ebay under aws983s, heatware under Mr. Tinker.
good read. didnt get through all of it yet.. but good read.
id say bill presents about as many questions as answers, but no one can blame him for that.. no one has all the answers.
I like this quote:
...a normal person would think [realistic testing] means putting the sink on a CPU and directly measuring its cooling efficacy, the presumed metric being temperature. But things are not so simple; which temperature? And where? And by what method? To what degree of accuracy? Dare I add repeatability?
ebay under aws983s, heatware under Mr. Tinker.
I thought it was a great read, I agree with Maxxx though that after reading it I came out with more questions than I had going in. Probably because a lot of the points haven't really been given much thought before.
Some thoughts I had coming out of reading it:
1) If he suspects that the IHS is more durable than a copper slug because of the nickel plating...why not nickel plate the copper slug?
2) Repeatability with TTV. If he is arguing that the IHS doesn't show the same wear as the copper slug, and argues that small differences in the flatness of the slug over time degrade the results to the point of uselessness....wouldn't small differences between different IHS skew the results between different TTV's? Also, wouldn't the slight differences in machining the probe groove, where the probe is situated (mere mm different seems like it might make a difference) and finally the epoxy or method used to secure the probe in place. Would this make each TTV as unique in their results as each die sim is today?
He suggested that the TTV is good because it allows for flexibility as processors continue to evolve, however these differences between each CPU seem like they could be as significant as the slight unflattening of the copper slug in the die sim. Would it be enough of a difference such that you need to retest every block every time you change your TTV processor?
3) Secondary losses. The % of secondary heat loss Bill talks about seems to be a figure received from testing done primarily with air heatsinks. What effect does watercooling have upon the percentage of this secondary heat loss. With less airflow wouldn't that secondary heat dissipated decrease significantly?
Once again, these are just some questions I had after reading the article, questions which I am sure dont have answers at least at this point.
P.S. : Reasing the proposal for testers to adopt the Intel TTV design and machining grooves into real CPUs...I couldn't stop thinking of MaxxxRacer's soon to be testbench
Erasmus
1. I dont think nickel plating will help much of anything either way. Its a very thin layer and is easily scratched, just like the copper. And the IHS being more durable is merely a suspicion I think. A slug and the IHS are made from the same material. So long as the Slug is the size of the IHS, durability wont really change.
2. The groove in the IHS deffinetly poses some interesting questions. Whenever something is machined there is some deformation. What I can say though is this. The heat put into the IHS during machining is very little due to the end mill having to be stopped on a regular basis during the machining process. In short, the IHS stays coool.
But anyway, that is neither her nor there regarding repetability. Repetability refers to a SINGLE TTV testbed. there is absolutley no two testbeds that are exactly the same and will give duplicate results. There just isnt. A few companies strive to get relatively close, but doing so is rediculously difficult. In short, your question was kind of off base as it seems you didnt understand what repetability means.
yes you would need to retest every block if you change the TTV.. no way around it.
3. A test done by overclockers.com showed that the outer edge of the IHS had a bigger delta to the core temp, the better the cooling used. So, you can relate that to secondary losses.. Better cooling means that the waterblock/heatsink is pulling out more watts from the CPU and thus secondary losses are lower. This can deffinetly be seen in the GPU world. When moving from air to water, the PCB can easily drop 10-20C, indicating a huge decrease in heat transfered to it, or as we call it, a decrease in secondary losses.
In response to your P.S.. I couldnt help but think that his article, was in some small way, aimed at me. Maybe its just my ego grabbing hold of me, but it did feel that way.
Actually the nickel is much harder than the copper (which is a "soft metal") and Bill's statement is based on actual observation (with a monochromatic light source and optical flat); it's quite real, not just a suspicion.
Nickel plating the copper slug would be a solution. I'm opting to cap my copper slugs with an actual IHS (btw, I'm still in need of some, pm or email me). The problem is that I'll be working with a 10 by 10 mm and 14 by 14 mm slug, and while it might be similar to many processors out there, it doesn't actually replicate them. In fact, it doesn't replicate anything exactly, except for an old P4.
IHS's are manufactured to very strict tolerances. You'd be hard pressed to find an IHS that is measurably different than another one. But... a P4 IHS is different than an LGA IHS (the latter is thicker).
The IHS groove specifications have a tolerance that is also very strict; hard to deviate from it. I think Maxxx is going to hook me up with his guy, at 0.001" (thousandth of an inch).
As for secondary losses, it's really something hard to quantify. It's assumed that there are more secondary losses from an HSF than with a water cooled (or better) solution, but there's no actual measure. I've got a plan to put a mobo/CPU assembly in a vacuum chamber, and running some tests this way, if I can figure out how to put it together without frying the whole thing, or have the thing implode on me.
