CPU cooling with PEP66

[lifeguard1999]

Given:
1. An Intel FC-PGA on a slotket.
2. Clearance in the case & motherboard

Goal:
1. Allow the maximum in air cooling for overclocking without raising the noise level.

Assumptions made:
1. The Intel version of the Alpha PEP66 is the same as the AMD version of the Alpha PEP66 reviewed on overclockers.com.
2. C/W does not change if the Watts output by a CPU decreases.

Then:

An Alpha PEP66 with a 27 CFM fan has a C/W of 0.23. An Intel 550E MHz radiates 14.5W = ~3.3 deg C over ambiant.
An Alpha PEP66 with a 27 CFM fan has a C/W of 0.23. An Intel 1.13 GHz radiates 35W = ~8 deg C over ambiant.
An Alpha PEP66 with a 33 CFM fan has a C/W of 0.19. An Intel 550E MHz radiates 14.5W = ~3 deg C over ambiant.
An Alpha PEP66 with a 33 CFM fan has a C/W of 0.19. An Intel 1.13 GHz radiates 35W = ~7 deg C over ambiant.

Yes, I know that the Intel 1.13 GHz is nonexistent. I am using it as a worst case.

It is intersting to note that the temperature drop is at most 1 deg C. This is not too surprising given that the test setup for the reviews on overclockers.com used a processor generating a heat load of 69.2 Watts. The difference between the 27 CFM and the 33CFM fans was 2.6 deg C. Half of the heat load is approximately equal to a 1.13 Ghz; half of the temperature delta is 1.3 deg C. This is close to the 1 deg C calculated above. So the math checks out.

Now a Papst 8412NGML 80mm fan is rated at 21.0 dBA and 26.5 CFM. That is approximately equal to the 27 CFM flow used above. The noise level is less than that of a whiser (23 dBA). The only problem is that this is an 80mm fan, and the PEP66 uses 60mm fans. No matter. All I have to do is buy/make a converter from 80mm to 60mm.

Now a Papst 8412NGM 80mm fan is rated at 26.0 dBA and 34.1 CFM. That is approximately equal to the 33 CFM flow used above. The noise level is louder than that of a whiser (23 dBA) and is the same as that of a Golden Orb. Again, the only problem is that this is an 80mm fan, and the PEP66 uses 60mm fans. No matter. All I have to do is buy/make a converter from 80mm to 60mm.

This should allow me the maximum in air cooling for overclocking without raising the noise level. Depending on which fan I get, I will also replace the 80mm fan in the power supply with the same fan so that the gains made in using the heatsink are not washed out by a noisy power supply.

Questions:
1. Does C/W change if the Watts output by a CPU decreases? I do not think so.
2. Anyone see a problem with anything?

 

[Mikewarrior2]

Where did you get the C/W Numbers?

Those are awfullly low, but it also depends on how it was determined? If it was determined with a thermocouple drilled into the heatsink, it won't really accurate represent the C/W of the primary heat patwhay(ie Core to CoreTop-Heatsink), which is more like ~.34-.35ish.


Mike

P.S. You'd probably see a 933@1.7V run somewhere around 9-12C over ambient case temp, depending on what kind of grease you are using.

 

[lifeguard1999]

I was going by what overclockers.com said in their review to get the C/W numbers.

Alpha PEP66 Fan Options

Their testing methodolgy was described as :



<< I ran Prime 95 for about an hour for each test and recorded CPU temp and ambient temps using the Omega HH23 Digital Thermometer. CPU temps were recorded by a thermocouple drilled into the heatsink exiting at the CPU's core top. >>



So why the difference between their numbers (0.19 ~0.23) and what you say 0.34~0.35?

 

[Mikewarrior2]

because they're using a thermocouple in the heatsink. They are measuring &quot;heatsink&quot; c/w, but it doesn't take into account thermal resistance between core and core-top. Taking that into account, the rough C/W(which is impossible to really calculate) is approximately .3x ish for the entire primary heat pathway.

Ie, you will likely see a warmer core temp than 8C over ambient with a P3-1.13ghz. (especially if you calibrate your diode temp reading).


Mike

 

[lifeguard1999]

That makes sense.

It would also make sense that the hottest part of the CPU may not be exactly in center of the CPU or where the diode is located in the CPU. I wonder where the hottest part of the CPU is located. The L2 cache? The ALU? I wonder where the diode is located in the CPU. Just thinking out loud...

Anyway, back to my project, which is having a quiet computer with excellent air cooling.

Assuming a C/W of 0.35 (your number) and a C/W of 0.40 (just a guess for the 27 CFM fan case) with a 33/27 CFM fan for a mythical P3 1.13 GHz/550E radiating 35W/14.5W, then...

With a 27 CFM fan and a C/W of 0.40. A 550E MHz radiates 14.5W = ~5.8 deg C over ambiant.
With a 27 CFM fan and a C/W of 0.40. A 1.13 GHz radiates 35W = ~14 deg C over ambiant.
With a 33 CFM fan and a C/W of 0.35. A 550E MHz radiates 14.5W = ~5.1 deg C over ambiant.
With a 33 CFM fan and a C/W of 0.35. A 1.13 GHz radiates 35W = ~12 deg C over ambiant.

So maybe with a P3 1.13GHz chip an Alpha with a large amount of airflow is needed. With a 550E, the difference of 0.7 deg C is negligible. I wonder if a fan is really needed for a 550E or if radiant cooling &amp; the air being exhausted out by the PS fan is enough. I will bet that a (mythical) 0.13 micron 550E would not even need it.

 

[Mikewarrior2]

12-14C over ambient is very good, and doesn't require large amounts of case cooling... Two fans(my prefered minimum) is probably all that you'll need.

The hottest part isn't the L2 Cache, its probably a mix of ALU and FPU functions. THe diode is embedded in the core(the core beling that slightly raised &quot;nudge&quot; on hte p3 pcb). Internal diode reports core temp better than anything externally mounted. L2 cache, however, is one of the cooler parts of the cpu core.


Mike