Apogee from Swiftech...

[nikhsub1]

Bill I thought the heat die has the simulated IHS built into it? No need for anything else. As for protecting the surface of the die, this is prob a bad idea, but I'll chuck it out there anyway. 3M makes aluminum tape, not sure of the gauge but it is pretty substatial yet bendable/plyable. Obviously this introduces an added layer but a simple test could be done for correction factors. I think the tape would do wonders in at least protecting the surface of the die, but perhaps more trouble than it's worth?

[Marci]

to use a heat die, and then affix a salvaged IHS to that...

Not my point - point was so Bill had an A64 to remove IHS off and assess quality of current mounting solution (bond of IHS to CPU / Wafer and TIM Joint) to base a more qualifiable replication on in terms of mounting method only.... not to actually use any of the parts for a test.

The testers here in this thread wouldn't be stupid enough (I'll assume on your collective behalves) to use a salvaged IHS.... they know it'd immediately rule out any data produced.

[BillA]

we have multiple subjects going on (I think ? lol)

if the IHS is loose it solves the wear/maint issue (by replacement)
if the IHS is the die top then there is a problem that may preclude progress

how will one 'know' that the die is flat ?

in ANY case, how is the IHS edge to be supported ?
is the mechanical issue understood ?
put this heat die into a BTX mounting frame

[Marci]

understood... no answers/suggestions... but understood!

[nikhsub1]

Ahh yes Bill, I too was confused but do understand now. If a real IHS is to be used, should be fairly easy to get the Intel spec for die size above PCB... have the heat souce 13mm sq or whatever is agreed upon, raise up that height above surronding area. Now, what if an IHS 'groove' were created so the IHS justs sits in the groove firmly, IE no movement in any direction possible besides UP. I actually think you are onto something here Bill.

[bigben2k]

A proposal:

Let's use a heat die, and a replaceable IHS.

Since the mating surface is only good for 20 mounts (as per Bill), that means replacing the IHS after 20 mounts (i.e. 2 to 6 rounds of testing, depending on the number of remounts one does), then every time that the IHS is replaced 20 times, re-surface the heat die.

Advantages:
-easy to replace an IHS, instead of the whole heat die.

-temp probes can still be used (if desired) to measure the heat flux (Incoherent style) within the heat die section.

Disadvantages:
-replacement IHS must be precisely the same.

-Makes Intel style temp measurement hard to reset exactly the same way.

-Adds difficulty if a temp probe is used for a core side temp measurement, within the heat die, due to the regrind (do the core-side temp measurement Intel style instead).


We'd have to find a good steady TIM material for the heat die to IHS connection.

We ought to contract someone for making the IHSs, pre-load them with a thermocouple, and run a calibration test, with deviation report.

Note that the option of using something other than copper for the IHS, or the heat die for that matter, is still open.

(There was something else... I'll update this post later.)

[nikhsub1]

WEll ben since the IHS's are copper, I see no reason to use something else?

[BillA]

an IHS lasts indefinitely, its surface is huge and the edges non-critical
if you ever get a flat die, leave it alone
if a die is used it must have multiple in-line sensors to extrapolate the face temp (see Incoherent's setup for an eg, Intel and IBM have published examples)

this is a monster already, I'm gonna let someone else make a prototype

some new CPUs have significantly reduced clamping force
and why ?
once the IHS 'crown' is deflected into the same plane as the IHS edges (i.e. the TIM joint is compressed as designed), the additional load is transmitted to and through the package to the mobo
this is pointless hence the reduced clamping forces IMNHO

I do think a parting of the ways is appropriate for those using an IHS vs those not
testers can cater to one group or another or both as they see fit
this could perhaps 'justify' lower resolution testing on CPUs, and more rigorous work on a heat die
??

[nikhsub1]

I have the old style Intel IHS if anyone wants it for this.

[bigben2k]

Originally Posted by nikhsub1

I have the old style Intel IHS if anyone wants it for this.

Yeah, unless someone else is going to put it to use right away.

I can see mounting becoming an issue, real fast.

[Albigger]

well it seems finally some good progress.

I like the idea of heat die with a standard IHS. should use some insulation board or similar for the IHS edges to rest on, so as not to transfer any heat through the supporting edges of the IHS. The insul. board would be permanent and grooved for locating the IHS edges possibly (as per Nik).


With regard to making the heat die flat - what methods have been used before? Is anything proven? The only thing I really know about is for example using a fine grinding wheel, but I'm not sure that meets the tolerances required here?