You're on the right track Maxxx; if you change the testbed, you're starting over. There's a repeatability factor to everything (including the Swissflow flow meter btw) so you ought to be able to swap out a dead component with negligeable impact, if you know the repeatability factor. Otherwise, I'm hoping to grow the WBTA block collection, so that any member can come back and re-test, if necessary. You should have one block as your calibration block.
The scariest part of the article for me was the pronouncement that without a flatness measurement, I'm going to be in trouble real fast. I need a method for measuring flatness (probably the mono light and optical flat if I can figure it out, and get it cheaply, or contract that out). I've been scouring eBay for months and have only come across the necessary equipment a handful of times (at an insane price). Right now I'm too busy putting the rest of my testbench together to even spend any time on this. I gave some thought to some kind of laser beam reflection assembly, but I haven't figured out all the details yet; I'm not even sure it would work.
Maxxx, I don't think anything in the article was aimed at you, but I definitely sensed a shot at Cathar.
Last edited by BigBen2k; 04-06-2006 at 03:02 PM.
Ben.. a vaccum chamber. now that is dedication. the tygon might not like the whole vaccum idea.
After I am done with my WB roundup I can donate some waterblocks to the WBTA block collection. I have about 12 right now.
One thing that I found puzzling is that Bill said NOT to lap the IHS after the groove is cut in it.. Other than removing the nickel coating, there isnt gonna be much difference. that is unless Bill knows something about the IHS we dont.. As in maybe its not meant to be perfectly flat to account for stresses place on the system. When presure is applied things flatten out?
while your slug doesnt represent any one cpu, that isnt of much concern. Die sizes are changing on monthly basis and the key here is to find a die that is similar in size to the average die for the time, and if you can, for a bit of time ahead.Nickel plating the copper slug would be a solution. I'm opting to cap my copper slugs with an actual IHS (btw, I'm still in need of some, pm or email me). The problem is that I'll be working with a 10 by 10 mm and 14 by 14 mm slug, and while it might be similar to many processors out there, it doesn't actually replicate them. In fact, it doesn't replicate anything exactly, except for an old P4.
Core sizes are also doing a wired up down in average size over short periods of time. First they go to smaller process on relativley the same design, thus making it smaller, and then the redesign the core with the same manufacturing process, making the core bigger, or more recently, adding a second core. and then the process restarts itself with the next step in manufacturing.
How would not having a flatness measurement kill your testing. The testing is based on how the waterblocks is presented to you. if itsn ot flat, then that will be accoutned for in the test results. (poor performance). While its nice to know if the block is flat I do not believe that it is imparative to performance testing. Getting the test die / CPU flat on the other hand would be a good idea.
BigBen2k - how are you planning to attach the IHS to the slug? Solder?
CPU Temps: Forget MBM, if it don't crash it ain't too hot.
For lab testing it presents quite the challange for sure, so many variables to control and some that are even hard to quantify. Definetly interesting read. Then there is real world testing. As noted there is quite a difference between product design testing and comparative product testing. When tough questions like this are brought up sometimes it is necessary to think outside the box and create a new scale for rating based upon the usual metrics. Obviously the usual units of measure are used in the lab for product design, and the new scale for comparative testing(At a cost of testing resolution but more meaningful to the consumer).
i.e. Grade B heatsink, S rating waterblock, this product scored a 6.7 on the "aicjofs scale", whatever brings a more meaningful quantifier to the average consumer. Me I like to read the hard data numbers, but I think I am a minority.
Getting a sector of the scientific community to agree on what criteria would yield what on the new scale would be almost impossible. Especially when pride, ego, and profit can influence the desire to adopt a new system, but often times once a few people adopt an idea others follow.
3770K @ 4.8Ghz 1.42V | GSkill 2x4GB @ 1066Mhz | Asrock Extreme4
2600K @ 4.5Ghz 1.38V | GSkill 2x4GB @ 933Mhz CAS 9-10-9 1.5v | Asrock Z68M/USB
1090T @ 3.8 Ghz 1.42V | Crucial 4x2GB 666 CAS 8 | Asus M4A79XTD EVO
Q6600 @ 3.66Ghz(8x458) 1.35V | GSkill 2x2GB @ 550Mhz 5:5:3:13 2.1V | Asus Maximus Formula(Rampage 0401) |
Actually, it's not the block's flatness that's interesting, it's the testing device (heat die or otherwise). If it changes, then I end up measuring a new block in a different way. The effect is measurable.
