OK, lets talk about die size... some more... :)

I have recently completed my second thermal die simulator. (My first attempt was modeled on a Thourobred core, made out of brass and was designed to be mounted into a dead socket 462 mobo. I found it more trouble than it was worth to try and rigidly mount the die just to use an existing socket - and maybe the brass wasn't such a good idea either!)

OK, a year later and I made a new one. Based on prevoius comments in this thread it struck me size really wasn't too important - it's a simulator after all. Given all the different CPU core sizes (now and to come) and the question of how big a role does the IHS play = no obvious decision.

For my second generation die I decided to go with copper (like everyone else) but make the die area much larger so it would be more robust and hopefully better able to stand up to repeated testing - maintain surface flattness and finsih without the relatively soft copper deforming over time. I picked 1000 mm2 (actually 1,024 mm2), which is 32 mm x 32 mm - almost the exact size of a P4 IHS (31 x 31) And I like even numbers... :)

Pics of my new thermal die simulator

This "seemed" like a good compromise as it better represents the actual contact area a HSF or waterblock will see when mounted to any of the new generation CPUs with IHSs. 1,000 mm2 is bigger than any of the actual cores, about the same size as a P4 IHS and smaller than AMDs new IHSs.

I don't think it is realistic to assume we can truly simulate the thermal characteristics of an actual CPU. Hot spots, thermal spreading, etc. By definition a simulator has to make some compromises so one size will more or less fit all.

I built my current prototype to mainly test insulation effectiveness and mounting techniques - haven't been too worried about die size. Since posting some details in my recent Reserator review it seems like a few folks are getting their pantys in a knot about how big my die is... :) Sure it will give much different C/W values than other die sims but so what???

I'm open to suggestions and can easily mill down the existing die surface area if given proper justification. (I'm already working on a new design that incorporates an automatic, universal loading mechanism.) So what do you think? What size is the best compromise for a CPU thermal die simulator???

I think a better idea than making the heater larger is to make the "core" the size you want and then put a heat spreader on top of that, to get an almost exact replication. Hell, butcher an old P4 and take that exact heat spreader...

HUGELY different. That makes all of your numbers about the reserator COMPLETELY invalid.

How so? I din't present any C/W data in the review! All of the performance testing was done on the same test rig (Xp-2400 in a real computer) under the same conditions that I test other HSFs and waterblocks. The thermal die data was added just to show how the Reserator's water temps responded to higher heat loads. It had nothing to do with the actual performance/comparison testing.

The 1000sq mm die is irrelevant unless you can relate to the IHS.
One of my guesses
http://www.jr001b4751.pwp.blueyonder.co.uk/IHS.jpg

However as mentioned may be interesring for Peltiers.

You compared it to something? Link? It still doesn't matter, though. That will disadvantage a lot of blocks, and enhance others (Zalman, I think) as well as heatsinks.

I don't like that robotech

That die sim seems to test how well it will cool a TEC, not a CPU die

Try testing a WW or a cascade on there

die simulators will have to have some sort of comprimise, no matter what, you just want the one with the least, and still upholds relevance to what we are trying to accomlish, which is to find out how well a particular cooler cools today's modern cpus, whether it be 10mm x 10mm in size, or have a IHS or not.

do we want to know how well a cooler cools a 10mm x 10mm square under uniform load? or how well it cools a barton cpu die (rectangular, not square), which probably isn't under uniform load? well, the difference is probably not very great for it to be a big comprimise, but it is still a comprimise, and er...............

ok, i have no idea what the hell i'm talking about, but I guess robotech can now test waterblocks for TEC usage with that thing

"Based on prevoius comments in this thread it struck me size really wasn't too important - it's a simulator after all. Given all the different CPU core sizes (now and to come) and the question of how big a role does the IHS play = no obvious decision."

this is the source of your errors
not obvious ? perfectly clear to me
you need to figure out who to listen to perhaps, someone has led you astray

OK Bill, if it's clear to you, please enlighten me... :)
Besides, YOU were the one I WAS listening to:

"take your pick, matters little through the sizes you listed,
and makes the results no 'better' "

Maybe I didn't put enough emphasis on the "through the sizes you listed"?

