Stren's 2012 CPU Water Block Roundup

[rge]

Nice job with the rotation data Stren! Helps to explain some of the testing differences.

[stren]

Nice job with the rotation data Stren! Helps to explain some of the testing differences.

Thanks RGE - I'm retesting the alphacool block now. The MX2 data was rotated and even though it's a block with pins, I guess the bow still favors rotation slightly (again about 1C). But this puts it in the "normal" range for IX to MX2 performance delta of 1-3C. The only blocks outside this are the DD M6 due to a poor mount, the Rasa and the Heatkiller.

FWIW I finally published my *long* "how to do the IX application" video guide:

[Church]

Hmm, out of curiosity, can you redo confidence plot with results sorted by awg temps? I wonder if it can improve readability or at least neatness

[stren]

Yes:



Like I said though - I wouldn't read too much into the exact width of the bars, we really don't have enough data to accurately show this.

[Church]

Why not read into? Wide bar = read - crappy mounting system.
Imho this chart looks well fit for main one, as showing most thermal info within one and in readable way.
(Also i wonder if with this one needs separate CPU Block Bow/Mounting Performance chart?)

[stren]

Why not read into? Wide bar = read - crappy mounting system.
Imho this chart looks well fit for main one, as showing most thermal info within one and in readable way.
(Also i wonder if with this one needs separate CPU Block Bow/Mounting Performance chart?)

Yeah this one I put in alphabetical normally just so that people could easily find a block of interest, but I agree this is easier to read.

The danger with this plot is that someone will judge performance based on the lowest end of the bar, not on the average and the width. Which as you said is a good indicator of mount/bow performance

[musicfan]

Agree with church that post #179 is an awesome graph with helpful information. It's my favorite. Perhaps I am wrong, but I suspect "enthusiasts" who don't get it the first time are willing to make the effort to understand...such as reading the thread or asking questions.

Plus any "danger" is only a relatively minor financial inconvenience. As you know, folks sometime buy on what looks good rather than thermal data anyway . So it's good all blocks are getting better and converging toward the mean over the years.

Your IX demo tape was fun too - always wondered about IX. Again many thanks for your countless hours of effort. It's amazing how IX TIM, block turning, block bowing, and CPU socket type all interact. Please pace yourself, take breaks, and keep it fun for you. Your testing will be the magnum opus of CPU block evaluation for some time. Great job Stren.

[Church]

stren: people imho usually turn to testing roundups to find several ones which do better and choose accordingly their block of interest from those, not vice versa. Such ordering helps to see it with quick glance. Also, if you sort other charts (eg. normal mx2 temp one w/o 'confidence bars' or flow rate one) by results, why this should make exception? All this chart might need extra - small comment beneath it, explaining what result wideness bar means, which looks to me still better then separate charts for average temps / best temps / block bow&mounting performance.
P.S.
Can you make Another one looks evaluation variant - with added two vertical lines from best temp and best average temp, then three lines, from which middle one is average temp of all average temps sum divided by count of blocks tested, and lines by it's sides - +/- 1C from it. Imho it might help to quickly show how much temp for particular block differs from very top one, and also illustrate better how close are most of blocks and which very few are above or below that 'averageness'.

[stren]

Agree with church that post #179 is an awesome graph with helpful information. It's my favorite. Perhaps I am wrong, but I suspect "enthusiasts" who don't get it the first time are willing to make the effort to understand...such as reading the thread or asking questions.

Plus any "danger" is only a relatively minor financial inconvenience. As you know, folks sometime buy on what looks good rather than thermal data anyway . So it's good all blocks are getting better and converging toward the mean over the years.

Your IX demo tape was fun too - always wondered about IX. Again many thanks for your countless hours of effort. It's amazing how IX TIM, block turning, block bowing, and CPU socket type all interact. Please pace yourself, take breaks, and keep it fun for you. Your testing will be the magnum opus of CPU block evaluation for some time. Great job Stren.

