Temperatures in relation to flow-rates. Data on several CPU-Blocks

[HESmelaugh]

Hi there.

I just completed a roundup of a handfull of CPU-Waterblocks and since users of this forum were a great help to me, answering many questions about how I should conduct the testing, I want to publish some data here.

Since I'm too lazy to translate the entire article (it's in German), I'll pick out the data that I think you might find to be most interesting.

First, a quick word on testing methods:


Test-rig and Testing

The test-rig consists of an open build with the following hardware:
- Q6600 Prozessor @ 3.2 GHz
- Asus P5E WS Pro Mainboard
- Magicool Slim 360 Copper Radiator (with fans @ 1100rpm)
- Laing DDC 1T+ (Swiftech MCP355) Pumpe with XSPC Reservoir Top
- Corsair HX 620W PSU
- T-Balancer bigNG plus Sensorhub
- 13/10 Masterkleer tubing (that's between 7/16" and 3/8" inner diameter)
- 2x analogue temperature sensors (measuring water temperature)
- 3x digital temperature sensors, measuring ambient temperature below the radiator
- Flowmeter Digmesa FHKUC 70
- Arctic Cooling MX-2

For test-runs and logging, I use the T-Balancer software as well as Everst. The Everest stress test is used to stress the Processor. Each test-run consists of a 45 minute warmup and following 20 "test-window". The logged data from all the external and internal data from those last 20 minutes is then averaged out for results. The temperatures shown are the differences between CPU-Temperature and Water-Temperature.

Why only 3.2GHz on that Q6600?
Unfortunately, I could not push this processor any further. No matter what I did with voltages and FSB/Multiplier fiddling, I couldn't go further than those 3.2GHz and get a stable boot to OS. Guess I got unlucke with this CPU.

Why CPU temperature and not hottest core or all cores averaged?
The CPU-Temperature turned out to be the most reliable. i.e. if I do a test run today and another one tomorrow, without changing blocks, TIM or anything, the CPU-Water temperature will be almost identical for both runs while hottest core-water and all cores averaged-water will deviate ever so slightly. I figured I would simply use the data with the smallest error.


Tested blocks
Here's a list of the tested blocks:
- Swiftech Apogee GTX
- Swiftech Apogee GTZ
- Ybris A.C.S.
- Aqua Computer cuplex XT di
- Aqua Computer cuplex hd
- EK Supreme
- Zern PQ+
- Thermaltake "Kandalf" block

I'm sure you're familiar with most of these. The Zern PQ+ is an austrian made block that might not be known in the US and other parts of the world. Here's a picture of the PQ+:


I also included a block taken from the "Kandalf" case from everyone's favourite LC-manufacturer Thermaltake! :-P
I wanted to see how it measured up against "real" waterblocks.


Restriction

Since I don't have the tools for pressure-drop testing, I measured flowrates instead. Here are the flow-rates of the entire loop with each of the blocks (I hope I got the conversion to GPM right):



As you can see, both Swiftech blocks did really well here. Especially the GTZ was suprisingly un-restrictive considering the micro-structures inside it.

I also measured flowrates running the pump at lower voltages, but the results are unspectacular: The loss of flow-rates with lower voltages is practically linear for each block. Graphs will be provided if anyone wants to see them anyway.


Temperature vs. Flow-Rates

But now for the really interesting stuff: I measured temperatures in relation to flow-rate for each of the blocks in increments of 30 litres per hour (that's about 0.13 GPM). The flow-rates were regulated using a ball-valve in order to factor out the influence of the pump's heat dump. Here are the results:



On the graph, flow-rate increases from left to right. The steeper a curve, the more benefit comes with increased flow-rates for that particular block. Two things struck me as especially interesting with these results.
1. The EK Supreme, while being very restrictive, still delivered the best temperatures at any flow-rate.
2. The Apogee GTZ goes from bad to great along a very steep curve. I wonder if it delivers better temperatures than the Supreme, given enough flow. I might get an opportunity to test that soon.

