TEN YEARS of Water Cooling Performance Tested! - Through The Ages Ep. 1

Linus Tech Tips ·Linus Tech Tips ·2017-05-06 · 1,553 words · ~7 min read
Floatplane YouTube

Transcript

JSON SRT VTT 130
0:00 Are they launching another
0:04 series? Okay, whatever. If you don't like all the series that we have now,
0:08 then just treat this like a standalone video because it's cool either way. The
0:13 idea here was to take topof-the-line products from a particular category. We
0:18 decided on CPU water cooling for the first one, and measure just how far
0:23 we've come thanks to improvements in design and manufacturing. Welcome to
0:29 episode 1 of Through the Ages CPU
0:32 cooling
0:42 blocks. Massdrop is now featuring the
0:46 AKGM220 semi-open studio headphones.
0:49 Check them and other drops out at the link in the video description.
0:54 So, the first step in this undertaking was to gather up CPU blocks from as far
0:58 back as we possibly could. By teaming up with SwiftTech, we were able to scrape
1:03 together seven of their nine blocks going all the way back to July of 2005,
1:09 a span of over 10 years. So, time to get
1:12 started testing them, right? Nah, that would have been too easy. The intention
1:18 was to use a modern test bench to eliminate as many variables from our
1:22 test as possible. And as the astute
1:26 among you might have already realized, Intel's latest flagship X99 platform not
1:32 only didn't exist yet, but Intel wasn't
1:35 even the performance king back then. So, we needed to source modern mounting
1:40 brackets for many of the blocks. Time for some DIY metal fabrication. Step one
1:47 was to glue the hold down plate drawings that Swiftech modified for us onto a
1:52 piece of sheet steel. Then we went to work with the aviation shears, cutting
1:57 the outside edges. We drilled out the middle as much as we could with a step
2:01 drill, finishing off again with the aviation shears. And finally, we center
2:06 punched the holes and drilled through them with a titanium bit, cleaning
2:10 things up a little bit with the Dremel. This was pretty tedious, but we didn't
2:14 want to bother our friends over at Protocase to make them for us if we
2:19 didn't have to. And yeah, so the Apogee GTZ required a
2:25 specific plate thickness and a pretty precise cutout. So, they ended up
2:29 whipping one of those up for us in the shop. So, so much for that. But anyway,
2:33 thanks Proto Case. As always, you guys were lightning fast and the plate fits
2:37 perfectly. Let's introduce now then the test bench that we're using. We used an
2:41 Intel Core i7 6850K 6 core processor
2:45 with a 246 mm squared die and a 140 W
2:50 rated TDP. Though it should be noted that according to this Tom's hardware
2:54 article, it's closer to 100 watts in the real world. This might be important for
2:58 later. The rest of the components, in no particular order, are an ASUS X99 Deluxe
3:03 2, 32 gigs of Corsair Vengeance DDR4, a random GTX 980 Ti that wasn't connected
3:09 to the water loop, making it somewhat irrelevant, and our water cooling gear,
3:12 an Alphacool Nexus XP 45 triple 120 mil
3:16 rad with Noctua industrial PPC fans, a D5 pump with an EKX top, and some 3/8 in
3:23 Primoflex tubing. We used IC Diamond for our thermal compound. So each block was
3:28 mounted using M4 screws with nylon washers to achieve the desired mounting
3:33 pressure. Yeah, we were a little bit approximate here, but as I think you'll
3:38 see from the results, it likely didn't matter much. Idle temps were recorded
3:43 using IDA 64's logging tool after 10 minutes. Then load temps were recorded
3:48 using a 3minut mean average after allowing the water temps to reach
3:53 equilibrium. Room temperature was maintained by our zone controlled HVAC
3:57 system at 23° C. Wow, that was boring. All right,
4:02 let's meet our first contestant. This is the legendary Swifttec Storm. Designed
4:08 by Cathar himself and sold to Swift for
4:12 mass production back in ' 05. It features an impingement design with a
4:17 midplate sporting 35 microjets that
4:20 blasted turbulent water into 35
4:24 corresponding divots in its thick copper base. It was a nightmare to manufacture,
4:30 making it very expensive. And its performance, while superior on bare
4:36 small die single core processors, fell quickly behind as dual core and quad
4:42 core CPUs with integrated heat spreaders became the norm. The story behind 2006's
4:48 Appleo G was less about raw performance
4:51 with 1°ree Celsius being optimistic and
4:55 more about optimizing for enthusiast CPU
4:58 design trends and ease of manufacturing.
5:02 Thanks to the relatively simple uniform
