heat die dimensions - a public worklog

[bobo5195]

As always with you guys im going to have to go over this I H S stuff but at the moment I’m in a “I couldn’t disagree more mode, I H S good” but I’m sure the argument has been beaten to death.

Here’s a thought though. Why cant you use a die/I.H.S. sim block. Get a block of metal (silicon if your being pedantic) and attach it with thermal paste to you heater assembly. Measure the temp in this block of material and you need not worry about die size of the heater anymore as my swapping around this top piece it can easily be changed. As long as things are well insulated all the heat is going through this core so you’ve maintained Q. You can easily make this little assembly into a nice I H S rig (pre-made and stuff so bolt on and test/ test various versions all pre cured. You could pre cure all the blocks I H S assembly for a week quite easily) as well and if it gets scratched build yourself a new one!

[Incoherent]

Some comments

Assuming a calorimeter type sensor placement, what is the necessity to use copper for the heat die ? (i.e. does the matl's conductivity matter so much ?)
while Ben's suggestion to use iron seems radical, why not use a more durable material ?

I think the only criterion for the material conductivity is that it is known. Steel has many grades and different k's which is why I didn't use it. If I could be sure of it I would have no hesitation using steel. Silicon carbide is another good option, machining aside. We use a variant at work with a very low CTE which has a k of 170, which I have been looking at. Not in my price range though, especially if machining is factored into the equation. If we could set up an accurate tester of thermal conductivity it would be rather useful. Secondary losses in this test are even more critical than the heat die.
There are other more trivial issues. Copper is so far ahead of all other suitable materials in terms of conductivity that I have been a little concerned about how much heat is generated over these long heat paths. Using heater cartridges mitigates this concern somewhat (versus the metal film power resistors I use). A positive aspect of this is that with the higher temperature gradients comes a lower dependence on measurement accuracy. Water temperature measurement resolution and accuracy remains of critical importance if heat to water needs to be measured.

Guard heaters are a direction I like, possibly an easier way around the secondary losses than proper insulation (vacuum).

Sensor holes and position. Depend of course on sensor size. Heat shadowing and other similar effects are problematic as is the fact that the waterblock itself will effect the temperature gradients in the die. One suggestion, one I will act on in future, with centred holes drill all the way though. The temperature cross section will be symmetrical, even though the heat shadowing is worse the compensations are much easier.

BTW Bill, the phenolic arrived, thanks. I will not be able to do anything with it for a while, work and move dominating life right now.

[BillA]

I'll investigate mal\tl costs and fabrication shortly
what are your thoughts on sensor # and placement
I'm looking at 3 to 5 sensors in these calorimeters, why not 3 vs. 2 ?

my last setup at TMT used a guard heater (a flat iron on Low with a Variac put in an insulated box with the heat die inside and level with the top)
with a PID controller it reduced the time constant to an hour or so - 75% reduction from super-insulation alone
- pretty crude though

[BillA]

AlSiC ?? http://www.alsic.com/papers/cpseuro992.pdf

how much interest might there be in this ?
the sintered piece would have a big pedestal which could be cut down as desired

[bobo5195]

Forgive the somewhat newbieness of this post but why on earth does it take so long for loops to water up to a stable condition?

Surely use of guard heaters as extra heaters and pre heating of the water in the loop should make heating of water far less than an hour. It might require you to be ready by the off switch but I see no reason why the loop cannot be prodded in the direction of thermal equilibrium very quickly.

I assuming here that your using a closed loop to test with as opposed to a open and quasi open loop with a large tank of water and the use of a secondary heat exchanger to obtain accurate water temps is not used (heat exchanger used in closed loop to cool water to pre-described level in a short closed loop ie pump with restrictor, heat exchanger and pump. Allows also for secondary measurement of heat input by level of heat addition to well insulated secondary tank)

[Incoherent]

Originally Posted by unregistered

what are your thoughts on sensor # and placement
I'm looking at 3 to 5 sensors in these calorimeters, why not 3 vs. 2 ?

