First off – a big thanks to the sponsors because without them this never would have happened. If you like reviews like this make sure to support them so that they will support future reviews!
Performance PC’s were generous enough to sponsor six blocks – so big love to them:
Thanks to my other sponsors (in alphabetical order):
DT Waterblocks for sponsoring the 5Noz, Direct and Sniper blocks
Danger Den (sadly now closed) for sponsoring the M6 Block
EK Waterblocks for sponsoring the Supremacy Block
Indigo Xtreme for sponsoring the TIM
Koolance for sponsoring the CPU-370 and CPU-380 Block
MIPS for sponsoring the IceForce HF Block (and Frozen/Aquatuning for getting me it)
Sidewinder Computers for sponsoring the Bitspower Summit HF
Spotswood Computer Cases for sponsoring the tech station
Swiftech for sponsoring the Apogee HD Block and a MCP35X2 pump to enable the measurements vs flow rate!
Motherboard – Rampage IV Extreme – stock heatsink, fan underneath board to avoid VRM throttling
CPU – Intel i7-3930K (Unlapped) @ 4.7GHz 1.45Vcore
RAM – 4x4GB Gskill Trident DDR3 2000-8-9-8-24 1.65V @ 1600-9-9-9-27 1.5V
HD – WD RE4 250GB
Case – Rich Chomiczewski’s Test Bench
Radiator – XSPC EX560
Fans – Gentle Typhoon 120mm 2150rpm in push/pull with BGears 120->140mm adapters
Reservoir – EK 400 Basic
Pump – Swiftech MCP35x2 – PWM controlled by CF635
Temperature Logger – CrystalFontz CF635 USB + SCAB module, probe data logged with WinTest
CPU Core Temp Logger – RealTemp 3.7
Temperature Probes – Dallas DS18B20 “One Wire” Sensors modified to fit inside a Bitspower 3/8” barb
Flow Meter – King Instruments 7520 series 10” 3.5gpm range
Quick disconnects – Koolance VL4N series (4 pairs)
Pump power was kept constant for the MX-2 runs, and it was swept over 7 points for the Indigo Xtreme. The MCP35X2 is a dual pump (in series) but was run at 39% PWM setting for the MX2 runs which roughly corresponds to a single MCP35X running at 100%. I.E. This test is meant to reflect a typical cpu only loop setup with one radiator and one strong pump. If you know your setup is significantly stronger or weaker than this setup then you will want to look at the results with Indigo Xtreme and modify them based on the MX2 to IX delta.
Because ambients were fairly well controlled and because my data was taken multiple times and measured vs coolant temperature rather than vs the ambient temperature then I think my results are about as accurate as you can be without calibrating the cpu temperature sensors themselves. I would estimate that my margin of error is no worse than 0.5C, and therefore averaged data should be a little better than that.
It should be noted that this data is taken on a 3930K and that results may differ significantly for different CPUs with different heat signatures and bows on the IHS.
Results Part 1 – “Normal” TIM (MX2)
Here are the results of the first round of testing using the MX-2 TIM. MX-2 is pretty much the standard TIM for watercooling testing because it:
- Has zero curing time
- Is reasonably independent of mount (i.e. reduces human error)
- Is easy to apply and clean
- Easily available and cheap in large quantities
It is not the best TIM in the world. However I will be showing data on all blocks with Indigo Xtreme to show “best case performance data” in part 2. 5 clean mounts were completed on each block with MX-2 and the CPU was loaded for an hour. Temperature data was logged every second. The last 20 minutes of data were averaged and the delta between the average cpu core temperature and the water temperature was used as the most important metric for test results.
Performance Data – Temperature
Here is the raw temperature data for MX-2 showing each data point for each block:
As you can see many blocks were close and this really shows how close they are as it’s hard to
pick out which are better than others. However some blocks were very inconsistent while some
were fairly flat. It’s also hard to tell which block is which if you’re color blind so in order to understand this data we can then process it in various ways. As different people think different methods are fair I’ve included some of the more common ones.