Thanks

stren: people imho usually turn to testing roundups to find several ones which do better and choose accordingly their block of interest from those, not vice versa. Such ordering helps to see it with quick glance. Also, if you sort other charts (eg. normal mx2 temp one w/o 'confidence bars' or flow rate one) by results, why this should make exception? All this chart might need extra - small comment beneath it, explaining what result wideness bar means, which looks to me still better then separate charts for average temps / best temps / block bow&mounting performance.
P.S.
Can you make Another one looks evaluation variant - with added two vertical lines from best temp and best average temp, then three lines, from which middle one is average temp of all average temps sum divided by count of blocks tested, and lines by it's sides - +/- 1C from it. Imho it might help to quickly show how much temp for particular block differs from very top one, and also illustrate better how close are most of blocks and which very few are above or below that 'averageness'.

Yeah I agree with both of you - I'll be using that style on all plots going forwards. I'm not sure why I chose that particular plot to be the one that would be different. My problem with the width of a bar is that I think it's only a good indicator of poor performance. Sometimes the stdev was only poor because of one mount. If that mount wasn't poor and instead I was lucky and got 5 good mounts then that block would look better. I just don't think 5 is enough mounts to give an indication of reliably good mount.

I'm not sure exactly what you mean by the looks evaluation? You mean give my opinion on the aesthetics of the block?

I finished MX2 testing on the rotated alphacool block:



It improves slightly despite being a block with pins rather than channels. It could be the bow of the block being rotationally sensitive, however I also suppose it may come down to where the input/output ports are relative to the cores on such a design.

Now that we've fixed the errors due to block rotation we can sort by the average delta:



Here's the plot with the actual average rather than the discrete data if you're interested. Personally I prefer the previous plot:



I also updated a few of the other plots:





I'm moving on saturday, so I'm going to wrap this up here and pack everything away for moving. I'll be updating the first page data and my website too with these plots in the next few days. I wish I had the time to test every block rotated, but I don't think that's going to happen and this is most likely the end of any data collection. The top performing blocks at least got rotated on MX2 and I think the conclusions are now more clear and I'm much more confident of my own data now.

Once I move into the new house the only thing left do will be the flow/restriction curves and then this "2012" roundup will finally be complete LOL.

[Church]

"Sometimes the stdev was only poor because of one mount."
Yes, there can be "unlucky" mount. IIRC Skinneelab for those reasons dropped worst (and maybe best - but imho best should be left, i easily can imagine user reseating few times to get tiny extra) mount from computing average. If most of mounts were all over the place then bar still would be wide.
"I'm not sure exactly what you mean by the looks evaluation? You mean give my opinion on the aesthetics of the block?"
No, i meant looks of chart itself

[stren]

"Sometimes the stdev was only poor because of one mount."
Yes, there can be "unlucky" mount. IIRC Skinneelab for those reasons dropped worst (and maybe best - but imho best should be left, i easily can imagine user reseating few times to get tiny extra) mount from computing average. If most of mounts were all over the place then bar still would be wide.
"I'm not sure exactly what you mean by the looks evaluation? You mean give my opinion on the aesthetics of the block?"
No, i meant looks of chart itself

Yeah I'm not as fond of dropping one result, because I don't think most people remount based on being 1-2C lower than expected. Most people don't take a baseline anyway, and if they do it's rarely measured well enough (i.e. datalogging of coolant temps with a non-curing TIM) to make a useful comparison. Those that do remount are usually worried about a large temperature change that most likely isn't a mount unless they really screwed it up

I'm still not sure what you want in that plot - sorry! Can you give me an example pic?

[Church]

I thought something like this:

Hmm, but then again it might not only illustrate how close most blocks are, but unnecessary put up very few leaders, so maybe better not

[StAndrew]

Not to hijack the thread, but Im interested in a review that runs all of a company's blocks from beging to end, ie for Swiftech, start with the strom and go through the Apogee's, etc... Would be interesting to see how far we have come.

[McDown]

Outstanding job stren! Thank you.

[Nightprowler_77]

Outstandig job!

I justed picked up a Phobya UC- Extreme for my 3960X/RIVE and noticed that my temperatures are way more higher than on my EK supreme witch I had to swap cause of corresionissues. After sevral remounts the temps seems to have droped a bit but i think the Phobya UC- Extreme has contact problem on the RIVE board. Adding alot of thermal grease seeems to help a bit.. Anyonelse has experienced the same problem with the same combination RIVE/ Phobya UC ?

And also the Phobya UC- Extreme came without instal instructions.. Any Idea where to find theme?