5:05 diamond pin matrix in its copper base, it was less restrictive, contributing to
5:10 better liquid flow rates for multiblock loops. It handled and continues to
5:15 handle large CPU dyes with heat spreaders better than the Storm. And
5:19 thanks to its cheap injection molded acetal top, it was one of, if not the
5:26 first legitimately mass-roduced CPU water coolers. This right here, this
5:31 exact one, was my first block when I got into water cooling about 10 years ago.
5:36 The followup, the Appleo GT, looks outwardly identical to the Apoge, but
5:42 thanks to manufacturing improvements and an enthusiast inspired tweak to bow out
5:48 the base with a thicker O-ring to compensate for Intel's less than stellar
5:51 IHS flatness. It boasts slightly better
5:55 performance, not Swift 1 to3° C from our
5:58 measurement, but I'll comment more on that in a moment. The Apoge GTX,
6:03 released shortly afterward, was basically a hot rod version of the Apple
6:06 GT with the same copper base plate and
6:10 then marketing that was focused more on the cool looking aluminum top
6:14 that Oho ho. I guess that's probably why
6:18 I couldn't find a working one for this video. Don't mix your metals, kids.
6:22 Moving on, the Apple GT Z brought about
6:26 significant improvements in manufacturing that allowed the small
6:31 diamond pins of the Apple GT to be shrunk to only 250 microns for a claimed
6:38 performance gain of 2 1/2°. Holy crap.
6:42 Too bad on our 6850K, it really didn't
6:46 amount to much. But back to my earlier comment about these disparities that are
6:51 starting to pop up. The GTZ design was
6:54 as much about optimizing for the die orientation of Intel's then flagship
7:00 QX6700 quad core as about anything else.
7:04 A trend that we're likely to see continue. The Apogee X-T from 2009 was,
7:10 from what we can measure, the last big
7:13 step forward in CPU water block performance. Sure, it was kind of
7:17 restrictive, but it looked amazing. And
7:20 thanks to its redesigned upper housing with the inlet centered over a much
7:26 larger 250 micron pin matrix, it crushed
7:30 its predecessor with our numbers nearly
7:33 backing Swift claim of a 3°ree improvement. as long as you had a
7:38 whopping $80 to spare. Apogee HD, which
7:42 followed two full years later, rocked tweak 225 micron pins, 30% lower flow
7:49 restriction to improve GPU block performance, and according to Swift, 2°
7:54 better temps. Though, I suspect we'd have to pull a 3960X or something out of
7:59 our hats to observe the specific optimizations that were done to achieve
8:03 that result. I'm actually still happily running a limited edition gold-plated
8:06 Apogee HD in my personal rig. Which leads us then finally, sorry we got lazy
8:12 and skipped the Apogee XL to the Apogee XL 2, the current flagship block that is
8:18 only available as part of Swift's H220 and H320 X2 kits. It features the same
8:24 225 micron fin/pin base plate as the XL,
8:28 but tweaks the inlet and outlet design of the top cover for improved flow
8:32 characteristics. SwiftTech claims about a degree and a half of performance over
8:36 the HD, but we saw a mere half a degree
8:39 improvement in spite of the five-year-old design of the HD on a
8:44 modern processor, leading me to believe that most of the performance to be
8:48 gained with future water blocks will come from compensating for the weird
8:53 idiosyncrasies that Intel introduces with each processor generation rather
8:57 than pure thermal transfer improvements. Though, I will be interested to see if
9:03 the prototype SKF block that Swift Tech showed me at CES will change my mind.
9:08 The fins on that sucker are so small that they can only be produced by
9:13 shaving copper and bending it up into little tiny fins. Maybe there will be
9:18 another leap forward after all. I'll be sure to update you in another 10 years
9:22 or so. So, thanks for watching, guys. If you liked this video, hit that like
9:26 button. If you disliked it, you can, of course, hit the dislike button. But if it was awesome, get
9:31 subscribed, hit the like button. Okay, I did this out of order. Or click that
9:35 link to where to buy the stuff we featured at Amazon in the video description. Also linked in the
9:40 description is our merch store, which has cool shirts like this one and our
9:43 community forum, which you should totally join. Now that you're done doing all that stuff, you're probably
9:47 wondering what to watch next. So, click that little button in the top right to
9:51 check out our latest video over on Channel Super Fun. I guarantee it will
9:55 terrify you.