactly perfect. but at the same time it's extremely hard to cut such small pins, hence the mess ups.

[rge]

When i spin the block around 180, the two hotter cores were cooler, and the cooler ones were hotter giving a third pattern, so should be either the outside of block or inside. After seeing the inside of the block, that is my best guess. The tim pattern looks ok, dont think screws are interfering, but yeah taking the socket out is too much of pain for now.

And because I lapped my cpu, I did use a screwdriver and tightened down as much as possible.

If milling was not so different from top half to bottom half, I would give my wifes i940 (non-lapped) a whirl, and will probably at some point, but seeing that difference was like "no wonder" for me, so lost my testing enthusiasm for now.

[Spawn-Inc]

since it's lapped i would look into that before the block. gabe said the small imperfections don't effect it's performance.

the block is meant to bottom out so it's not really a fair test.

[baditude_df]

My new XT does not yield any abnormal results in regards to individual core temperatures. In fact, I've never had such a close grouping b4. Max temperature deviation between cores is an avg 5 degC. Currently sitting atop an unlapped i7-950.
The cores do tend to pair up with one pair ~2 or 3 deg higher than the other, but that's no different than the GTZ, or my EK Supreme for that matter.
All blocks with intentionally convex bases seem to exhibit this type of behavior.
http://img35.imageshack.us/img35/292/loadrosetta42ghzxtd.th.gif (http://img35.imageshack.us/i/loadrosetta42ghzxtd.gif/)

[rge]

One day when my wife wont be around for day or so, I will try her unlapped i7940, best way to be 100% sure.

baditude, did you see same pattern of core temps on all your blocks?

[x509]

Hmm
For me it's temps, then price, and looks are wayy last. It's just a computer, not a dancer at a strip club.

Agreed! :thup:

I'm going to actually spend the extra 20 bucks for the optional no-window side for the Corsair 800D case. I figure that my wife hardly cares what the inside of a computer looks like, and I only care that it's neat-looking and dust-free. :salute:

I also figure that the solid metal side is better for sound-proofing and for heat dissipation.

[baditude_df]

baditude, did you see same pattern of core temps on all your blocks?

My GTZ tended to have a larger temperature gap between the pairs, and temperature spread from lowest to highest(maybe imagination). EK Supreme was comparable to the GTZ.

The XT yields noticably better temperatures, and seems to be a better overall build than the GTZ or any other block I've used to date.

Here's some screencaps from a comparison I did the day I installed the XT.

Idle XT
http://img689.imageshack.us/img689/6916/idlextcputemp.gif
Load XT
http://img196.imageshack.us/img196/8093/loadxtcputemp.gif
Load GTZ
http://img197.imageshack.us/img197/5176/loadgtzcputemp.gif
Some more of the testing here:
http://www.ocforums.com/showpost.php?p=6293153&postcount=49

[rge]

Thanks baditude...your ratios are similar on two blocks as one would expect, first is highest, then 2 middle, core 3 is lowest.
max temps XT....68, 65, 65, 63
max temps GTZ..70, 67, 68, 65

Since when I rotate my XT 180, core pattern changes, half block is cooling different then other half. I just scribbled temps down as I was in a hurry to check surface with topograph at work after seeing change...but reflection was not good enough to measure surface topography....then looked at insides thinking there might be some obstruction and found a difference there...need to go back and do few mounts with vertical and 180 vertical with non-lapped i940, just to rule out any other variables.

Though I already answered my main question, my HK performs better than my XT on my lapped cpu (HK was ~3C better than GTZ, so both perform better than GTZ), now interested in why core temp patterns change with XT, and why change again when rotating block 180. Clearly something is different between 2 halfs, and only visible difference is the very different milling channels split right down the middle. Top is small pins throughout, bottom is much thicker and some partially cut, almost like channels in some of bottom half areas that would allow water only one way.

[baditude_df]

YEah, that is very odd. And your pictures show it very well.
I understand what your problem is, and agree that the different milling could certainly be the major contributor.
Have you spoken to Gabe about it? He seems to be quite responsive to user's comments and requests.
I notice that the latest in the XT thread over at XS (http://www.xtremesystems.org/Forums/showthread.php?t=236172&page=13) is Gabe pulling random samples and testing backplates based on someone's claims of sub par manufacturing. It's pretty amazing that they first pulled about a dozen or so random samples from Swiftech USA, then checked 200 more at their manufacturing facility just to be sure.

[rge]

At this point it is just more of a curiosity. I thought my block must have been a production error from first few runs when first looked at it, but after googling around, looks like milling defects are pretty common in it. And worst case scenario the defects are only causing a degree or so temp wise, and most wont notice that. Unless, like me, they took time to accurately measure with equipment, and were expecting a degree or 2 in other direction comparing to my HK. But given the milling defects are common...not going to worry about it, just will stick with my HK cu, despite the annoying "thumb screws". Also from varying pressure on my HK and realizing their ideal pressure was not the best pressure for my particular cpu for best temps even before lapping it...I would really prefer a mounting system exactly like Swiftechs, except instead of a hard barrier, maybe some adjustable/removable barrier nut or clip that limited pressure for dummy proof, but then could be adjusted or removed for those who wanted to try varying pressure or for lapped cpus.

[Springbok]

I'm gonna be sure to check my block interior when it arrives later this week. The easier install & prior good experience with Swiftech won me over to the XT. Heatkiller would look good on my Asus Sabertooth tho...

[davidm71]

I bet you would beat the HK Cu temps if you mod the inside of the XT with a washer and silicone that was detailed in the Skinnee labs review. I just ordered one even though I have a heatkiller cu in my draw as I have no desire to completely remove and disassemble my pc. Should be a relatively easy install to swap the Fuzion block I currently have in there and recycle the backplate...

[rge]

It might reduce temps overall...but with half block not milled well and raising temps by few C on one side, not that interested in using it. Even if could send it back, too lazy to do so when I might get similar base or even if not, temps wont be noticeably different, less than a degree either way.

As for silicone, I saw that, I will wait for Swiftech or HK to use 0 ring if it helps:p.

I am back to waiting for next revision of any block that is at least 1-2C better than what I am using. Going from fuzion to GTZ gained 1-2C and ~3C from GTZ to HK. From HK to XT was only non-gain, but fun to test nonetheless. Not that disappointed, sometimes you hit and sometimes you miss when you pounce early, 2 out of 3 is not bad.