Swiftech Apogee HD Review and Comparison to other Top Blocks

[eXa]

So improvments in flow from XT rev2 is about the same as from XT rev1 to XT rev2...

[stephenswiftech]

So improvments in flow from XT rev2 is about the same as from XT rev1 to XT rev2...

that sounds about right.
But keep in mind that:
* The Apogee HD is optimized to perform very well on a wide range of flow rates.
* Combined with several water blocks (GPUs, chipset, etc) using Multi-Port connectivity the flow rate in the Apogee HD is still going to be higher than the blocks from competition with a classic serial setup.

[meanmoe]

I'm very interested in multi-port options beyond the 2 presented by Swiftech/Stephen

[PiLsY]

Yeah pressure drop is pretty much taken out of the equation, though in some cases you pay a minor temperature penalty on they sub loops. I very much doubt enough to cause any reduction in stability in any situation though. The penalty you pay is having the extra tubing clutter around the cpu block.

[stephenswiftech]

sub loops give you the advantage (especially true with GPU) to have the same inlet temperature in each block. As opposed to in series, each GPU will add up 1-3C. I am preparing a test that illustrates this well.

the extra tubes on the block could be seen as a negative, but keep in mind that going sub loops after the CPU allows for 3/8" ID tubing without compromizing performance.

[jayhall0315]

Posted a different picture of the top and internals so viewers could see them better.


Bds71 - I used TX-2 because that is what Martinm used and I wanted to continue in the same vein and with the same equipment/measuring units, as much as possible. So one could look from his nice graphs (I hate to fool with Excel) and look at the new data and slot it right in.

I will have to look into the IX stuff. My (admittedly limited) past experience with low melting point metallic TIM is that they have been a royal pain the ### and that their cons often outweigh their heat passing properties. I will look into it.

[Church]

jayhall0315: apart from high price IX indeed is close to ideal TIM (well, there are minor cons like it being only for specific sockets, so not usable anywhere, eg. on gpu-s) - almost liquid metal efficiency but without it's cons (at least i hadn't yet heard that IX also diffuses LM-like into IHS/block surfaces), very consistent results between mounts and so on - so i can advise it for personal use. But i don't advise it for use in testing, as apart from it's high price that makes multi-mount tests expensive: 1) it's not what majority of users (which probably testing/roundup is targeted to) will use & 2) as i stated above, it changes block relative performance by making quality of contact less important - as result blocks are ordered by efficiency differently then with TIM (eg. check 2011 cpu block tests of skinneelabs. Different leaders with TIM and with IX).
P.S.
Seeing how some best top TIM pastes got very close to IX (within 1-2 degree) imho no need to pay for IX these days. 1-2 degree is definitely not worth double price and loss of universality (eg. on chipset blocks/gpu-s), so i kind of retract initial advise for personal use too

[jayhall0315]

Thanks for the info Churchy. I definitely try to apply what is easily available and commonly used. For years it was AS 5 and now TX-2. But just for personal curiosity like you said, I will look into it for my home rig. $20 a pop though.... I wonder if they ship it with some caviar as well

[Vinas]

I like the manufacturer mounts. Using the vice isn't real world -- we don't need lab results as much as real world. The stock mounting mechanism plays as much of a role as the original block design so that's what it should be tested with IMHO. But I can see the value in trying to keep the variables uniform too.

yes, I am a real engineer..... no, my degree is not from ITT Tech

^^ that's not convincing anyone

[Vinas]

As to fluid dynamics, I have mentioned this in three or four detailed posts, but not trying to be arrogant, it requires some mathematical background and knowledge of Navier-Stokes, which most dont have. arguing with someone who doesnt know how a Reynold's calculation is done, or what a partial derivative or pde is. Doesnt mean I think they are stupid, just that like so many North Americans, their math/physics background is sorely lacking.

I gotta say, these kinds of posts can seriously damage one's credibility. There are those among you with advanced degrees in physics and mathematics, so I think you can save the lecture and theory, and just get to the data. If people question your methods -- by all means defend it with partial derivatives, etc. That's high school anyway. IMHO it's better to present the data in a clear and concise way. Let the people draw up their own conclusions after that.

[PaganII]

And, yes, .... take 100 average folks off a typical American street and compare them to 100 average Koreans and they would be massacred in math/physics knowledge.

Did you learn this on "Jay Walking"?

Edit for non Americans - http://www.ebaumsworld.com/video/watch/80691723/

[PiLsY]

sub loops give you the advantage (especially true with GPU) to have the same inlet temperature in each block. As opposed to in series, each GPU will add up 1-3C. I am preparing a test that illustrates this well.

the extra tubes on the block could be seen as a negative, but keep in mind that going sub loops after the CPU allows for 3/8" ID tubing without compromizing performance.