Brief conclusions:
- The EK Supreme is very reliable in that it delivers good temperatures almost independantly of flow-rates.
- The Apogee GTZ needs lots of flow, but given that, it performs great. Since it isn't very restrictive, it's not a big difficulty to provide that needed flow either.
- The cuplex hd isn't only slightly cheaper than it's predecessor, the cuplex XT di, it also performs ever so slightly better.


Feedback on this is highly appreciated. If you would like more information and/or pictures, there are two options.

A) You know German and check out the original article (or you don't know German and just browse the pics).
B) You ask for it here. I'll post more data and pictures if there's any demand.

Cheers,
Shane

[evilsponge]

Nice work thanks!

-sponge

[Giannis86]

interesting testing there. nice to see that you used more mounts.
However there is one thing to consider. At different voltages a pump will dump different amounts of heat in the loop.

Also the flow rate-temp graph looks a bit weird to me. I would expect the results from different flow rates to be far less than illustrated (it could be because some of the blocks are restrictive though so..)..

Anyway my cpu temps (fuzion v2) are barely affected from turning the pump down. In fact they seem to improve at some point by a small margin when the pump (DDC3.2) runs at slightly below its full rpm. I know its not a rather restrictive block but maybe its a thing to consider

[generics_user]

interesting testing there. nice to see that you used more mounts.
However there is one thing to consider. At different voltages a pump will dump different amounts of heat in the loop.

Also the flow rate-temp graph looks a bit weird to me. I would expect the results from different flow rates to be far less than illustrated (it could be because some of the blocks are restrictive though so..)..

Anyway my cpu temps (fuzion v2) are barely affected from turning the pump down. In fact they seem to improve at some point by a small margin when the pump (DDC3.2) runs at slightly below its full rpm. I know its not a rather restrictive block but maybe its a thing to consider

180l/h = 1 GPM

the scaling is so good because it starts with such a low flowrate, testing needs to be extended to at least 360 l/h (2 gpm)

[BlueAqua]

Very nice testing results. I like how you varied flow rate with the different blocks to see how it differs on performance. From your testing that Aqua Computer Cuplex HD block is tempting me now. The Ek Supreme did really well, that's great to hear too.

Thanks for posting this and I hope you keep posting your test results.

[HESmelaugh]

Thanks for the feedback!

However there is one thing to consider. At different voltages a pump will dump different amounts of heat in the loop.

Thanks for pointing that out. Very important: I used a ball valve to lower the flow-rates. That way, the pump's heat dump stays the same for all flow-rates tested. I will edit the original post accordingly, since this is a very important factor and needs to be mentioned.

180l/h = 1 GPM

the scaling is so good because it starts with such a low flowrate, testing needs to be extended to at least 360 l/h (2 gpm)

I agree, would have been nice to see a 2gpm comparison.

After seeing the GTZ's trendline heading for the Supreme's, I want to test with higher flow-rates too. :-)
Though you gotta admit, the differences above 1 GPM are minuscule. There's a bit of a cultural difference going on here, too. In Germany/Europe, having a single loop for the entire system (CPU, MB, GPUs,...) is the norm. Most people don't use very powerful pumps either and the general agreement seems to be that flow-rates above 60 l/h (0.26 GPM) are unnecessary. By those standard, my test-rig is extremely highflow, having a powerful pump with just the CPU and radiator in the same loop.
Here on XS, I often see multi-loop setups and tubing sizes are also generally bigger (no pun intended).

How do I get 2 GPM anyway? Will a dual-Laing do it? I'm hoping that I wont have to replace all the tubing since that's a lot of bother... -.-

[evilsponge]

Hi HESmelaugh,

Question, I was under the impression that with the hot running GPUs we have today a single loop is just not enough. Especially if you consider crossfire and sli system that are typically volt modded and overclocked. Has there been testing performed in Europe/Germany to support the premise that one loop is sufficient opposed to a dedicated GPU loop? or is it an academic exercise based on flow rate numbers and heat dump calculations. I agree it has come to my attention that the benefit of 1/2 barbs opposed to 3/8 is minuscule. I am new at water cooling and look forward to hearing your perspective.

