Yea perhaps a bit harsh but jeez.
We've been through this a few times before for sure. If really interested in some practical testing for those with more sense than to try and do it with error bars perhaps some logic could be applied.
1) Everyone will need the same CPU and IMO it should be a ceramic based AMD (TBird or MP). Take that CPU and thermal epoxy a thermocouple or thermistor under the core's center. Then cut away a bit of the socket to run the wire out. Then you can use AMD's tech doc to extrapolate the CPU die temp from the reading of CPU under. You will want to do a voltage modification on your board so you can tune the voltage under load to the desired amount. You'll want to do a mod to the socket so you can record Vcore as well rather than relying on the motherboard's circuitry. Use a CPU loading program that is insensitive to fsb and ram timings (I suggest CPU Burn). That's a decent start towards cross-person comparability.
Is any of this really "bench testing"? No because the power of the CPU cannot be quantified.
|