PDA

View Full Version : Testing under vacuum


bigben2k
03-31-2006, 04:31 PM
As I've mentionned previously, I'm planning a test run (or a water block) under vacuum, in an effort to quantify secondary losses.

Now it seems that I'll have to run a mobo CPU combo, for comparison purposes.

What difficulties does anyone foresee in doing this?


Assume for now, that I'm using a rotary vane vacuum pump to bring the pressure down to 1 millibar (atmosphere is ~1'000 millibar), and that I have a safe vacuum chamber, with the required feed-throughs (power, water, and temp probes).

qet77
03-31-2006, 08:02 PM
Realize that the reduction in absolute pressure means that the water block will be subject to >15 psi water pressure. This may need to be taken into consideration depending on the area inside the gasket on the waterblock (15 pounds of force for each square inch). The other thing that comes to mind is the large amount of power (aside from the cpu) that is running through that motherboard. I work with a UHV system and although there is much less pressure in the system I use then what you would be using remember that virtually no heat will be collisionally transfered away from the motherboard from the air inside the chamber. It may be interesting to see how hot things get and how fast, but you might also see how fast things melt, so it is something else to think about.

bigben2k
03-31-2006, 08:49 PM
Thanks for the reply.

I really haven't given this idea a lot of thought, that's why I threw it in here. Maybe if I throw a few more thoughts:

First, there's (essentially) not going to be any air, so there will not be any convection cooling.

The other type of heat transfer is radiation (i.e. the infrared stuff), but it's not a lot of energy.

So everything will heat up, and (essentially) none of the heat is going to go anywhere. That means that the temperature is going to keep on rising, in each and every component of the motherboard, through which power flows.

Let me throw in another bone:

Let's assume that 20% (as a guesstimate) of the power applied to the mobo-CPU combo comes out in secondary losses.

And another one:

In UHV (Ultra High Vacuum), everything works differently: there is no movement of air based on pressure differences. All of the components are rated for their leak properties (at this level of vacuum, everything actually leaks, even solid SS walls). To top it off, all of the elements are often baked, to degass them, otherwise they would add some kind of gass in the vacuum chamber.

What's going to degass from a mobo?

Should I be using a different type of TIM material?



I'm still months away from setting up the vacuum chamber, and I want to make sure that I do it safely (it can be dangerous). I'm ruling out a glass bell jar as unsafe, even with a protective cage, unless someone can explain to me how it could be used safely.

bobo5195
04-01-2006, 08:33 AM
radiation is T^4, so at low levels it is not alot but if things start getting hot then they could transfer a significant amount of stuff this way. might be worth pluging some numbers.

Also you sure your not going to get secondary additions in a vacuum chamber. ie the heat from stuff like voltage regulators going through the water block.


I am very worried about the mobo frying. are you sure there are no better ways of doing this?
Hmm need to have a little think of a better approach. COuld probably do something with case temp and water temp over various runs to quantify results need to use a bit of the old science as well.

bobo5195
04-01-2006, 08:52 AM
Had a little think any you should be able to quantify secondary losses. Send me a PM with an email and i'll give you a paper i handed in for CFD.

Basically heat transfer rate from a vertical plate by convection is

Nu = C*Ra^n

c we will never know but n is between a 1/4 and a 1/3.

Ra = Rayleigh nuber, reynolds number for natural convection

Which is
Ra= Gr*Pr

Gr is grashof number and is a non dimensional number based on the amount of energy added to the system by convection
Gr = g * (1/T) * (T surface - T air) * L^3 / viscosity^2

We can hence say that the heat transfered from the board varies by a certain range and with a few emperical results could possibily work it out.

qet77
04-01-2006, 01:38 PM
Well, the system that we use is UHV so there are certianly big differences. An ion pump is used to get to very low pressures (10^-10 torr), but we do not have anything inside of the chamber that is powered. The one temperature dependant thing that is of concern is desorption of metals on the inside of the chamber. The rate of the desorption scales with the double exponential of temperature. I know of some people that had some problems with heat removal on a system that used much lower vacuum then we use ~10^-7 torr and a measurement component kept melting (very little current was being run through this ~microamps). So as the others mentioned the heat can add up fast. Is there some other calorimetric way of doing this in a highly insulated container?

bigben2k
04-02-2006, 12:29 PM
Hey thanks guys, this is good stuff. It's starting to look like the mobo would just fry.