[stren]

I'm assuming that what you have is just the full metal version of the UC1-LT that I used. I didn't have problems mounting it, although I don't particularly like the mount. I also used it on the R4E so I don't think there is a specific incompatibility. You can screw the bolts through the back of the board and uses the thumb nuts, springs and washers to tighten the block down, or you can just screw the bolts in from the top of the board (again with the springs and washers). The latter is more likely to give you a bad mount because you have to do so many more full turns (25 ish if I remember). What TIM are you using?

[Nightprowler_77]

I have mounted many different cpu blocks in my life but this one seems be a hard case. First I tried the stock TIM from Phobya. Now I am testing the artic silver 5. Not a big differece. Only if I apply a lot the temps drops a bit. I have tighned the bolts to the maximum.. I have also added a brand new MCP655 pump to the loop get better flow, but it made no difference on temperatures..

I wonder if the CPU cooler gets in conflict withe CPU bracket on the R4E.

[stren]

Well in fairness AS5 has a very long curing time, no idea about the stock tim though. I would suggest MX2 if you really want to do an apples to apples

[stren]

Finally updated the results at the start of this thread and also on my website. The website has much more info than the "basic" quick read at the start of the thread.

So that's pretty much it for the CPU roundup. I'll be taking flow measurements in the next month, and alphacool may send me a slight variation on the NexXxos, but apart from that, this is done!

[Martinm210]

Would you mind retesting on the 2600K as well?
Sorry I couldn't resist...lol Just kidding!

Awesome work as always!!
Martin

[Church]

stren: have you counted how much time it took overall for all the tests/retests?
And thanks for your fine work

[stren]

Would you mind retesting on the 2600K as well?
Sorry I couldn't resist...lol Just kidding!

Awesome work as always!!
Martin

Lol I sense you've been asked similar things many times before!

I'm trying to get my new testing station setup in the garage in our new place. Tomorrow's job is to get the manometer hooked up and tested to see if everything looks right!

stren: have you counted how much time it took overall for all the tests/retests?
And thanks for your fine work

Well I think I started in June of last years so I suppose 9 months, but I lost a bunch of time testing at 4.8GHz before the chip degraded. Then the IX testing was slow because it was so hard to get a good reflow until they changed the instructions. Of course my work would also be a high priority interrupt as would the holidays. I'd say 80% of the "good" testing was from september to december.

[Church]

Martin: imho ivy bridge 3770K is more actual for lga1155 if testing on platform requiring fully different optimisations is considered .. simply as even worse corner case valuing bow (even smaller/narrower chip + ty stock TIM1 below IHS).
stren: if you decide to resume tests and on lga1155, then you can lessen workload, not redo whole gargantuan batch - simply test just most interesting, not whole batch. 380, Supremacy, UC1-LT, XP3, maybe something from worst ones or 1-2 of interest (my guess - blocks that seem to have more bow then average, as they might benefit more on this platform). Maybe even new poll can be set up, as to which blocks @lga1155 people are most interested in. For at least 380 - also with 90deg test.

[stren]

Martin: imho ivy bridge 3770K is more actual for lga1155 if testing on platform requiring fully different optimisations is considered .. simply as even worse corner case valuing bow (even smaller/narrower chip + ty stock TIM1 below IHS).
stren: if you decide to resume tests and on lga1155, then you can lessen workload, not redo whole gargantuan batch - simply test just most interesting, not whole batch. 380, Supremacy, maybe something from worst ones , and 1-2 of interest (my guess - blocks that seem to have more bow then average, as they might benefit more on this platform). Maybe even new poll can be set up, as to which blocks @lga1155 people are most interested in.

Yeah I don't have an ivy cpu/board right now, so unless a generous corporation wants to sponsor me with one I'll be taking a break from block testing until at least haswell. Next up is pump/res/top roundup, and I have a few other things in the pipe too...

[Martinm210]

Yes, sorry I just thought i would toss in the usual "we want MOAR" response before anyone else got a chance out of humor because anything this good someone will want more. I have a 3930K myself so I have no interest in smaller core performance. I did see some interesting changes in regard to bow preference between the 2600 and 3930 though. Generally the 2600 favored extreme bow where the 3930 preferred a more mild amount.

More than anything, I am in awe and happy to see so much awesome work put into this. That's really awesome work!! Kudos for hanging in there and giving us so much of your time!