I had three on the previous fluxblock. I found that it the temperature gradient was very linear so dropped the third sensor on the current version. Still, having more would be a good idea, especially if people want to calculate heat flux, the more data points the better. The disadvantage is that if you have inline (drilled into the heatpath) the effects of the sensors on the heatpath become more dominant. Also, a longer heatpath is then required to accomodate them, in a 10x10mm channel this would lead to very high temperatures, increasing secondary losses etc.
As far as placement relative to the interface, I am in two minds. One way is to have them quite far away, like 3mm so that the sensor effect on the surface is minimised, more length to even out the temperature at the actual surface. On the other hand going as close as possible minimises the error of any extrapolation. I think that whatever one does, assumptions need to be made and the problem modelled in order to correct for the heat shadowing. Les and I have been exploring this a bit in the Fluxdie thread.
I of the opinion that the only really safe way to position the sensor is in the centre of the heat path despite these shadowing effects. Of course, this means that the sensor needs to be as small as possible and the die CSA as large as possible to minimise errors. Since we want to test 10x10mm this limits us. Small thermistors are also not very cheap. I was intending to use Betatherm thermistors (0.457 or 1.01 mm diam.) but the cost stopped me, at ~$10 each with my tendency to destroy perfectly good sensors through mishandling this is untenable. I use Mitsubishi, RH16's which are 1.5mm in diameter and under $1.00. RTD's I am not up to speed with. Thermocouples can be any size so are probably a good solution for most people, especially since to get thermistors accurate/linear there is a convoluted calibration procedure.


Re AlSiC. I'd be very interested.

[Incoherent]

Originally Posted by unregistered

...(a flat iron on Low ...)...- pretty crude though

I'm all for crude, if it works.

[BillA]

looky here: http://www.tc.co.uk/news/news_minisensors.htm
info on 4-wire 0.5mm dia RTDs requested

bobo
water chillers, chamber, inst., etc. all good to go in an hour
the problem was with a heat die having 4+" of insulation coming to a steady state condition (reading to hundredths eh ? with a probe in the insulation)

my preference would be for a heat die designed (or adapted) to have a 20% secondary heat path loss, this a nominal mobo/socket (through the traces) sink 'value' described at an IDF by Intel
the actual value is different for every board

[bobo5195]

Ah i see.

Im worried about this secondary heat path as you are no longer exactly sure (for a well laged die) of the power input. Surely it would be easier to just apply an analytical model and say there is 20% less heat input and such. It is far easier to insulate everything well than have a well designed secondary path with debatable characteristics.

Could also try and use emprical relationships using a watercooling loop to measure energy output into it from a cpu versus the actual cpu power. Given that the heat die system can give you everything else.

[BillA]

the insulation is going to be a big deal for the riser w/sensors
best might be to pot the bare element w/leads into a filled cross drilled hole
the phenolic insulation could be cut for one side with RTD lead holes and sealed to that riser face, the remaining to insulate the other 3 sides (and locate ?)

this is what I was asking Stew, if the rad Q is measured via flow and temp - why not also the wb ?

[bobo5195]

Thinking about things i am abit uneasy about this whole concept of hole shadow, especially given that this thign is in equilibrum and well insulated.

[redleader]

Originally Posted by unregistered

10x10 and 14x14
ok by me, 14 should serve for GPUs moderately well ?

I agree.

Prescott was one of the largest CPU cores in recent times (ignoreing dual core), and the ball park for the core is just 70mm^2. The rest is just cache, which doesn't actively produce significant heat, but does dissipate some heat via conduction from the core itself.

100mm^2 should be a fine middle ground between current dual core processors and current single core processors. Future dual cores will likely be closer to 100mm^2 anyway.

I'm not sure about GPUs. I've never seen a layout of their functional units. I'd guess its fairly uniform over the whole die though since there is no cache and many symetrical pipelines.

[bigben2k]

Originally Posted by unregistered

looky here: http://www.tc.co.uk/news/news_minisensors.htm
info on 4-wire 0.5mm dia RTDs requested...

