Measuring CPU heat output.

[Les]

From curiosity have included pHaestus's latest( 02-07-2004) MCW5000A results. The data was "eyeballed" from graphs presented here .

[pauldenton]

Originally Posted by pHaestus

I think I might have an explanation for the high W with the Cascade. When I removed it the other day I accidentally touched the northbridge hs (passive) and burned my finger (ok not to the point of a blister but it was red for a bit). The cascade and other 4 hole blocks are really close to that nb heatsink while the MCW5000-A is not (it's square rather than rectangular). So my guess is that extra heat from nb (as much as 10W though?) is being dumped into the cooling loop, and that the better the wb the more heat goes in (from copper traces on mobo and possibly from air as well. I'll put a fan on that hs and see if the numbers are a little more sensible. My cascade sprung a leak though and currently my test system is being air dried.

hmm - maybe you need to w/c the northbridge (on a separate loop) to make sure it can't affect things.....

[wimms]

Hi, somehow reached this thread.

P4 has 2 special pins: VccSense and VssSense.
VCCSENSE - Output VCCSENSE is an isolated low impedance connection to processor core power (VCC). It can be used to sense or measure power near the silicon with little noise.

VSSSENSE - Output VSSSENSE is an isolated low impedance connection to processor core VSS. It can be used to sense or measure ground near the silicon with little noise.

If one makes precision measurement of resistence from that pin to voltage regulator (with CPU onboard, unpowered), one can use that resistence as a known shunt resistor. During operation, measuring voltage on these pins and voltage reg must give you exact voltage drop across the vcore power path, thus it must be possible to calculate power consumption. Vss might be even well suited to be wired directly into analog input of a sound card if that supports DC input.

Wouldn't that be enough to get decent measurement of core power consumption?

Even if voltage drop along the Vcore path might include power going out the cpu via data/address pins, Vss would include only power dissipated inside the core, thus should be representing 100% of CPU heat output. No?

[Incoherent]

It's a good idea but I am not so sure that you can do this. The unpowered resistance will not be correct I think. When the power is on there are lot of things that change, as transistors start switching. Could try it though and see what kind of reading you get.

[wimms]

Quoted description was from P4 478 datasheet. As per Intel, these pins are specifically to "be used to sense or measure power near the silicon".

I guess that they have near-zero resistence to points closest to the core where all the current flows through. Resistence to the Vreg would be in milliohms as per wiring resistence. The problems I guess would be high frequency noise and low voltage difference (100mV). There aren't any active components associated with these pins, thus powered or unpowered, no difference, unless conductivity of wires changes under the load.
I wonder if its possible to actively (re)measure ground wire resistence while on power...

Anyway, this should be equivalent to what has been suggested here before (inserting known shunt wire in series with the Vreg inductors), only that the shunt is already there, you just need to measure it.

[BillA]

pH
wimms' suggestion seems to make sense, have you tried it ?

[pHaestus]

Does AMD socket 462 have this as well though? I can't do the homemade reader thing on P4s (pins too small and they don't even give external access to the good diode...

[Butcher]

Originally Posted by pHaestus

Does AMD socket 462 have this as well though? I can't do the homemade reader thing on P4s (pins too small and they don't even give external access to the good diode...

From looking at the datasheet, it doesn't appear to have them.

[pHaestus]

yea I didnt see them either.

[Groth]

The whole reason they give the option of sensing voltages at/near the silicon is because the copper traces on the board have a big tempco, and you want your V-reg system to supply the same voltage at the silicon regardless of mobo temperature.

Ain't too useful, unless you plan to actively cool the mobo to a specific temperature, or you have a scheme to measure the trace resistance vs. a temperature map, or...

[BillA]

damn, I keep forgetting you're on AMD
sorry, should have left this to sleep

[wimms]

Originally Posted by Groth

The whole reason they give the option of sensing voltages at/near the silicon is because the copper traces on the board have a big tempco, and you want your V-reg system to supply the same voltage at the silicon regardless of mobo temperature.

Ain't too useful, unless you plan to actively cool the mobo to a specific temperature, or you have a scheme to measure the trace resistance vs. a temperature map, or...

No Groth, check your math, check some MB design guides. Temco of MB wiring is low enough to fall into cpu tolerance range. Its not compensated like that. Most (if not all) MBs don't even use these pins to regulate Vcore. Intel reference designs advise to take feedback from specific VCore pin under the socket, not from these Vsense pins. Resistence of power path is in 3 milliOhm range, impedance of P4 Core is in 30mOhm range. The voltage drop due to tempco is under millivolt or few.. totally nonissue.

For some reason they do not want to use Vsense as Vreg feedback signal. It must be good reason, though I don't know it. Perhaps to avoid oscillation issues.

These pins allow one to measure combined resistence through the socket for power path, which is specified with quite strict tolerance. But they are meant to measure precise voltage across core. Maybe they are used to report the Vcore by chips like Winbond.

Anyway, even though AMD has no Vsense pins, the principle is still sound - take any pin on socket for both Vcc and Vss, measure voltage across it, and measure voltage across Vss path back to Vreg. Knowing resistence of power path to Vreg, you can arrive to cpu power. By any means, precision of measuring electrical signals wins over measuring temperatures directly any time.

[Groth]

Yep, your right, can't be used by the on board volt-reg. And, yep, oscillation is the issue.

On the other hand, while the trace resistance has to be controlled tightly to keep the Vcore in spec, they aren't concerned with the tempco (results in mV changes).

At ~0.004/K the temperature coefficient of resistance is way too high to for the power traces to be used as a current sense resistor. My error budget for current was 1%, a 2.5°C (averaged through the trace, weighted by current) deviation would blow that before I even got to the instruments.