BTW I took a look at your testing page very nice article though I can't read German


-sponge

[HESmelaugh]

Hi evilsponge,

There seem to be several questions in there. First: "One loop for whole system?"
If you ask for advice on selecting watercooling components in a German forum, what generally happens is that people will recommend 9x120 Radiators. Examples from MagiCool, Watercool and Aqua Computer.
Since the heat load of a gaming rig is huge nowadays, and everyone here uses single-loop setups, you simply need huge heat-exchange surface to get good temperatures.

Now, I don't know of anyone who tested single-loop vs. dedicated loops and published results, but A) hundreds of people have CPU, GPU and more in one loop and get good temperatures and B) I'm sure the dedicated loop setup gives you better temperatures. Probably not by much, but certainly better.

I agree it has come to my attention that the benefit of 1/2 barbs opposed to 3/8 is minuscule. I am new at water cooling and look forward to hearing your perspective.

This is a different subject all together. Size of barbs isn't really relevant, but tubing diameter is for flow-rates. Here is an article I wrote, comparing 8mm ID to 12mm ID tubing. I got about 50% more flow with the larger tubing.

I hope I managed to adress your questions with this.

[MpG]

Great to see this data! Although looking at some of those curves, it makes you really wonder what would have happened further to the right...

180l/h = 1 GPM

the scaling is so good because it starts with such a low flowrate, testing needs to be extended to at least 360 l/h (2 gpm)

Depends on your country, technically. In the US, a gallon=3.78L, so 1gpm~226L/hr. In Canada and UK, the gallon equals something else entirely - over 4L, iirc. But yeah, that's why the the scaling is so impressive.

Do they really get by with only 60lph in Europe?

[Giannis86]

i dont know about in germany, but here in the netherlands i have seen a lot of people running high flow systems (im also one of them ). I think that you are a bit generalizing there about europe as a whole

[HESmelaugh]

i dont know about in germany, but here in the netherlands i have seen a lot of people running high flow systems (im also one of them ). I think that you are a bit generalizing there about europe as a whole

True. What I should have written is "German speaking Europe" which is Germany, Switzerland and Austria. Outside of that, I have little experience with what people build, since I only know what's going on on German forums. I apologize to anyone who was unjustfully marked as a non-highflow-fanatic.

Do they really get by with only 60lph in Europe?

In German speaking regions, yes. High flowrates just aren't the thing here. For many users in this region, silence seems to have the highest priority. In fact, if you ask about multiple loops on the forums I frequent, you mostly get told that thats pointless/a waste of money/not worth it etc.
I have to say that it's really interesting for me to see this difference between the preferences on the XS-Forums and the ones in German forums.

[Bojamijams]

I don't know where you got that "More then 0.26GPM is unneccesary" but that is a totaly wrong idea to have. Educate the Germans for us

Martin has showed repeatedly that the temperature drop drops off heavily after 1GPM but until that point there is HUGE temperature decreases to be had. Look at his site www.martinsliquidlab.com

[HESmelaugh]

Well, I guess that also depends on your definition of "huge". But with these tests and my reviews, I do try to show facts as clearly as possible, so that everyone can decide for themselves, how much flow, what size tubing, what pump etc. they want.
And thanks for the link, though I have to tell you I already know martinsliquidlab.com. Dude, do I know it. Martin is the reason I test the way I do. Before I came across his site, I didn't have a clue about running a proper test.

By the way, another thing I still want to improve in my testing is finding a way to make the differences between blocks more apparent. Greater heat-load should do it, but I'm afraid that will have to wait for a test-rig upgrade including a more oc'able CPU. Any other possibilities for making differences between blocks more pronounced are greatly appreciated. :-)

[evilsponge]

Thanks for the info HESmelaugh. OK after seeing the size of that radiator I can agree one loop would probably work but most people I see around here try to internally mount their water gear that rad looks like it would have to be on the outside of a case

-sponge

[Bojamijams]

There, you can point all your german naysayers to that As you can see the gain from 0.5GPM to 1GPM is over 2 degrees. After 1GPM i'd say its not worth it but its DEFINETLY worth it to have more then 0.26!

