[bobo5195]

Nice to have you back stew. Unfortunate that Procooling seems to be reverting to old ways. There is not even creative use of swearing anymore tsk tsk. I thought we were ending personal attacks. Everyone hear is coloured by experiences that is why we are hear we all know something different. Maybe we should pick a random newbie off the [H] forums to decide and argue? At least they won’t have bias.

First off I quite agree with the bit tech review intel cpus are heat load limited if intel was willing to produce TDP max 200w cpus I’m sure they could get to 5ghz quite easily (a real world tech article pointed out that 65nm chips have 500mhz chopped of the design case simply from heat), but users would hate them and most computers wouldn’t use them. It was an example case of what good cooling allows you to do.

This is no where near as complete as I wished but here goes. Sorry about my grammer and stuff but the red bull is wearing off.

My take on all of this is that the TTV is a very good setup, I do take mgfr data from the TTV as near perfect, jaydee. It is a factory made industrial design thermal testing rig after all. The results for all TTV setups must be identical if correct methods are used and I disagree with any talk of them being flawed as they are unknown. Intels only possible course of action is to use the best methods available so not knowing what the test does is less important. For intel there is too much riding on it for it not to be a god test. A few dodgy heatsinks get through it is not worth it for them as a company. The methodology in using it is perfect the results are repeatable and it is accurate. It is perfect for testing. If every website in the world used it for testing then all reviews should come up with the same numbers and there are smilly faces all round. It can also seems to be a true model of real life. Secondary loses through mobos are included TIM and heatspreader is included.

So why then if it such an uber well made test do I think that it is flawed? There are a few little things all related to the fact it is a verification not testing device as such it is a flawed platform for testing. In short it is not as accurate as other methods, less variability between results but less resolution 0.5 ± 0.1 degrees(TTV) compared to 0.1 ± 0.2c(DIY) say.
• Results only give as degrees. No reporting in Watts. Alright this may seem like a nitpick but it is an exceptionally important variable. It means that you can compare waterblock A to B but you can’t test them under different loads. You can’t tell performance on different chips as you can’t do a C/W conversion.
• I think that its accuracy is low for two reasons.
o No modelling of actual CPU. I don’t believe cpu die variability is included and a standard heater is used. This is a fine approach for validation and admittedly not many other testing rigs have this but it is still a flaw.
o Inclusion of IHS skews results. The use of something in the way is going to modify results. Accuracy is going to be lower as heat has to travel through the IHS and its TIM, this can be seen in the reduction of gap between the apogee and storm in tests carried out on real cpus. Like working out engine power measured at the wheel compared to crankshaft.
• We don’t know how the TTV compares to real case CPUs. I presume it’s a worse case value. Ie this is simulation of the worst CPU from a heat and IHS interface point of view that intel would release into the wild. As such it does not measure the case of real cpus well and we do not know what correction factor to use to convert up to the “average” cpu in the wild.
• I am worried about the thermocouple and groove effecting results. It is not a true model of a cpu and as such prone to error.

As such DIY made rigs are better because we know more. The best rig for CPU testing is a bare die precision rig. Heat in is accurate, no other effects going on and as such because we have followed KISS less potentially to go wrong.
• We can model other effects and as such vary them. Switch between bare die and IHS easily. New IHS comes out we can correct and compare to previous data. New testing rig results can be directly compared as they are both accurate heaters their results should be identical. New cpus, new mounting methods all correctable.
• Heat input can be varied and quantified. We can dial up the heat and test high thermal loads. This is very important as while non-dimensional analysis means we should be able to scale result this is often not the case and in any case there aren’t the people around to do the conversion formula accurately.
• Accuracy is greater as we know exact numbers and there is greater resolution of the testing device. The real world is getting in our way.

In short we are trading accuracy and ability to examine different cases for repeatability. For testing this is a good compromise as accuracy is the most important but for a verification rig that intel has produced in the TTV it is not an acceptable trade off as repeatability is key.

What we need to do is make sure that other variables are accounted for. Data on secondary losses should be quantified. TIM / IHS data be put back in and corrected for.