Regards to many testers and if there should be a standard heat die design - well standards are good yes, but if some flaw is inherent in the design then this will be replicated. Maybe better to produce guidelines for heat source and explain differences later (obviously things such as how flat the surface are need to be specified, but for example as discussed the heat source size may very depending on what size core one is trying to replicate).

[bigben2k]

Grinding wheel is fine (lapmaster, ...) but measuring the end result requires equipment, which only BillA has, and 8ball has access to.

I've been trying to figure out a way to measure flatness, but I'm down to a rotating laser beam reflection, to a projection screen, which would require a nice reflective surface (on the die). Not sure if it's practical yet. :shrug:

[talcum]

Originally Posted by pHaestus

wow still the same thing again about the ttv?

Am I hearing from Bill that using the Intel CPU would be preferable to an AMD one for my test bed? Because the limitations of the intel platform are pretty large ones: 1C resolution, complete reliance on the motherboard temp monitoring, and no access to the "hot" diode only the cooler diagnostic one.

I would think ditching the IHS (or popping it and reaffixing it somehow?) on an AMD processor would still be preferable to the above problems?

Not sure how easy it will be to "pop" the Intel IHS if it's one of the new diamond/copper composite (like GLIDCOP) ones. Maybe stick to AMD if IHS is to be removed, pH?

[Albigger]

Well, here's the lines I was thinking of for a setup, not sure where temp measurement would occur (several places accoring to Bill - I'll have to read on Inchoerent's setup...)


It should not be too hard to source someone to test for flatness, or to send to Bill for appraisal?

[Albigger]

Originally Posted by Annirak

I still maintain that through the judicious use of insulation, a closed system could be made in the coolant loop, and the power out of a conventional CPU could be found, allowing conventional CPUs to be used for WB testing purposes.

I don't see why this couldn't happen either, if you can measure flow and temperatures (temp differentials at least) very accurately. I may not have followed your earlier analysis completely but here's the way it seemed to me:

One could measure the radiator water temp differential in/out to know how much energy is dissipated there. Also measure the temp differential across the pump inlet/outlet to know how much heat is added (or setup an inuslated closed system previously and measure the temperature increase just do to the pump and friction of the water). Finally measure the block inlet/outlet temp to see how much more heat is added inside the block (mostly presumable due to the cpu, somewhat due to the block friction - this could be attained as well).

And the final check should be
pump_Energy + cpu_Energy = rad_Energy
Assuming no losses through tubes etc....

[shadowing]

Originally Posted by Cathar

We shall see.

I agree with that. I am eagerly already waiting for some true tests instead of discussions.

[snowwie]

so what are the true tests?

roscal has already said what robotech's results will be and why. AND how it fits into the picture.

who else is there?

[BillA]

well,
Cathar said the results were due to a flawed test methodology
and I said the methodology was fine and the Swiftech results 'good' data (quite reproducible) but not (apparently ?) intelligible - missing a correlation permitting these numbers to be compared to others

the greater picture ?
if the Apogee cannot beat the Storm in CPU testing (as Swiftech's testing indicates), they have created for and by themselves a PR debacle
does Gabe have the balls to admit an error by him ?
and Cathar and I are quite in agreement as to the specific error

(I failed to think about what Roscal was saying, and why, else this would all have been more clearly defined from the onset)

[Marci]

and Cathar and I are quite in agreement as to the specific error

Care to share that biscuit??

[BillA]

removing the Storm from the Apex kit

[RoboTech]

OK, I’ll take a little break from testing and share some of my thoughts…

Fact: Thermal die simulators are just that – simulators. They can provide useful data (especially for waterblock designers) but rarely give a true representation of how a waterblock will perform on a particular CPU under specific conditions. Intel has the resources to build a better CPU simulator (the TTV), which should more accurately reproduce some of the dynamics of a real processor - but its still just a simulator. I agree with what others have said: what we really need is much better correlation between the simulators and real CPU’s. IMHO simulators will continue to be useful and different simulators will produce different data.

I have been wrestling with many of the issues being discussed in this thread for some time. I know my current 14mm naked copper heat die is becoming outdated and have given a lot of thought to its replacement. Here is what I have come up with so far (see attached sketches).