So what would be your preferred size and/or recommendation??? (Please)

the understanding is that the thin IHS does little to laterally spread the heat, so a reasonable die simulator could be slightly larger than a 'typical' CPU
(but never so large as an IHS)

I am still abusing my 10x10mm, and as you observed the maintenance of the 'die' area is hugely important, and quite difficult
- despite the best care I can give it, after about 40 mountings the numbers start to degrade;
with professional equipment, resurfacing takes ~2hrs (.5hr on lapping machine, 1.5hrs diamond lapping) + 3hrs teardown and setup

I used to spend days hand-lapping w/o satisfaction - inspecting with optical flats

your setup looks good, just shrink that die size

EDIT:
you doing flow rate/pressure drop measurements ?
JoeC, pH, and I have started a round-robin test series - e-mail if interested in joining
- and anyone else trying to take accurate measurements

Thank you, sir... :)

So if 1,000 mm2 is too big and 100 mm2 is too small (fragile) how about ~200 mm2???
14 x 14 = 196 mm2
14 x 16 = 224 mm2... ;)
15 x 15 = 225 mm2

I'd go 12 or 13mm sq

I thought this was about standardising. surely these suggestions do not.
Your die 100sq mm,JoeC's 140sq mm

You can't standardize something like this, unless you can standardize actual die sizes. There is the Athlon T Bird, the Palomino, the T Bred A, the T Bred B, the Barton, the Willamette, the Northwood, the M0 one, the Prescott, the Athlon 64, there's soon to be a 90nm version of the Athlon 64, and the list goes on. I imagine that I missed quite a few. Then there's the issue of the heatspreader.

Why 12 or 13?

He is talking about standardizing testing procedures and equipment used so test beds can have comparable data. If everyone use the same die size on their die sim the results would be one step closer to comparable.

So they're comparable to other test beds (well, not really seeing as JoeC's is 140mm^2, Bill's is 100mm^2, pHaestus' actual thing is 98mm^2, Robotech's is 1000mm^2...) but not to what is actually important...

Well obvioulsy the other people will have to modify their die sims when they get everything worked out and a plan made. Not like this will happen over night eh?

Not sure standardization is ever going to happen in any case. I actually don't mind dies of different sizes; if we standardized position and resolution of the die temperature collection then I'd think a clever modeler (e.g. Les) could go to town with some sort of relationship between heat density (W/mm^2) and cooler performance using all our die sim results.

The problem is you can't get there until you can get everyone individually within calibration and understand the offsets among the different rigs. In principle heat density is what separates our results; in practice mounting force, insulation procedures, many other difficult-to isolate things do.

Actually it's 83.5mm^2 (T'Bred B). Quite a different challenge to cooling even a 100mm^2 die (or a Barton at 101mm^2), and different again to a P4 (147mm^2) or an XP64 (176mm^2 ??)

On a related issue, what about a diamond coating on the top of the die simulator to try and preserve its shape/size and avoid the need for constant lapping and maintenance? You could have the copper and phenolic resin be flush with one another on the top and then put a diamond sticking up to come in contact with waterblock or heatsink bps...

Actually, air is a better insulator than the resin, by a factor of at least three (last I checked, could be wrong).

I was just thinking about a nickel plating, like the Intel IHS. Opinions?


AA: it's best to be different for now, until we can sort out all the differences. That's where the WBTA is at: wait and see... ;)

I'm still going to go ahead with my plans to run test series on 10 by 10mm and 12 by 13mm, for a number of blocks. I should have my chiller remotely operated within a week or two.

I'm in support of things being different to (hopefully) better represent all the processors. What are you talking about?

A die sim cannot represent a CPU even if it is the same size.

...dependent upon the budget...

Can do lots of things correctly, given enough money (and expertise).

Generally though, totally out of reach to achieve for all but those with less than 6 figures of cash to throw at the problem.

Would take some serious effort to replicate the heat output of an actual CPU. Hot spots, cool spots, dead spots, wattage variances of different mobo's, etc...

That's true, but the aim is to get close. We can't get close by being 10 times the size of the die.