EK Supremacy EVO

We were excited to receive the EVO – a new CPU block from a manufacturer that already has one of the top performing blocks from our 2012 roundup means that we expected awesome results.  We weren’t the only ones – EK warned us to expect a 2C delta in comparison to the old Supremacy.

As we received a pre-release sample – the fancy outer sleeve for the box was not included. The packaging and included goodies are however the same as the old Supremacy:

At first glance the block even looks very similar though immediately we can tell that it’s not the same:

Like the original Supremacy you can order the Intel or AMD version – this of course is the Intel version and it comes with screws and backplates for 1150/1155/1366/2011 socket boards. Instructions were too fresh to be included in the packet – but I was sent a PDF instead and they are up to the usual EK standard – clear, descriptive and an example to the competition.

The block is protected in a sealed bag and the base is protected with an easy to peel sticker:

So far then nothing is new – even the mount mechanism and the thumbscrews and springs are the same. Because they are the same we won’t bother to talk about them except to say that they are the same usual high quality and ease of fitting. We are however going to have to look at the internals to figure out what is up:

Comparing to the old Supremacy (top) vs the EVO (bottom):

Here we can see that the design is similar in architecture. The nickel plated copper base is in fact identical. A similar stainless steel jetplate is used to force water into the center of the machined channels in the copper base. The jetplates are still changeable and different ones are optimized for different families of CPU. The difference is in the acrylic section where instead of it being one-piece there are now three pieces. The cover of the block is now separate and simpler in design. There is then a large insert which fits to the cover and creates the path to and from the base.

The small insert fits into the larger one and narrows the channel of water before it hits the jetplate. It also forces the flow to split into two halves before merging again and hitting the jetplate.

Once the small insert fits into the large insert they look as if they are one piece:

The naming of the block then becomes clear – this really is a minor internal change to the Supremacy – it is an evolution of the design.

Performance

Given that the external look and options (copper or nickel base, acetal/plexi/copper/nickel top, csq or clean design style) on the EVO is essentially the same as the old block then our only real concern therefore is the difference in performance and the most important metric is cooling! It’s been a while since we’ve had a new CPU block up for test.  The last mini cpu block testing we did was on the Impact Monoblock.  For this review we will still be testing on the 4770K/Max VI Impact combo, but at a different heatload due to having to replace the first failed Impact board.  Therefore results can not be compared apples to apples and existing blocks like the original Supremacy and CPU-380 had to be retested.  It’s also worth mentioning that the Supremacy had jetplate 4 in that test while in this comparison we are using jetplate 2 which performs better on the 4770K.

For the thermal performance of the block each rotation of each block got 5 thermal runs at a fixed flow point. Each run consisted of a fresh mount on a fresh application of MX2 TIM. The CPU was loaded with the latest Prime95 for 1.5 hours. The cpu temperatures were logged with realtemp, while the water coolant temperatures were logged with an Aquaero 6 running the latest firmware and software. The 4770K CPU was clocked at 4.0GHz and a Vcore of 1.25V – a low power load but due to Intel’s mistakes this is all we can safely get out of the CPU without delidding while running prime.

Each run averaged all four core temperatures subtracted from the coolant temperature. Referencing vs the coolant temperature takes out more sources of error meaning that we have better data on the performance of the block than if we referenced it to ambient temperatures. The flow was set by a MCP35x2 dual pump with it’s PWM setting set to 34 which equates roughly to a regular DDC. Flow was roughly 1.25-1.5GPM depending on the CPU block in question.

The correct setup for the 4770K was insert 1 and jetplate 2 – here abbreviated to I1/J2. We found that a “regular” mount was the best orientation. For more on orientation check out this explanation of what “regular” and “goofy” means. Interestingly the EVO did a little worse on average than the original Supremacy. Unlike our previous testing the original Supremacy beat the CPU-380 because it now used the better jetplate 2 which gives the block a pretty extreme bow.