As for the nickel coating, it's all about how Nickel is a harder metal than copper, which leaves me with a low maintenance surface (but I still have to check it). Using a bare copper heat die to test, I'd scratch up the surface in no time, just from the roughness of everything that I'm about to test on it, either as a scratch, or as an imprint. The Nickel plating is going to make it a lot more durable. Every time we mount a block, there's a good chance that we're changing the surface geometry, so it's got to be checked periodically. An easy way to check this, is to run a test series with a "calibration block" (a block of your choice) prior to each test run, to see if everything is running normally. You keep your calibration block, forever; it's your reference point. If you start seeing different results with your calibration block, then you can either adjust the test results, or send the die out for maintenance.
I don't know exactly how the IHS flexes, but I'm pretty sure that it does (to make a flat contact with the CPU core). The old specs for clamping force of an HSF were much lower, prior to the introduction of the IHS. What that translates into (to me) is that some of this force isn't applied to the CPU core at all. In what proportion is this force applied to the core? Only the designers know. Knowing that, and being limited by the lack of info, what I have left is ... an IHS capped copper slug. I'm sure I could clamp the whole thing down as much as I wanted, because I'm not going to be risking crushing a core here.
Right now I'm not sure what I'll be using for that TIM joint. A liquid paste is going to be variable, each time. An epoxy would be permanent (not practical). a TIM film product would be an easy solution.
As for the vacuum test, depending on the level of vacuum, I might have to plumb some hard lines. I'm not looking forward to that, because then I have no flexibility for the mount. I might be able to get away with high pressure tubing, dunno yet. Right now what's stopping me is the cost of the pump, and the safety of the vacuum chamber. It'll get done, in time. Got a little help from my friends...
I don't think that there's a need for a new scale. The basis on which everything is measured (in electronics especially) is usually provided by the manufacturer, following industry standards. In this case, the Intel TTV. Note though that there's been a progression in the proposed testing methods: the old testing method included modifying the cooling device, not the CPU, because that wasn't an option back then (no IHS). It was also more consistent with standard methods, because the temperature probe was exactly at the junction of the CPU and cooling device. Now, with the groove, the temp reading is (ever so slightly) recessed on the CPU side. The groove construction instructions specifically call for the temperature probe to sit at the bottom of the groove (on a step). That's not the actual junction... but it's pretty close.
Ben, as a "Pro-lurker," it's great to see you posting here.
Q: How can you get mounting pressure to be as consistant as possible?
ebay under aws983s, heatware under Mr. Tinker.
Bolt down retention and torque wrench?Originally Posted by Mr. Tinker
CPU Temps: Forget MBM, if it don't crash it ain't too hot.
Pesonally I think using the same mounting on every block is not a good idea. While it does allow you to test the block itself more accurately, it does NOT allow you full testing of the mounting system that is used. It doesnt really matter if the block works well with "X" mounting when it comes with "Y" mounting which doesnt work very well.
Easy; use a pressure transducer.
Joe C (Overclockers) has an elegant solution, involving a Postal Scale:
http://www.overclockers.com/articles870/
Of course you could always use the mounting solution from the manufacturer, but there are a lot of variables there.
I'm leaning towards the postal scale: easy, simple, and accurate.
If I were doing testing on a purely academic scale, or for internal testing, a specific mounting system would be paramount, but for reviews that will be read by the end user, using each waterblocks mounting system is neccesary. Otherwise, the user will not really know what kind of performance to expect unless they use the same mounting as I did.. which would be custom.
I see your point about using the product's mounting system, since that is what will be used by the end user. What I mean is, many coolers have bolt-down mechanisms, and I wonder how to keep that type of mount consistant.
ebay under aws983s, heatware under Mr. Tinker.
bolt down as in a nut with a spring? like danger den, and cooltechnica. Swiftech made it ocnsistent with the collar between the spring and mounting plate.
What would ever make you think BillA and Cathar would have any slight feelings towards each other?
Yeah, I think in some ways I agree with Maxx and in other ways BillA. In some ways it's nice to see how all the blocks perform on equal "footing," but that really doesn't help the end user when the mountings are different. The ol' academic vs. "real-life" argument all over again.
I thought the article was written very well. I always thought BillA's work was done very well. I just think personal feelings have gotten in the way recently of good, hard data (and it's not just BillA). And, really, that's the most important thing to me--good data.
Oh, and I'd also like to extend a welcome to BigBen2k. Having read many of your posts over the years, I find your opinions to always be of upmost value.
Thanks Raven, I really appreciate that!
I came across some info that might be usefull for others. A "pressure transducer" is a vague term: a proper one would be "load cell". Plenty of hits on eBay for those.
Posting Permissions
Reply With Quote
Bookmarks