temps and TDPs

[VirtualLarry]

If you replace one CPU with another, and use the same cooling solution, and the temp as measured by CoreTemp goes down, does that necessarily mean that the overall power draw went down as well? I believe so, personally, but I wanted to get comments on my theory.

The interesting thing is, my E2140 @ 2.8, 1.425v BIOS, got up to 86C in CoreTemp on both cores during the summer, running SeventeenorBust.

I replaced it just recently with a Q9300 @ 3.0, 1.225v BIOS (stock voltage), and under full load (F@H SMP+GPU), it gets up to 71C max so far.

So from the temps alone, can I conclude that my slightly overclocked C2Q 45nm draws LESS power than my heavily overclocked and overvolted C2D 65nm chip?

What's even more interesting, is that the E2140 is a 65W TDP, while the Q9300 is a 95W TDP.

http://ark.intel.com/Product.aspx?id=29738
http://ark.intel.com/Product.aspx?id=33922

 

[Ben90]

Maybe and possibly. There are a few things I would consider for the change. The first is that the load does not appear to be with the same program. There really is no realistic application that produces 100% load on a CPU. Even the beastly Linpack has some stalls here and there. Distributed computing may not be the best to compare temperatures with as I have known drastic changes of temperature with different projects of the same program. Something repeatable is really the best way to measure temperatures.

Another possibility is the difference of the heatsink mount. Perhaps something went slightly wrong with the E2140 mount or got loose over time.

The last thing that I can think of that could be making the E2140 to appear to be using more power is the location of the DTSs. I haven't read up on these in a long time, and even then it was only with the Nehalem architecture and things may have changed. I'm too lazy to reread the Intel docs so I'm probably using a term wrong here. There are a few Digital Temperature Sensors placed around the core at different locations. Ideally these would be right smack in the middle of the core as to know the exact temperature, but since there are more important things like transistors in the way they get placed in non-preferable spots. They are not symmetrical. The PWM controller takes either the hottest sensor or an average (forgot) and reports the value as the core temp over PECI, so if your E2140 has its sensors placed in a particularly hot spot compared to the Q9300 it would definitely create this effect.

Then again, the E2140 may be very well using more power than the 45nm Q9300. There is quite a bit of voltage going through it on an inherently more power hungry process.

Its a mix of either one, some, or all of the above things that could be happening.

 

[VirtualLarry]

Maybe and possibly. There are a few things I would consider for the change. The first is that the load does not appear to be with the same program. There really is no realistic application that produces 100% load on a CPU. Even the beastly Linpack has some stalls here and there. Distributed computing may not be the best to compare temperatures with as I have known drastic changes of temperature with different projects of the same program. Something repeatable is really the best way to measure temperatures.

Even with CoreTemp, that keeps a running max temp? I wasn't taking point-in-time temps, these are the max recorded temps.

Another possibility is the difference of the heatsink mount. Perhaps something went slightly wrong with the E2140 mount or got loose over time.

I doubt that the heatsink was loose, it didn't appear to be when I took it off to mount the Q9300.

The last thing that I can think of that could be making the E2140 to appear to be using more power is the location of the DTSs. I haven't read up on these in a long time, and even then it was only with the Nehalem architecture and things may have changed. I'm too lazy to reread the Intel docs so I'm probably using a term wrong here. There are a few Digital Temperature Sensors placed around the core at different locations. Ideally these would be right smack in the middle of the core as to know the exact temperature, but since there are more important things like transistors in the way they get placed in non-preferable spots. They are not symmetrical. The PWM controller takes either the hottest sensor or an average (forgot) and reports the value as the core temp over PECI, so if your E2140 has its sensors placed in a particularly hot spot compared to the Q9300 it would definitely create this effect.

Then again, the E2140 may be very well using more power than the 45nm Q9300. There is quite a bit of voltage going through it on an inherently more power hungry process.

Its a mix of either one, some, or all of the above things that could be happening.

Interesting. I have no idea about the DTS location issue.

 

[veri745]

Probably, but not necessarily.

There could be differences in thermal contact of the HSF to the CPU

or

There could be a more efficient thermal profile on the CPU die that allows better heat transfer to the IHS

Also, average temp is probably a better indicator of power usage than max temp.

 

[Accord99]

What's even more interesting, is that the E2140 is a 65W TDP, while the Q9300 is a 95W TDP.

It is possible that the overclocked Q9300 simply uses less power than the overclocked E2140. Measurements in the past have shown the Q9300, and 45nm Core 2 CPUs in general, to have full load power measurements well below their TDPs.

For example, Anandtech has the Q9300 using less power than the E6750:

http://www.anandtech.com/show/2489/13

while BeHardware/Hardware.fr measured Prime95 power consumption of 45W for the Q9300.

http://www.behardware.com/articles/717-2/intel-core-2-q9300-e7200.html

 

[jihe]

If you replace one CPU with another, and use the same cooling solution, and the temp as measured by CoreTemp goes down, does that necessarily mean that the overall power draw went down as well? I believe so, personally, but I wanted to get comments on my theory.

The interesting thing is, my E2140 @ 2.8, 1.425v BIOS, got up to 86C in CoreTemp on both cores during the summer, running SeventeenorBust.

I replaced it just recently with a Q9300 @ 3.0, 1.225v BIOS (stock voltage), and under full load (F@H SMP+GPU), it gets up to 71C max so far.

So from the temps alone, can I conclude that my slightly overclocked C2Q 45nm draws LESS power than my heavily overclocked and overvolted C2D 65nm chip?

What's even more interesting, is that the E2140 is a 65W TDP, while the Q9300 is a 95W TDP.

http://ark.intel.com/Product.aspx?id=29738
http://ark.intel.com/Product.aspx?id=33922

Just stick in a kill-a-watt and find out for yourself.

 

[Brunnis]

Another thing to keep in mind is that die size will affect temps as well. Given the same power consumption, a small die will run hotter with the same cooling compared to a larger one. I don't know the die sizes off the top of my head, so can't really comment on how it affects this particular case.