Should've perhaps placed more emphasis on "in some cases" in my post. Serial GPU's is never good in any situation, however if you are at the lower end of the pump power scale I can see where a parallel gpu setup in a serial loop will have better gpu temps - as I said though this will make no real world difference and its an unlikely combination. Once you get around 0.25gpm things go south fast (cant speak for the mcw82, only EK and AC full cover blocks). I found the difference between 0.6gpm and 0.3gpm to be 2c at load but between 0.6gpm and 0.2gpm is 23c. That's within the limits of my not at all scientific real world experience . Tri-sli with ek 470 blocks in parallel is a no go with an apogee xt (rev1), heatkiller chipset and 5 x 120 of rads in the loop when using an EK DCP4.0 . Dual sli in parallel runs great. The difference between 0.3gpm and 0.2gpm per block was massive. AC 470 blocks react the same. MCP355 has no such flow issues, also removed the chipset block as its a sandybridge build now anyway.

Anyway, rambled on a bit there, but the short story is if your gpu sub loop gets below 0.3gpm per block it starts getting very inefficient, at least with some popular full cover gpu blocks. Different blocks will have different drop off points, but they're probably all somewhere near this value. Traditionally swiftech designs are very resilient to low flow, so ill be interested to see what you guys can do here .

Remember though its worth checking out other manufacturers blocks to see how low a flow situation they can tolerate, or at least request the customer to do this before use and check their flow rates in sub loops. Someone is bound to attempt drip level flow rates in a sub loop. You should probably recommend a minimum flow rate for sub loops or something .

[jayhall0315]

Vinas - If you check in the middle of the review, there are numbers for the top three blocks with the manufacturer's OEM screws and backplate. I tested it both ways (uniform 240 N clamp and stock OEM parts).

As to the other comments, you dont have to believe me (of course you could look up my homepage at the university), ...... test it for yourself. Like any good skeptic, repeat what Martinm did (and I followed that pretty closely but with a bit different equipment) and verify for yourself. I only mentioned the fluid comment previously, because any discussion I have seen here at XS concerning fluid dynamics tends to quickly get muddled. It was just an observation, although in hindsight I can see how it appears I am being arrogant. I can say that a bit of attitude has probably seeped into my comments within the last year or so because I have seen several products where apparently there was either no or very little testing done before they were released (ie...they were designed in CADS, the files were sent to Taiwan or China, etc... and then they started selling the stuff). I can think of three notable examples off the top of my head. I actually dont think there are as many people with advanced math/physics knowledge (as you note), in the watercooling world as you may believe. If there were, some of these products would have been redesigned to work properly. I believe this also bears witness in the copycatting of flat pinned arrays from one block manufacturer to the next. That is not calling anyone a cheater or intellectual property thief, but rather just noting the obvious.

Pagan - About ten months back I was reading a review of a research publication about the dismal performance of adult Americans concerning math and science. I believe it was done by academics at Brown University (but I will see if I can locate it). If memory serves, it had some really funny stuff in it. For example, when sample adults in the USA were asked to add 2/3 + 5/7, something like 60% came up with the wrong answer (whereas only about 3% of Koreans and 5% of Japanese get the answer wrong). There were all kinds of hilarious things in their report. Many adult americans were asked to name how many planets were in the solar system and the most common answer was seven, followed by six. I think America has its strengths, like fostering creativity, but......most American kids know way more about Battlefield 3, their Xbox 360 and Lindsay Lohan's latest jail stay than they do about math or science.

[stephenswiftech]

Should've perhaps placed more emphasis on "in some cases" in my post. Serial GPU's is never good in any situation, however if you are at the lower end of the pump power scale I can see where a parallel gpu setup in a serial loop will have better gpu temps - as I said though this will make no real world difference and its an unlikely combination. Once you get around 0.25gpm things go south fast (cant speak for the mcw82, only EK and AC full cover blocks). I found the difference between 0.6gpm and 0.3gpm to be 2c at load but between 0.6gpm and 0.2gpm is 23c. That's within the limits of my not at all scientific real world experience . Tri-sli with ek 470 blocks in parallel is a no go with an apogee xt (rev1), heatkiller chipset and 5 x 120 of rads in the loop when using an EK DCP4.0 . Dual sli in parallel runs great. The difference between 0.3gpm and 0.2gpm per block was massive. AC 470 blocks react the same. MCP355 has no such flow issues, also removed the chipset block as its a sandybridge build now anyway.

Anyway, rambled on a bit there, but the short story is if your gpu sub loop gets below 0.3gpm per block it starts getting very inefficient, at least with some popular full cover gpu blocks. Different blocks will have different drop off points, but they're probably all somewhere near this value. Traditionally swiftech designs are very resilient to low flow, so ill be interested to see what you guys can do here .