[Waterlogged]

Boj, the problem is, they're not using the blocks we use. Most of them are using blocks designed for their flawed thinking.

[HESmelaugh]

Well, you can't really say that's true, Waterlogged. For example, the EK Supreme is just as popular here as it is in America (and, apparently, everywhere else in the world). It is true that Swiftech and D-Tek are used less around here, though.
Also, I didn't get a chance yet to test a D-Tek Fuzion V2, but I'm working on getting a sample.

Bojamijams, thanks for those stats. As always, Martin's work is great.

[Lagaaja]

http://translate.google.com/translat...-8&sl=de&tl=en

And there's the translation (sort of). Nice info!

[Waterlogged]

Well, you can't really say that's true, Waterlogged. For example, the EK Supreme is just as popular here as it is in America (and, apparently, everywhere else in the world). It is true that Swiftech and D-Tek are used less around here, though.
Also, I didn't get a chance yet to test a D-Tek Fuzion V2, but I'm working on getting a sample.

Bojamijams, thanks for those stats. As always, Martin's work is great.

Ah, didn't realize that. Just sort of figured that with companies like Zern, Heatkiller, Innovatek and some of the others I can't think of ATM, there was very little to no room for anything from outside Germany or the surrounding NW region. My apologies.

Zern and HK do make some neat stuff though. I've got an Innovatek G Flow and....I'm not impressed.

[HESmelaugh]

http://translate.google.com/translat...-8&sl=de&tl=en

And there's the translation (sort of). Nice info!

I had a look at this and it made me laugh. Lots.
Verry Engrish. Read the translation at your own risk... :-)

Ah, didn't realize that. Just sort of figured that with companies like Zern, Heatkiller, Innovatek and some of the others I can't think of ATM, there was very little to no room for anything from outside Germany or the surrounding NW region. My apologies.

Zern and HK do make some neat stuff though. I've got an Innovatek G Flow and....I'm not impressed.

It's true that there are quite a few regional manufacturers, but when it comes to CPU waterblocks, they aren't more popular than international manufacturers. For GPU blocks, Aqua Computer and especially Watercool are very popular here (almost everybody uses full cover blocks, too) and I'm sure you find greater differences there than with CPU cooling.

Innovatek... nobody who knows anything likes them much. They make stuff that is generally somewhere between medicore and awful and charge tons of money for it.

The reason that American-made coolers are rare in my roundup is simply that it's more difficult for me to find a sponsor for those coolers. Gabe from Swiftech was the only contact here who agreed to sponsor my testing right off the bat. Big thumbs up! :-)

[NaeKuh]

[IMG]After 1GPM i'd say its not worth it but its DEFINETLY worth it to have more then 0.26!

so lets see.

After 1gpm to 3gpm, which is probably less then what im running.

I get:

.76 + .44 + .21 + .21 + X

= 2.5C + X(being how many more gpm i am above)

Oh wait... Bojam, but is done on a D-tek, which isnt pressure senstive as the EK.

Meaning i will most likely pull BETTER then what martin tested.

*looking again*

Yeah sorry to pop your bubble, but with that much gain, and this forum being XS and all, my RD-30 and dual DDC setup stays.

[Bojamijams]

Haha point taken. I was just giving my example in 'german terms'

[HESmelaugh]

This really makes me want to push the flowrates and do some more testing.

Can someone point me to something like a "how to maximize flow-rates" guide? Maybe some examples of how to set up a loop with two pumps for maximum effect? This would also be great to know since I'm going to test Dual-Laing tops soon and would like to compare them to other loop setups.

[NaeKuh]

This really makes me want to push the flowrates and do some more testing.

Can someone point me to something like a "how to maximize flow-rates" guide? Maybe some examples of how to set up a loop with two pumps for maximum effect? This would also be great to know since I'm going to test Dual-Laing tops soon and would like to compare them to other loop setups.

i hope you are also aware on the EK testing, that the block is VERY VERY tempermental.
Has a worse attitude then the girl representing my avatar.

meaning it acts differently each time you rotate the block 90 degree's.

[hotdun]

Thanks HESmelaugh! Great post!