In this test, I'd probably run a full CPU load test, so there would be a significant amount of heat there. If I only ran a low power test, we're still talking about a lot of power.


Following your suggestion (get77), I gave some thought to an immersed test run, but I can't imagine what fluid would be both dielectric, and a low thermal conductor, for a reasonable price.

It finally just hit me; I started this idea with a target vacuum of 1 millibar; what if I progressively lowered the pressure, starting from atmosphere, take some temp measurements around the mobo's surface, and see what happens as I drop the vacuum in 100 millibar (10% of atm) increments?

I've got some mylar sheeting that I could use to isolate various components from radiation heating.

bobo5195
04-02-2006, 06:28 PM
Get me the properties of air at those pressures and i'll work out how hot things get.

Using dimensional similtude analysis / voodoo magic

Long Haired Git
04-02-2006, 07:07 PM
Dumb idea.

Take one motherboard you don't really like.
Take some perspex sheets.
Glue a box around the CPU socket and seal it up good with goop.
Run the pump for this box only, and let the rest of the motherboard run in normal atmosphere.

bigben2k
04-02-2006, 09:03 PM
Get me the properties of air at those pressures and i'll work out how hot things get.

Using dimensional similtude analysis / voodoo magic
U mean heat capacity?

bobo5195
04-03-2006, 05:42 AM
I mean

prandlt number, viscosity, thermal conductivity, beta

bigben2k
04-04-2006, 05:07 PM
This page should have the answer to all the info you requested:
http://www.engineeringtoolbox.com/air-psychrometrics-properties-t_8.html

Specifically:
http://www.engineeringtoolbox.com/air-properties-d_156.html
and
http://www.engineeringtoolbox.com/gas-air-systems-t_22.html

Not sure how you'd compensate for lower pressures.

Let me know if you need more.

bobo5195
04-18-2006, 07:48 PM
Eeek I’ve left this too long. Got no time now my exams are starting a week odd and works pilling up. Feel kinda guilty if im not at it 12 hours a day.

I was thinking while washing the dishes that all this vacuum stuff is a little excessive and will melt things. Surely you can do it with a standard waterblock and some thermometers.

We know the heat going to the water block (dT across it, strictly there is some heat from restriction but this can be found out using pressure drop data). We know that there is some heat going through the mobo. Ie

Qcpu = Q water block + Q mobo.

There are some implications that could be made off that.

We know wb performance on a heat die so should be able to calculate Qcpu from that using scaling calculations and its temperature as well (bare die, know interface stuff thanks to bill, know water block). Un sure need to think about this abit more.

We also know that heat is going through the mobo. If water block performance varies (vary radiator fan speed to make water inlet hotter or lower pump I’m not sure yet) we would expect more heat to via the mobo than the waterblock. We can measure this difference by using dT across the block.

VRM and heating stuff on mobo is a problem. Was thinking that we can just go linear superposition to get rid of that but my heat transfer is a little rusty.

Bit of heat transfer voodoo magic and it all should be illuminated.

(it’s a vague idea hence the vagueness of this post)

Long Haired Git
04-19-2006, 04:14 AM
I like the idea of vacuum testing, if only to prove or disprove the burn-up theory.
Get an old slapper of a motherboard and CPU you're prepared to toast and see what happens.
Video and post, you'd get some hits for your Google ads...

bigben2k
04-20-2006, 04:25 PM
Yeah, the more I think about it, the more I believe that this idea is crazy.

What's really needed is a way to isolate the CPU thermally, from the rest of the mobo. Heat travels through the pins.

I know that there's a CPU socket extender out there (we've linked to it before), but it doesn't solve the problem.

I'm stomped; I need to give this more thought.

As it is, the mobo components are just going to hit a higher temperature, as the vacuum drops. Then the components overheat, and blow. What's the point? Would I be able to make temp measurements of various parts of the mobo, under various small vacuums, and be able to make something out of the data?

