Waterblocks on i7 - Part 5

[HESmelaugh]

Here, finally is my next installement of waterblock-testing. Because of delayed samples this one has been a long time coming.

New Presentation

I've decided to try a new presentation for this data. In this first post, I will simply show the most simplified version of the results. The idea is to make it a more accessible reference. If you just want to quickly check the essential data, you'll find it straight away in this first post. All the details and pics as well as more in-depth data you'll find in the second post.

One thing is important here: If you look at the data in this first post and have a question, read the second post. It's likely that you'll find the answer in there.


Flowrates:




Average Temperatures:

[HESmelaugh]

Here, finally is my next installement of waterblock-testing. Because of delayed samples this one has been a long time coming.

So let's get right into it.


Test Setup

Hardware used:
- Intel Core i7 920 @ 3.7GHz, 1.28V
- Asus P6T
- 6GB Mushkin XP12800
- T-Balancer BigNG and Sensorhub for measuring and logging temp data
- Three digital thermal probes measuring ambient temps
- Arctic Cooling MX-2

Cooling loop:
- Swiftech MCP355 with EK X-Res top
- Digmesa Flowmeter
- Four inline thermal probes
- XSPC RS360 radiator
- 3x Nanoxia FX1250 fans @ 1200rpm
- 13/10mm tubing

Sftware used:
- T-Balancer Navigator
- RealTemp 3.0
- CoreDamage


Testing Procedure

The testing is done with the above hardware set up in a CPU-only loop so that there is no thermal interference from other components in the loop.
A test run has a duration of 40 minutes during which the CPU is fully loaded using CoreDamage. I did a lot of preliminary testing and determined CoreDamage to be the best suited program, because it produces very high and very stable temperatures. The 40 minutes might seem short, but in my testing I got identical, just as repeatable results after 40 minutes as after 60 or 90 minutes.
From the 40 minutes of CPU-burning, I collect all the data from the last 20 minutes of each run and average out all the data to get average Core temperatures, average water temperatures and average ambient temperatures.


Blocks tested

Alphacool Niagara


A very particular new block from Alphacool. The base plate is the same as with the Livingstone and Yellowstone blocks, but the internals of the top are different. Uniquely, it's equipped with G3/8" threads.


KoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAP CPU-350AC


The notorious KoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAP block. Most demanded block for me to test for this roundup...


Watercool Heatkiller 3.0 LC


This is the low cost version of the Heatkiller 3.0. It has fewer channels in the base plate than the other HK 3.0 versions and also comes without the jet-plate. I'm sure someone is going to ask, so let me state right away: The reason the HK 3.0 LT is not featured is because it's simply the same as the copper Heatkiller I have already tested. Only the LC version actually has different internals and therefore different performance.


Ybris Eclipse


The Eclipse is a fully chrome-plated block made by Italian manufacturer Ybris. As you can see on the first two pics it's available with either an acrylic or an aluminium mounting-plate.


Flowrates

I simply measured the flowrates achieved in the loop with the pump at 12V with each block. Since all other parts of the loop are always equal, this represents the restrictiveness of each block.

Results:


In terms of flowrates, there is a new record holder among the four waterblocks added in this round: The KoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAP waterblock is the most restrictive block I've ever tested. The Ybris Eclipse is also considerably restrictive while the HK 3.0 LC and the Niagara are both quite unrestrictive.
NOTE concerning the Niagara: I tested it using fittings with 3/8" threads. then did another few test-runs using 3/8" to 1/4" adapters and 1/4" fittings, just to see if it makes a difference. With the adapters, the flowrate is 1.73 GPM instead of 1.75 GPM. This has no measurable effect on the temperatures.

You'll see that I also added the Zern PQ + Pro here. Unfortunately, I could not do thermal testing with this block, since my sample was faulty (concave base plate). I didn't want to withhold the flowrate-data, though. For thermal performance, I can refer you to Bundymania's test on HWLuxx.


Orientation

Orientations

Each block is mounted and tested six times. Three times in one orientation and three times in an orientation rotated by 90°.
This means that the blocks were either installed with horizontal/vertical flow accross the CPU or diagonal flow with the outlet on the top right/top left, depending on how the threads on the block are oriented.
Keep in mind that the socket's orientation on the mainboard isn't always the same. Some mainboards (e.g. Gigabyte) have 90° rotated sockets. What ultimately matters (if anything) the direction of flow accross the CPU itself. On my mainboard, it would look like this:




Averaging out the results from three mounts per orientation, this is what we end up with:

Diagonal Flow:


Horizontal/Vertical Flow:




Temperatures

Next, let's take a look at temperatures achieved. (all temperatures shown are differences between CPU and water temps)

Here are the three runs in the optimal orientation for each of the blocks. All of the results are sorted form lowest to highest temperature for each block to keep the graph tidy:




Here's what we get as an overview, when all the results from the three runs above are averaged out:



While the KoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAP CPU-350 is amazingly restrictive, it also offers very good cooling. The HK 3.0 LC also does a great job in this regard, especially considering it's low price. The Ybris Eclipse doesn't offer a stellar cooling performance, but finishing next to the EK Supreme isn't half bad, either. The Alphacool Niagara's performance is, unfortunately, pretty bad.


Temperature Range

As you can see from the above graph, there is practically one block for each tenth of a degree along the scale. This can be a bit misleading since on such a bar graph, a block will be placed ahead of one that averaged .1 K higher temperatures. So it can look like one block is better than another even though the both basically performed almost identically and ended up with different results due to margins of error in the measurement devices. To give a more realistic impression of the thermal performance, I made a candlestick-graph.
Here, the bar represents the temperatures measured during the three better test-runs and the sticks reach out to +/-0.5 K to account for the imprecision of the thermal sensors. All in all, this can give you a visual idea of where you can expect the performance of each block to lie.