Average of all 5 mounts
This data assumes all mounts are equally valid and you that the average of all 5 is a good
choice for performance metric:
Best out of 5 mounts
This shows the best case performance scenario, however if a water block has poor mounting
repeatability then it may not be a good indicator of performance for you. More on that later:
Average of the middle 3 mounts
This discards the best and worst results as outliers and focuses on the “more” likely data
95% Confidence Intervals
This takes the average and attempts to show how wide the area is around it that you are 95%
likely to hit based upon the standard deviation of the 5 mounts. However bear in mind that the distribution of results is unlikely to be a normal distribution. It is also reliant on only 5 data points so should be taken with a pinch of salt and in the context of the other plots!
This does show is just how close things are.
Performance Data – Other
Temperatures aren’t everything. For more complex loops with other components, such as GPUs ,
block restriction affects the flow of water and hence performance of these other components.
Here is the flow rate at this fixed pump power setting:
Mount repeatability is also a concern. It is always hard to know without substantially more
data whether a poor result here is truly from the mounting system, the bow of the block or
simply human error. After mounting blocks over 100 times just in this test I’d like to think
I’m not an idiot at it but I’m always willing to be proved wrong. Some blocks such as the
Swiftech Apogee HD and the EK Supremacy do let you bottom out the thumbnuts and so do provide a more reliable mount and this can be seen in the data. In this plot we show the standard
deviation of the thermal performance of each block across the 5 mounts. A higher score is an
indicator of likely poor mount/bow for socket 2011:
Another good indicator of how well a block mounts is to compare it to Indigo Xtreme which we’ll look at in part 2.
The raystorm really does win the value for money award here being priced considerably below the other top performers.
Block rotation does sometimes matter. It matters most for blocks that use channels rather than pins because the cpu cores are laid out on the 3930K in a way that favors the block water channels to be lined up vertically when mounted in a “normal atx tower case”. The improvement varies from block to block on the order of 1-3C. See the chart in the port threading section to determine which blocks are correctly aligned by default (horizontal logos/text when in a “normal atx tower case”).
The alphacool could not be recommended because it only had one complete thread on the G 1/4 threads. It should be noted that I did not see any problems with the threading nor do I have the time and tools to do accurate measurements of “likelihood to strip the thread” or “torque to failure”. Alphacool have stood by their design and stated that it meets required specifications however personally I would feel more comfortable with a fully threaded block. Here is a table showing the other blocks and their amount of thread in the ports. It also shows whether the blocks are aligned for maximum performance when mounted with manufacturer’s typeface correctly orientated in a standard atx tower case:
Part 2 – Indigo Xtreme
This is the second part of the results and focuses on testing with Indigo Xtreme which should take out mounting/block bow effects as well as measuring block performance vs flow.
The test setup was identical however in this case multiple mounts were not used for performance results. Indigo Xtreme when applied and reflowed correctly with the right mounting pressure has very similar results from application to application. This is one of the reasons it was chosen in order to test vs flow. Multiple mounts would make the data collection prohibitive as there would be too many variables.
First a good mount was ensured by multiple ETI reflows at varying pressures. The block was tested at one fixed flow rate and the data was analyzed in conjunction with the appearance of the reflow. Once a good mounting pressure was determined the block was then tested across 7 different pump settings. Flow was recorded as well as the water temperature and cpu core temperatures.
The results taken at 39% PWM can be directly compared with the MX-2 results.
The DT Direct block is not a “normal” waterblock. Instead of the water cooling a copper base that transfers heat from the IHS of the CPU with a layer of TIM in between, the DT direct allows water to directly come into contact with the CPU IHS. The advantage is that there is no thermal resistivity due to the TIM and copper base. The down side is that there is less surface area for the water to contact and less heat spreading effect. Here is the data for the blocks vs pump power setting:
As you can see there are only a few blocks that have different slopes. I.E. A few respond better to high flow than others. However performance of *every* block depends not only on the block but also on the flow. You may wish to base your decision on block purchases based on the expected flow of your loop.
If you know your current block and current flow rate you can use this to compare blocks by using this plot to work out your effective pump power in relation to mine:
It is of note to see that there is a knee in the curve around 1gpm as would be expected and that performance increases are minimal after 2gpm. Do not be fooled by this plot into thinking that the Phobya block is better than the CPU-380. This plot by it’s nature gives an illusion that higher restriction blocks perform better. No one that I know of adjusts their pumps for an exact flow when considering different blocks. Instead pump power is usually kept constant when changing blocks and so the plot of temperature vs pump power is much more useful.