Remember though its worth checking out other manufacturers blocks to see how low a flow situation they can tolerate, or at least request the customer to do this before use and check their flow rates in sub loops. Someone is bound to attempt drip level flow rates in a sub loop. You should probably recommend a minimum flow rate for sub loops or something .

great input there thanks, .3gpm seems about right.
We could certainly make a note with a minimum flow (which would be around .25gpm with our blocks based on previous data, still need to update that with more current gpus), though unless you've got a way to guestimate the flow rate, this might not be as useful. Apogee HD + 3 Komodos gave about 1.7gpm in the HD or slightly over .5gpm in each gpu. We could also do some testing with a few blocks from the competition, I should add this to do to list

[PaganII]

Jay, thanks for the review. 1 in, 3 out is a great new addition.

You'd be suprised how many people (anywhere in the world) have no clue where their water comes from and I don't mean the company they pay for their water.

[bds71]

2) as i stated above, it changes block relative performance by making quality of contact less important - as result blocks are ordered by efficiency differently then with TIM (eg. check 2011 cpu block tests of skinneelabs. Different leaders with TIM and with IX).

that was kind'a my point it takes mounting pressure and TIM application (two of the largest variables in this game called W/C'ing) out of the equation and tests **only** the ability of the block to do its job. granted, since most W/C'ers don't use these types of costly products, TIM useage is more applicable to the general public. what it does (in my eyes) is tells you the performance limitations of the block in question - for those who DO want to go to those levels. personally, i can't understand how so many people would spend upwards of $500-1,000 for a custom loop that might get a 5C dT (water).....but then scoff at the notion of spending $20 on a product like IX which could lower system temps by another 2-5C.

P.S.
Seeing how some best top TIM pastes got very close to IX (within 1-2 degree) imho no need to pay for IX these days. 1-2 degree is definitely not worth double price and loss of universality (eg. on chipset blocks/gpu-s), so i kind of retract initial advise for personal use too

the top TIMs under Skinnee's admittedly "extreme" contact (read: pretty much unobtainable by the average user since he created a custom block specifically for doing those tests) settings, yes - TIM's do get pretty close. but, when compared to his "typical" settings....not so much - and these are the *best* results most users can expct to get. otherwise, this is very true (with regards to using regular TIM on other products) - and no, i would not expect most testers to use IX. it is very expensive (though you would not necessarily need to do multiple (3-5) runs to average out your results since they would all be the same)

[Martinm210]

So how the heck did a good technical discussion digress into cultural incompetence...

Back on topic...

[Philip_J_Fry]

nice test. quick question, i am moving from i7 920 LGA 1366 to LGA2011. don't 2011 and 1366 have the same mounting holes?
but according to swiftech:

http://www.swiftech.com/apogeehd.aspx#tab1
"A socket 2011 spring-screw kit will be available shortly."

why is that?

[Boulard83]

I just ordered one of these. Lets hope ill be happy with it

[stephenswiftech]

nice test. quick question, i am moving from i7 920 LGA 1366 to LGA2011. don't 2011 and 1366 have the same mounting holes?
but according to swiftech:

http://www.swiftech.com/apogeehd.aspx#tab1
"A socket 2011 spring-screw kit will be available shortly."

why is that?

the hole-to-hole dimensions are the same, but the hole diameter is larger and most motherboards are shipped with a top plate with M4 buts. We made a new set of mounting screws that bolts on the board directly - no additional back plate is required.
all Apogee HD owners will have their LGA2011 hardware. If they bought the block before we started shipping the LGA2011 hardware we will send it for free.
This same hardware is also compatible with XT, GTZ, etc.

[fantomau]

so out of those 3 blocks, The Raystorm is the better choice, Because thats the one I plan to get, but using red leds instead.

[SonDa5]

Good review made better with Swiftech matching data.


I'd like to see the some tests done in Series and parallel configurations. Just a simple test going from (CPU+series device), (CPU+GPU+Chipset) and finally (CPU+GPU+Chipset+VRM).

Besides the high performance CPU cooling the HD's most important feature is its ability to run the series and parallel configurations. I'm very interested to see how flow rate, head pressure and individual device block temps are all effected in different configurations.

[jayhall0315]

I rarely do testing (this article was a rare exception) SonDa5 because it takes too much away from our regular research schedule. But in this rare case, just bc I am interested and can work on it in the down time, I might agree if Gabe or Stephen ever send me the new MCR drive multiport radiator. So far Tiborr offered to send me the EK Supreme HF jet plate and Koolance offered to send me the 370 (which, I will include later this week), but perhaps Swiftech feels more comfortable with someone like Martinm, since they have known him longer, etc.

Cant test what I dont have.

[SonDa5]

jayhall0315 even if the tests were done without the MCR drive multiport radiator I think the HD will work in a regular loop. This could be the master builder missing link to help unecessary tubing, extra radiators and connections to cool more devices. Very innovative and unique idea.

[jayhall0315]

Added the results for the Koolance 370 cpu block. Which I like, but not as much as the XSPC Raystorm.

SonDa5 - The original post to start this thread is testing only with the Apogee HD in the single inlet/single outlet config.