A problem with c/w

[Diggrr]

I've noticed alot of hsf manufacturers and reviewers using the c/w rating to judge the effectiveness of hsf's.

Here's my question: How can anyone know what my cpu is using in watts? (even me)
From what I understand from a buddy who works at Intel in Portland (P4 fabrication), they test each chip after it's made sorting them into bins for their tested clockspeed. It's then that the chip gets it's Megahertz designation....so every single chip is different. They always get different clockrate cpu's from the same piece of silicon. I assume AMD does the same.
My board does not consume the exact same power as the next one off the assembly line, and my power supply has different output voltages than the next one off the assembly line. Heck, even my wall outlet has 5-8 volts variation from one minute to the next. All the minute electrical components even have a +/- of 5-10% tolerance in their values.

So how could anyone, using a database program, correctly calculate c/w? Presuming that I could take 3 "identical" mobo's and 3 "identical" power supplies, and using the same chip on each, consume different levels of power?

Couldn't 10 watts skew the number? Couldn't an incorrect number in a review effect a companies sales? Couldn't an incorrect number make the OCer's pursue a P.O.S.?
How could it possibly be accurate?

[Tiger]

Its not the input wattage its the output wattage of heat that is being measured. There is a programe called Radiate that calculates the heat wattage output for most of the common procs at all the different settings

[Diggrr]

Yes Tiger, I know that. But wattage consumed is directly proportionate to wattage radiated as heat.

Say an electrical conductor were 90% efficient. Thus 90% of it's power going in is conducted through, and 10% is radiated as heat. That's where it becomes proportionate.

If my particular chip were a 1.2G t-bird in fabrication, but didn't do well in the testing, they would program and label it as a 1G t-bird. The wattage of heat radiated would be different from the 900Mhtz chip that did so well in testing, it was labeled as a 1G.

So how could radiate know this?

I'm not griping, or complaining, but how could this number be accurate enough for everyone to follow it as if it were the 'end all' answer? Could someone please enlighten me?

[Tiger]

Here's an extract from a document about processor die manufacture;
Junction Temperature
The speed and volume advantages of MCM technology come at the expense of higher heat flux and increased failures. One of the most important considerations is keeping the IC junction temperature within its optimum temperature ranges. If the junction temperature exceeds its rated specifications no other performance specifications can be guaranteed. The junction temperature is equal to the ambient temperature plus the product of the component power dissipation and the junction-ambient thermal resistance. Several factors affect the junction temperatures, including the thermal resistance of encapsulation, the effects of surrounding components, and the type of heat removal used.
The whole document can found here;
http://developer.intel.com/design/smartdie/cob.htm#link7_0
Since the tolerances in this kind of manufacture process would be very strict I would think that the average specs for individual procs of the same speed would be very close. Since the components on the mobos are also manufactured to the same kind of tolerances the same would apply. Given this kind of manufacture I doubt one would see variations greater than 1%.

[BillA]

Diggrr

you are correct, but also for several different additional reasons
which not so many seem inclined to address
too indeterminate, most prefer a simple answer - even if it's wrong

suggest eyeballing C/W confusion, need a logic check (http://www.tekforums.co.uk/posts.php?threadId=3670)

then try How to make a simulated CPU heat load? (http://hardforum.com/showthread.php?threadid=266016) (missnamed, really a heat die)

pick the pony and we'll flog it to death

be cool

[Diggrr]

Thanks for the link Tiger, very informative reading. And I have no doubt that the manufacturing process is exact as can be. But, the examples I gave above are not supposition. It's what really occurs in manufacturing. The person I alluded to above is specifically a technician repairing and calibrating line equipment involved in testing and fabrication. This is not based solely on 'inside information' either. I've also seen film of the actual sorting process.

My contention is not with the chip manufacture, but in using computers with all the possible variations, to rate c/w using a database of averaged calculations. The only way I can see c/w being equal and consistent is in using standard equipment with a known, proven, and calibrated heat radiation. This is something that no reviewer has at his disposal.
I'm not knocking the reviewers either. They are using the equipment they have available, and can afford. They are also making a concerted effort to keep the playing field level by comparing one hsf to another on the same computer. Lemonade from lemons.