It is also worth comparing IX to MX-2. While improvements can be made by changing your waterblock or your pump pressure, they are also affected significantly by your choice of TIM. Do your research on this (skinneelabs is a great resource for looking at TIMs). IX is certainly one of the best if not the best, however it’s not the simplest to apply and I’ll be putting a guide together for each block that should ensure a good application (at least on socket 2011).
If we compare all blocks and all TIMs we can see the effect of IX vs MX-2 is about 2-5C which is a larger difference than between many of the blocks
We can also see that sometimes changing your TIM may help you more than your block. If we plot this delta we can also use this data as a way to judge the bow/mount of the block in addition to the mount variation that is given by the standard deviation of the MX2:
Here is the same plot but with the raw data points shown:
Again you can see that the mount can make a huge difference and this is shown by the Danger Den M6 which performs much better when using the Koolance CPU-380 mounting kit.
From the data we can see that both the new Koolance CPU-380 block and Phobya UC1-LT blocks stand apart from the crowd on all three temperature metrics. It is therefore fair to say that in this type of loop setup on a 3930K that these are the best performing blocks.
There are some surprises in here too. I was not alone in expecting the EK Supremacy to be in first place in this roundup. In fact it was beaten by many other blocks. In st0neds tests on Ivy Bridge it did very well. At first I thought this was an indicator of a flaw in my mounting, however the Indigo Xtreme tests confirmed that the data looked right. EK have since said that the bow is just a little too much to perform well on socket 2011.
It should also be noted that full copper blocks do not always outperform mixed blocks. For example the Raystorm Full Copper performed worse than the acetal version. I am not the only one to see this. Martin saw the exact same result in his testing. The bow of the block will be different when using different materials and may be worse or even improved depending on the socket. Don’t go ahead and automatically pay more for a full copper block because it may not be worth it. However the one advantage to full metal blocks is a reduced likelihood in stripping the threads of the block ports.
Quality and Mounting
These are both very subjective. Although the Phobya did well thermally, the CPU380 feels like a much better block. It looks and feels less “cheap”. The mount feels sturdier, doesn’t require access to the back of the motherboard, and requires only a few turns of the thumb screws to install. In short if I were to draw an analogy, the CPU380 feels more like a Mercedes than a Hyundai. They both may get you to work at the same time but the experience along the way is different.
Here are also some of my pet peeves in waterblock design that I hope manufacturer’s won’t repeat going forwards:
- loose mounting bracket e.g. Raystorm, NexXxos etc – why not attach it to the block? Floating around buys you nothing except something else to hold onto when mounting the block
- port spacing – making it very wide makes it much easier for me as a tester. At the least make it wide enough for 3/4″ OD compressions. The Heatkiller had the narrowest ports by far
- slow mounting mechanisms – e.g. DD, Alphcool, Phobya – You shouldn’t have to turn a thumbscrew a bazillion times to mount the block because the machine screw was too long. You also shouldn’t have to screw a machine screw from the back side of the board another bazillion times because you cheaped out and didn’t make a design that could be screwed in from the front of the board
For mounting – the EK Supremacy, Koolance CPU-380 and Swiftech Apogee were amongst my favorites.
Choosing the best waterblock for you depends on many factors and it isn’t as simple as just reading off the best performing block from one test. The factors you need to consider include and the weight of each will vary from individual to individual:
– CPU/Socket and the curvature of the IHS
- Flow rate/pump pressure of your current loop
- Water block restriction
- Manufacturing quality
- Ease of mount and repeatability
- TIM choice
The last I’ll ignore as you can use your own judgement!
The Koolance CPU-380 performed exceptionally besting every block on MX2 (nearly all of them by 1+C) while still being at the top of the heap with Indigo Xtreme. The quality and feel of the block is up to scratch too. Mount variability is good and the flow is decent too and I would heartily recommend it.
The following blocks also performed very well and have their own niche uses:
- the Phobya UC1 performs very well and is also priced very well
– Swiftech Apogee HD because it performs very well and has a very nice mounting system
- MIPS IceForce performed very well and has very high flow
- the XSPC raystorm performs well and wins the value for money award!
- the Watercool Heatkiller performed very well with IX