I'm still scouting eBay for a vacuum pump, because I need it for a couple of other things. This idea isn't dead yet.

bobo5195
04-20-2006, 08:50 PM
Might be worth doing some basic specing.

What is it your trying to test for and what gets in the way? etc.

There are loads of approaches to this beyond frying chips. The more i think the more i get worried about things like vid card vrms, southbridge, memory.

Heat conductive paths from nice copper cables out side the vacuum.

If yo had the cash the old asus pm converter might be useful.

bigben2k
04-21-2006, 01:48 PM
Yep, basic spec'ing. I'll try to put something together over the weekend.

Back on Monday.

What is "asus pm converter"?

bobo5195
04-21-2006, 04:17 PM
pentium m socket converter gizmo
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2382&p=2

bobo5195
04-22-2006, 10:50 AM
Another thing ive been thinking off under the haze of revising ultra lean burn stratified charge engines (shudder).

Could try measuring the area around socket for temps as don't they keep it clear for HS mounting these days

Ice Czar
04-22-2006, 11:11 PM
Glad youve abandoned the whole board idea
the VRM\VRDs alone on a modern mobo would burn the board
with no where for the heat to go but into the board, the heatsinks will get hotter and hotter through time till they start to shed more and more radiative energy (http://www.zianet.com/ebear/metal/heattreat4.html) and eventually melt, that is if the VRD doesnt die first
(I have a big autoclave but I wouldnt use it for whole board testing,Im fabricating vaccum test chambers out of dive tanks and other oxygen bottles)

the vacuum chamber idea really is tied to the CPU simulator idea
I picked up a Welch Duo-Seal (http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&item=7607168007&rd=1&sspagename=STRK%3AMEWN%3AIT&rd=1) the other day and have some ideas on both determining the heat gradient across an IHS and then replicating it with a simulator.

(I still don't see the need for a UHV system, Glacerbay's vacuum panels are .05 torr, and reducing it lower doesnt substantailly add to the insulative R value http://www.glacierbay.com/vacpanelinfo.asp we will have no common conductive edges like a dewar flask or panel, just the connections to the simulator and waterblock we could even make good guestimates as to the heat transfer through those by independently measuring them)

Been making quite a bit of progress on the heatsink\waterblock test front (but more on others)
picking up a Setra 230 DP transducer (0-5 PSID) as well as cutting a deal to have a sonic temperature immune "low flow" mass flowmeter fabricated by Eesiflow. ;) (0.3% of range and calibrated to our "typical" coolant mix)

the biggest hurdle I have in my vacuum chamber is finding affordable feedthroughs
the simulator is another matter

consider the problem is backwards
if you know how much energy the block is moving in isolation
and you can measure how much energy the CPU is employing
the difference from the energy being moved in the real world through the block would be the secondary pathways

its measuring the power the CPU is employing that is where it hurts
http://www.chromaate.com/english/product/detail.asp?gid=44&id=892

Colt357tw
04-23-2006, 09:30 AM
how do you cool of mosfets around CPU socket if you running a system in vacuum??

Long Haired Git
04-23-2006, 11:55 PM
BladeRunner water cools his mosfets...

Ice Czar
04-24-2006, 12:13 AM
point would be that youd need to cool every chip since the only heat path is into the board, the VRMs are just the biggest source after the chipset, just a matter of time till each chip dies if power is able to be maintained to the VRD

bigben2k
04-24-2006, 04:18 PM
You're on the right track, Ice Czar. If I could measure the dT across the water block (along with flow rate) and knowing the applied power, I could calculate the difference.

The problem is measuring that dT, because it's so small. Maybe I could make a run at a very low flow rate (i.e. 1/4 gpm or less) so that there's something to measure...

Good deal on that vacuum pump.