What I am getting at, is any c/w rating should be taken with a large grain of salt (or pellet if you will), and it's accuracy shouldn't be depended upon to be what you will receive on your system.

By the way, the article indirectly points out that case cooling is as important as the hsf's efficiency. It states that surrounding component temps are a factor, as well a heat transmission through the conductors on the board to and from the chip. This brings up the point of differing motherboards giving different results just from thermal conduction.

*edit* nice, BillA. (you posted while I was writing). Now if only all the reviewers had the same setup, and regulated power supplies to ensure wattage didn't change along with the wall outlet power.

[BillA]

and I'm even one step beyond
- use a Sola harmonic transformer
- 4 individual preset regulated DC power supplies (to 500W max)
- simultaneous metering to milivolts and miliamps
etc. etc.

now beginning to setup for a wb testing article

be cool

[Diggrr]

Excellent Excellent Excellent!

This is exactly the pony that needed flogging!

Now if only we could talk you into testing the top brand hsf's to give a real c/w rating. :D

And by the way, this info is not necessarily for me. I watercool using my own homemade blocks. I've just seen this c/w number influencing so many fellow OCer's and really don't want to see them wasting money on misinformation (however rightly intended). I've noticed that many of them seem to have a spare hsf or more lying around, which implies a bad purchase was made somewhere along the line.

Thank You BillA.:beer:

[BillA]

go back and re-read that C/W Confusion thread

I think a new "T/W" term might be more descriptive of the item, rather than the setup

thoughts ?

be cool

[Diggrr]

Yes.

I don't believe in accurate temp measurements in the waterblock. (as far as testing cpu temp, or heat load)
I think it would be much more accurate to know the exact ammount of heat goin in. No two people can place the probe in the exact same place for a waterblock, so your setup would be of much assistance here too.
I could, theoretically use these multiple measurements to make improvements on my water system, but I think exact testing is much better suited to comercially made systems. The cnc's used provide dimensional repeatability, so any arguments I made for my design over another would be moot. No-one could exactly repeat my hand carved wb's.

As for any hardware sensing, I'm waiting for the in slug sensor to work on the XP motherboards. Even then, it's all just a guess in my humble opinion. An engineer at DigDoc for example, could draw prints and specify components until he's blue in the face. It's what purchasing department actually gets a hold of that's going to determine accuracy. I've worked electronics assembly both in supervision and in quality control. You'd be amazed at what a little tolerance stack-up can do.

Just a couple of thoughts. Thanks again for the links.

*edit* it would be nice if DangerDen would include a temp probe hole in the cnc program, then we'd be comparing apples to apples.;)

[BillA]

"include a temp probe hole"

won't/can't happen

AMD specs a letter H drill hole, 2mm up from and parallel to the face, to over the die area

this is a damned difficult hole to drill

I'll work on the T/W thing

be cool

[Tiger]

Just to add some extra thoughts here. As you can from my sig I am working in the negative temperature zone which is providing some additional factors. The block size shrinks because of the cold and this reduces the pressure of the w/b on the die and therefore affects the transfer of heat to the w/b. So an additional factor in the testing of w/b's with a standard test temperature as well.

[BillA]

good to hear from another sub-zero cooler

as an ex matls engr working with adhesively bonded joints and cyclic stress testing,
I quite agree that differential thermal contraction can be a factor
fortunately it is not difficult to calculate (as long as the materials are not anisotropic and constrained !!)

use springs and be clever with the choice of materials
or chill the attachment pieces

unfortunatly the wb testing article will probably not get into sub-zero apps
for while it is quite easy to do so (I use a Haake A82 recirculating chiller, to -30^),
there is little interest and I doubt that wb performance is much different
(but I'll take a quick look, heat spreading should have a larger affect, maybe CuSil ??)

be cool