New Features:
+ Incorporates a removable copper heat spreader
+ Extended die height so extra sensors can be added to calculate heat flux
+ Thermocouple (or RTD) is mounted into the heat spreader (this eliminates variability’s in the heat die to heat spreader joint)
+ Heat die is spring-loaded to insure good contact pressure at all times
+Integrated 0.01 ohm, 50W shunt resistor for more accurate Amp measurement

Standard Features:
+ Two cartridge heaters (0~300 watts)
+ Copper die
+ Phenolic deck (very rigid, easy to machine, OK insulator
+ 6-32 threaded inserts (mounting patterns for popular AMD and Intel CPU’s)
+ Ceramic felt insulation (Cotronics)

To Be Determined:
· Heat spreader surface area (32mm sq?)
· Heat die surface area (14mm sq?)
· Spring load force (~20 lbs ?)
· Spacing and location of optional temp sensors (to measure heat flux)

This 4th generation thermal die is intended to better simulate current processors that use an IHS and have larger die areas. It also provides a much more robust test surface that should hold up longer before neading to be re-lapped or replaced. The temperature probe is mounted into a tiny hole drilled into the bottom of the copper heat spreader and is potted with thermal epoxy. The temp probe lead wires exit thru a hole drilled up thru the center of the heat die. This places the temp probe ~0.5mm from the top surface and eliminates the variability’s caused by the TIM joint.

I envision making/using a separate tool to hold the copper IHS during machine lapping.

To address the problem of consistently seating the waterblock being tested onto the thermal die simulator without adversely distorting the IHS, I included a spring under the heat die, which pushes up (exact force TBD) insuring the heat die is making good contact with the bottom of the IHS (and waterblock base) at all times. When the waterblock is first mounted it will engage the top surface of the IHS and push the whole assembly down (~1/8”?) until the outer rim of the IHS firmly seats against the insulated Phenolic deck. The spring pre-load force pushing the heat die up is adjusted by: (1) spring selection and (2) length of standoffs used to mount lower spring platform.

The top Phenolic deck plate uses threaded inserts located at the common CPU mounting hole locations (just like my current design). Alternately, a universal mounting frame could be attached to the deck to allow clip-style mounts to be tested (I don’t plan to go this route).

The lower heat die and upper Phenolic deck are covered with ceramic felt insulation (1/8” thick), which allows the heat die to slide up and down thru the square opening in the deck. The entire assembly is then surrounded with ceramic fiber insulation and housed in a small enclosure.

Side note: Today is Thanksgiving Holiday in the USA… HAPPY THANKSGIVING!

Thanksgiving is an attitude… I am very thankful this day for the MANY blessings God has granted me (family, friends, a job, freedom, good health… and access to a global community of PC enthusiasts like you!)

[BillA]

nice concept Lee
would set the sensors differently but I think the mechanicals are ok (max the thickness on phenolic)

yes, we are fortunate and should be a bit humble

[Marci]

Hmm... mini shock absorber... sealed pneumatic would poss produce more consistency? Less wear over time should spring lose it's tension (? if an issue at all... not an engineer *shrug*)

Know you can get mini-shocks for use with FloydRose style locking trems on guitar... to replace the usual 4x springs... can't recall exact name - ensures trem always returns to same point whether compressed or extended.

EDIT: Tremsetter - http://www.hipshotproducts.com/tremsetter.htm

[Kbofwy]

Originally Posted by RoboTech

+ Thermocouple (or RTD) is mounted into the heat spreader (this eliminates variability’s in the heat die to heat spreader joint)

I'm not understanding how this will remove the IHS-heat die variable from the end data, can you elaborate on this?

By having a fixed measured input wattage in the heat die and the TC in the IHS, a poor IHS/die contact is being directly measured and the final data will have this error included to only appear as an improved IHS->waterblock c/w.

[Orkan]

Got my apogee from sidewinder... and decided to strip it down this turkey day to see what it's all about.

This pic shows how they intend the water flow to head across the pins. Wonder how it would perform if you flipped it the other way... top->bottom instead of side->side as you see it now? More resistance maybe?


Here you can see it next to a poly top for a WW block. Roughly 2/3 the size.


Upon closer inspection... I found a sign of pure greed. Rushed manufacturing process. The pin array is riddled with metal shavings due to a rushed milling no doubt. I have circled a couple, but they are EVERYWHERE. They are not loose, but rather attached to the block/pins... normally caused when a saw blade, or machine control is pushing the mill bit too fast, and not allowing it to do its job.


I took a tiny flathead screwdriver and scraped out every single track between the pins. Took some time... but I wouldn't want this crap stuck in my radiator:


Its clean now, but that metal shavings crap really has me doubting my purchase. So if you purchased one of these, make sure to clean that sucker up before you stick it in your loop.