Ice Czar
04-27-2006, 12:37 AM
energy is small if you measure it in small amounts :p

you could for instance fill a bucket full of energy and measure that
the other great hangup I have for any block testing is no lab chiller
you cant measure a difference if you don't know where your starting from
measuring on the "fly" isnt as easy as measuring in larger volumes, you can always do a good job of insulationg
and you can workup an error correction as well

http://www.robertsoxygen.com/htmlfiles/VJ%20PIPE.html (can be had in flex too, dozens of manufacturers)
pump into a dewars container with a known rate of loss

the monday following Easter someone made an error and stuffed a NesLab HX-150 in the wrong spot on ebay, that combined with it more or less being a holiday led to a $5000 list (and it looked new) recirculating lab chiller going for some $350 :rolleyes:

of course shipping 320 lbs halfway across the country was how I got outbid by $5 in literally the last second (I hope that was you :p )
which brings me to ebay notifying me of an end to an auction one day after it closed
a tested and functional HP 6050A mainframe didn't sell with $49 starting bid :rolleyes:

but its all good, Im feeling much better now and have time to fix the holes in my wall
I landed this (http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&item=7611942757&rd=1&sspagename=STRK%3AMEWN%3AIT&rd=1)

I'm playing around with the software, memorizing the manual and counting the days till it arrives, Im also seriously pimping for 20 or thirty cases to review, I need to stop spending money (http://forums.techpowerup.com/showthread.php?t=11262) and start making it, need another specific DMM (HP 34401A) so I can use the software I have for my Autotest UTS-325

oh an Kyle shitcanned me :p
but W1zzard picked me up in a heartbeat
need to change the name of the lab

bigben2k
04-27-2006, 06:28 PM
Nice find on the Temperature monitor. I've got a couple of Fluke 2190A and scanners; am working out a deal to acquire a spool of type T thermocouple wiring and some RTD sensors.

Yeah, budget is limited here too, otherwise I would have grabbed at least one of those NI cards. Right now it looks like it'd be quicker for me to build that 24 bit ADC circuit!

Ice Czar
04-27-2006, 07:31 PM
anemometer array :p

I already can do temps, but being able to profile airflow and temps at the same time
now that makes me happy

here is your thermocouple wire guy
http://stores.ebay.com/AA-Appliance-and-Industrial-Supply_W0QQssPageNameZstrkQ3amefsQ3amesstQQtZkm
BillA turned me on to him, I picked up 100' of type T

bigben2k
04-28-2006, 02:15 PM
Yep, that's the one. I'm probably going to pick up the rest of that spool: ~185 feet.

Any idea if there are issues using stranded wiring to form a thermocouple?

Ice Czar
04-28-2006, 04:27 PM
I would assume so,
consider that there isnt a uniform contact at the junction without say welding, even then it likely pretty unique
but under a screw terminal its all compressed together and just another electrical signal, I still would have perfered single wire like my J type extensions

my T type sensors are actually surface mount, imbedded surface mounts at that
(little blocks of aluminum (http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&category=119119&item=7593721116)) exactly how Im going to employ them Ive yet to detrmine without some testing, but Im thinking I'll machine them as needed and directly mount to various chips with a "cut" thermal epoxy. Need to get a few similar new surface sensors for comparative testing to determine response lag.

there was no going wrong for $20 :p

the T type panel jacks and plugs (three prong w\ ground) from Omega will run over twice that,
plus sheilding, plus overbraid, plus the extension wire, but I have those already

bigben2k
04-28-2006, 04:50 PM
Hum... I was wondering... if the Omega probes are "butt welded", maybe I could zap/fuse them together, using a high voltage source? What do you think?

I've never made my own thermocouples before, and I'm finding myself lost here. Worst case I can get real probes, and use that wiring for extensions.

Ice Czar
04-28-2006, 06:04 PM
Ive had a thermocouple welder on my acquisition list for quite awhile, but they aren't all that cheap on ebay generally, and its not a very high priority, its factory automation and control engineering where they are doing hundreds per year Id imagine that it comes in handy, but then Ive also seen thermocouple "kits" with SS probes\wells, mineral insulation, and of course sensors.

The advantage there is you fabricate grounded, ungrounded ect packages yourself, but typically the sensor itself is factory made and you connect to it (which is where Id use a welder)

While I have no real idea of specifically what the difference would be between simply a wrapped vs welded vs factory fabricated sensor
I would assume its a closer repeatability with less error rate. and that a wrapped solution is a "close enough" general temperature type approach. Handmade welded sensors if they where subsequently calibrated and the error factor was coded into the aquistion system Id imagine would approach the same fidelity as a factory manufactured sensor that is consistently made from unit to unit in a more controlled manner. Simply with less error to plot those are "close enough" out of the box Id gues by at least a decimal place or two. (6.5 DMM)

Just a little dead reckoning ;)

bigben2k
04-28-2006, 06:25 PM
Looks rather simple, check out this oldie:
http://www.overclockers.com/tips437/

I'll make my own for the air measurements, for now.