This result surprised as we expected a performance improvement. In order to rule out some possible differences in ours and EK’s testing we also checked the other jetplate (J1) which is included in the above data. The other jetplate was worse as expected. Considering that there was a possible flow difference between the two sets of data we then swept flow rate to be sure:

Across the region of interest the EVO is consistently slightly worse. EK then suggested we may have used too much TIM (we had followed the instructions to use “2 grains of rice”) – EK suggested half as much TIM for a 4770K but this also made performance slightly worse still as the TIM now did not fully cover the IHS. MX2 is very easily spreadable TIM, so any excess TIM just gets squished out of the way meaning that a fully covered IHS is one with maximum contact to the block. At this point we wanted to test the blocks on a 4930K to see if this poor performance was simply limited to my 4770K. However as the NDA approached there was not enough time to get this data ready before launch. However we will be testing again on the 4930K to verify that EVO is in some way better than the presumably cheaper to make old style Supremacy so stay tuned for that in a week or two.  Sadly that data will be made irrelevant very quickly as Haswell-E will be paper launching in only a few days.

It is worth noting that both the Supremacy blocks had a significant variation in their results and this can be seen when plotting confidence intervals:

This gives us an idea that in reality the two blocks are not separated by much and that 5 data points for each is not enough to really distinguish their performance. However it does seem enough to say that on this CPU that the EVO has not lived up to the hype.

Having said all of this the CPU-380 was out top block in the 2012 roundup and it does seem to be handily beaten on by all version of the Supremacy when on the 4770K. So the EVO is still the best performing CPU block that can be bought for the 4770K simply because the old style Supremacy can’t be bought anymore!

Flow & Restriction

Flow and restriction was tested with the setup used in this test for the EVO, while the old Supremacy numbers are from data collected with jetplate 3. If this plot is meaningless and you want to understand it then follow this link to find out what it means.

As can be seen the Supremacy EVO is more restrictive than many CPU blocks, and particularly when configured with this insert/jetplate combination. It also gives up ground on the old Supremacy too which is disappointing but expected given the design changes. While this would be worthwhile if performance was increased, it’s a disappointment when performance hasn’t been increased.

Conclusion

So the Supremacy EVO is an evolution of the existing design that provided us with similar or worse performance in all aspects, but yet looks the same and presumably costs more to make. Fine tuning jetplates and inserts can give better performance but does make things more complicated for the user and I hope EK won’t go along the route that Danger Den did with it’s fully customizable jetplate design. Sometimes leaving too much to the end customer can be “too much” particularly if the benefits are small or non existent. This design has become more complicated but has not yet shown a benefit.

Given all of this negativity in the performance results it can be hard to remember that this is still the best block tested on our 4770K that can be bought today. In a void if the Supremacy had never existed we would have applauded it’s performance the huge range of options for the style and finish. It therefore wins a gold award, though somewhat reluctantly simply because we expected more. Hopefully the 4930K/5960X testing will prove EK’s performance claims correct, but for now it’s hard to get excited about this perceived downgrade to an already good design!

EDIT- I’ve been surprised by the amount of negativity towards the block that this review has created. At the time of writing there were no other reviews to compare to, nor did I have time to test other CPUs like the 4930K. Since the NDA lifted there are also reviews that show the block in a more positive light. Some of those reviews are of high quality (moonman’s for example is as far as I can tell very high quality) while some of the reviews should just be ignored as poorly thought out and performed. The reader has to remember that there is variation in the CPU IHS flatness and power density across parts, and even more across SKUs. In addition there is the differences due to TIM used and slight variation in the block due to manufacturing tolerances also. Just because this block happened to be slightly worse on average for this CPU doesn’t mean that every block on every CPU will be worse. This doesn’t mean that the test data is flawed, just simply that there is a limitation in time. To thoroughly evaluate a CPU block you would have to run multiple CPUs of every type with multiple versions of each block and multiple types of TIM, with enough mounts of each. Even then all you’d find out is that there is quite a wide spread of possible results. This is why I’m a fan of the confidence interval plot – when you look at that you see just how much overlap the blocks have in the possibility of their performance and that’s only 95% confidence and only for this one sample of each block and CPU. In summary no reviewer has the time, money, energy or will to be able to do as “good” a review as we would like to. 5 mounts per block is really a minimum to show broad trends on one cpu. A reader therefore is always wise to not only consult multiple reviews and see which might correspond best to their setup but also to learn what differentiates a quality review from a poor one while knowing that no one review will perfectly match their own setup. As the great tester Martin of martinsliquidlab once said “anything with 1C should be considered the same”. So is it a great step forward in peformance – no. Is it an evolution of the design and is it still very good – absolutely yes.

Where to buy: Performance PC’s – Price – $73-78