Oh please share any info you obtain!
Pricing:
http://www.tcdirect.co.uk/deptprod.asp?deptid=230/29


Going through the core sizes this morning, narrowed down to 90 nm only, we have:

Smithfield 90nm
206 mm2 / 95 to 130 Watts

Prescott 2M 90nm
135 mm2 / 84 to 115W

Prescott 90nm
112 mm2 / 103W

Toledo/Manchester 90nm
199 mm2 / 110 W

San Diego 90nm
115 mm2 / 104 W

Venice 90nm
84 mm2 / 67 to 89W

Winchester 90nm
84 mm2 / 67W

Palermo 90nm
> 84 mm² / 62 W

This would dictate:
-10x10 to cover:
Prescott (112 mm2), San Diego (115 mm2), Venice (84 mm2), Winchester (84 mm2), and Palermo (84 mm2).

-14x14 to cover:
Smithfield (206 mm2), Toledo/Manchester (199 mm2)

...which unfortunately (?) leaves out Prescott 2M (135 mm2) in an intermediate position (do we care?).

I am happy with 10x10 and 14x14.

[RoboTech]

In regards to concerns about thermal shadowing of sensor holes...

One idea I planned to try in a future die sim was to bore/drill a tiny hole (1/16") up thru the center of the heat die, to within ~1mm of the top die surface. The hole would pass between two cartridge heaters and the bottom opening would allow sensor wires to exit out the bottom. One sensor would be inserted all the way up to the top, blind end. A second senor would then be inserted behind the first one to a desired distance from the first ( say 10 to 20mm). Omega makes some very small RTD's (all flavors: J,K,T, etc) that could be thermal epoxied in place inside the bore.

The top surface area would effectively be a square with a tiny hole in it (although not breaking thru to provide better top temp measurement). Building the die in this way would eliminate any shadow. The top surface area calc would exclude the area of the hole. The hard part will be drilling a very small hole that deep - especially if its in copper!

Edit: Alternatively, two (or three) parallel holes could be used, each for its own sensor if there was concern about spacing, clearance for multiple sensor wires, and possible heat conduction thru wires from one sonsor to another). This could also allow for a tighter fit of the sensor in the hole.

Just a thought...

[Incoherent]

Originally Posted by RoboTech

Just a thought...

I like that idea. The thing that would concern me is the uncertainty of the measured temperature location. For what I am doing an important part of the equations is knowing where the sensor is in the heat path.

[BillA]

no, even the centerline assumption for a horizontal wirewound is a bit iffy for what we want
(poor boy all the way here)

hey Ben, you see the price !

[jaydee]

Any more ideas on the design? Will try to draw it up if you give specs. Will then send it to who ever wants it in and what ever format Solidworks 2003 can put out.

[bigben2k]

Yeah, pricey.

I'm perfectly comfortable with a square die, but the upcoming lines of dual-core offerings might throw a wrench.

[BillA]

FWIW, dual cores should have dual independant sources
sorry, that's how its done in the big leagues

and 4 core to come ?

suggested is that the technical sophistication may soon exceed the 'practical' participation of all but big players

jd
you have a go but we lack some of the bits yet, let's list what we need;
1) sensor #
2) sensor placement
3) sensor selection/dimensions
4) sensor hole definition + tolerances
5) heater selection/dimensions
6) heater hole definition and tolerances

5&6 are no biggie, 1-4 need resolution
I'm trying to source some sub-mm RTDs at a tolerable price, give me a bit
- anyone out there use such ?

[bigben2k]

Bill, I haven't even able to find one, until you pointed one out!

The closest item I have is an element (only) from Omega, 1mm diameter, 10 mm long. 2 wires:
http://www.omega.com/ppt/pptsc.asp?r...NTS&Nav=temc13
p/n 1PT100GX1510
$51 ea (in their bare configuration)

You're welcome to what I've found so far:
http://www.newportus.com/Products/RTDprobe/RTDElem.htm
http://www.newportus.com/Products/RTDprobe/1PT100G.htm
http://www.airpaxtsp.com/tspsite/tproduct.html
http://pyromation.com/products/prt.html
http://www.minco.com/support/ts103.php?section=11
http://www.pyrosales.com.au/thermocouples.asp
http://www.riedon.com/rtd.htm

or to try GlobalSpec

or to go through the WBTA weblinks of manufacturers of temperature equipment:
http://wbta.us/index.php?option=com_...tid=21&lang=en

[BillA]

someone makes them smaller, suspect Germany as the source for the UK
?