This welder's insanely expensive:
http://www.tiptemp.com/Product.aspx?ProductID=4002
(TIG welder)

Another one, slightly cheaper:
http://thomasregister.com/olc/smartcat.aspx?az=17183047&template=/olc/17183047/two.htm&type=list&ptno=10-A10120+10-A10240+10-A20000+20-A10120+20-A10240+20-A20000&stripecolor=99CCFF&backlink=/olc/17183047/hotspot.htm&header=Click%20on%20part%20number%20for%20details&textcolor=000000
with details here:
http://thomasregister.com/olc/17183047/twospec.htm
A 35 Vdc to 75 Vdc source, 250 Joules (watts/sec) capacitive discharge.

Ice Czar
04-28-2006, 06:41 PM
since Bill is away I guess I'll need to play devil's advocate :p

as you are of course aware, we are now talking pretty small differences in one DUT vs another, and Id imagine that even though the primary path would be whichever two thermocouple metals are employed some of the path would be through the solder. Id also want an "ungrounded" sensor for directly embedding in a heatsink and or multiple surface mounts

used welders on ebay run $200>300
"as is" nasty ones occassionally much cheaper, that would still buy one hell of alot of bare sensors

I bought 40 epoxy coated and 40 glass rod thermistors (http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&item=7608769628&rd=1&sspagename=STRK%3AMEWN%3AIT&rd=1) the other day planning on building some "grids" with them that can be installed infront of, in back of and inside of cases \ fans. I'd be able to do 50 channels at a time with the 7701 (would need to build 2 cables)

close enough for rock and roll is really a function of what it is undertest, measuring airflow Im just trying to get a basic picture, but in WB tests there are alot of "photo finishes" which I think demands as high an accuracy as can be managed, there are alot of unaddressed variables still, and the "general picture" pretty well mapped already

bigben2k
04-28-2006, 06:52 PM
I like the way you think ;)

If I went the way of the solder (which still looks dubious to me), I would expect the solder to impact the response time of the TC, but otherwise nothing else. A big assumption...

I much, much favor the welding method, just because it's a simpler no-nonsense worry-free approach.

I just have to figure out something, somehow. We've posted links before, to spot welders. There's got to be an easy way to weld thermocouples. Google, here I come.

Ice Czar
04-28-2006, 07:13 PM
if you want to know my real fear
its that the embedded chip cooling technology will change before we are really able to find an agreed on proceedure, IHS is just the first step, Im not even sure "waterblocks" as we know them are going to be that relevant for too much longer. There are so many developing technologies from Diamond Substrate Chemical Vapor Disposition to Carbon Nano advances in materials to nano heatpumps, dual phase microchannel and spray cooling that with the ever greater increases in power I think we are going to see integral thermal solutions sold with CPUs sooner rather than later. Intel's "waterblock" system was just the first. The IHS is basically a heatpipe base, its currentlty transferring to a seperate component but for how much longer? I think there will be a trend for the IHS to "grow" into a larger interface, that attaches to something else, possibly several options of something else.

remember alot of what drives thermal issues chip development comes from incresing densities in 1U and now blade servers
but once tooled into the fab gets transfered to the desktop

thats why Im really more interested in being able to determine the power consumption and develop a system that measures the total amount of heat whatever tech that is applied is moving to the case (or outside the case)
my "waterblock" test suite is really looking more to that direction. And I see vacumm insulation as an important part of quantifiying and isolating variables.

bigben2k
04-28-2006, 07:25 PM
Got'em:

http://www.5bears.com/welder.htm
and
http://users.frii.com/katana/spotweld.html

Ice Czar
04-28-2006, 07:30 PM
I wanted a spot welder. I nearly NEEDED one. :p

you cant link that here without also linking this
http://www.5bears.com/cnc.htm

everything duely bookmarked ;)