NOTE: The images are simply added so that it's easier to see which bar represents which block. I'm not sure if this is an optimal way of displaying this. Suggestions welcome.


Temperatures in relation to flowrates

The last part of testing consists of lowering flowrates by using a shutoff valve and measuring how this impacts temperatures. I lower the flowrates to predetermined values with 30l/h or 60l/h increments. converted to GPM, the increments don't make as much sense as in l/h, but the important point is that each block is tested at identical flowrates.
Because of this, the following graph does not take into account the different levels of restrictiveness each of the blocks has.

First, the graph just showing the four new blocks in this roundup:



The curves that the Ybris and KoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAP blocks are displaying are what I would call average. The Watercool and Alphacool blocks benefit slightly more than average from higher flowrates. Keep in mind though that the KoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAP block will require way more pumping power in order to achieve the same flowrates as the other blocks.

Here is the very chaotic overview of temps in relation to flowrates for all 20 blocks tested so far:



Conclusion

As always, I want to keep this to a minimum and leave judgement and further speculation up to you.
In my opinion, the HK 3.0 LC is the most remarkable block of this bunch. Not because of it's performance or flowrates alone, but simply because it offers these at such a low price. Way to corner a market, Watercool.
I like the design of the Ybris Eclipse a lot and it's a block where I can imagine many will go for it for the looks, knowing that they won't get top-notch performance out of it. The Niagara is interesting but unfortunately not very good. Time to experiment with new base plates, Alphacool.
The KoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAPKoolKRAP... great performance, super restrictive. I'm sure speculation will not rest after this test, either.

Cheers,
Shane

[Hondacity]

wow hk3 leads by that much? hmmm retest perhaps?

[ILikeCosmosS]

i think this clear that the heatkiller can get lower temps with less retriction while the cpu 350 needs more restriction to get almost equal temps

[Snyxxx]

Awesome job. This is a tremendous amount of work and very professionally presented.

Way to go.

[RedRaider]

Great Presentation...

Again, a quality test from a quality person...

My beloved HeatKiller 3.0 is once again laying down the PWNAGE...

These test results differ greatly from another test I've recently seen...

[Worthy]

I have swapped my hk3 and my k350 back and forth a million times in all different orientations for the two, and I could not tell any measureable difference between them.

My results for each were completely repeatable too. Not even enough variance to take an average between a few mounts for each.

I like them both just the same. But context is everything.

In a cpu only loop I will always use the k350, and with a full block loadout, I prefer the HK3 because it flows better.

but some are using the k350 in large loops and multiple blocks with great success and the flow is not that bad at all.

Good writeup.

[trfnj74]

As always greatly prepared and executed... Well done and thanks for all the testing you have done.

[MomijiTMO]

Error bars = Excellent.

[BDW88]

Outstanding work HESmelaugh thank you for this!

[HESmelaugh]

Originally Posted by Hondacity

wow hk3 leads by that much? hmmm retest perhaps?

Currently, I have no reason to believe further mounts would yield different results than the six previous mounts.
Also, my test-rig is not set up for testing CPU waterblocks anymore, so retests are not going to take place for a while.

Originally Posted by Worthy

I have swapped my hk3 and my k350 back and forth a million times in all different orientations for the two, and I could not tell any measureable difference between them.

My results for each were completely repeatable too. Not even enough variance to take an average between a few mounts for each.

I like them both just the same. But context is everything.

In a cpu only loop I will always use the k350, and with a full block loadout, I prefer the HK3 because it flows better.

but some are using the k350 in large loops and multiple blocks with great success and the flow is not that bad at all.

Good writeup.

If you take into account a margin of error, it's not unlikely that you'll get identical results with these two blocks.



Taking into account +/- 0.5 K the ranges of the two blocks clearly intersect.

Also depending on pumping power and restriction in the loop, the results may lie closer together or further apart.

[Worthy]

Originally Posted by MomijiTMO

Error bars = Excellent.

true.

[Hondacity]

Originally Posted by HESmelaugh

Currently, I have no reason to believe further mounts would yield different results than the six previous mounts.
Also, my test-rig is not set up for testing CPU waterblocks anymore, so retests are not going to take place for a while.




If you take into account a margin of error, it's not unlikely that you'll get identical results with these two blocks.



Taking into account +/- 0.5 K the ranges of the two blocks clearly intersect.

Also depending on pumping power and restriction in the loop, the results may lie closer together or further apart.

thanks

[Septim]

another excellent comparison. thanks shane.

[YerBuddy]

Firstly, this is outstanding stuff here.

Secondly, would it be out of line for me to request the use of this comparison for a small newbie computer building class that I am presenting? It's for non-professionals that want to learn about basic enthusiast computer building and water cooling; I've been stuck in the middle of helping out. It's a long, very boring story (almost sex in the city boring), but I would appreciate being able to quote your findings HESmelaugh.

[HESmelaugh]

Sure, you're allowed to use this. Just reference it properly and don't remove the watermarks. i.e. as long as you don't try to pretend the data is yours, you're free to use it.

[Crow]

Originally Posted by RedRaider

These test results differ greatly from another test I've recently seen...

I've seen other roundups that have dissimilar results, specifically one at Madshrimps, but that could be easily attributed to other factors (flow rates, pressure, heat load, etc.)

But, all in all, the block with the best price/performance seems to be the Enzotech Sapphire.

Which is good, because I can't afford a $50 block, much less an $80 one.

EDIT::
Oh wait, I see what you did thar.

[ballz0r]

wow, a phenomenal amount of work hes... 20 blocks x 4 mounts each is 80 fucking mounts...

more than ive done in my whole life.

Thanks for taking the time to bring us this valuable info