4.19.2007

Scientia's Flawed Analysis

Since most of you seem to have read Scientia's recent blog "Core 2 Duo - the embarrassing secrets" let me then respond to them here.

Scientia said: "The two things that Intel would rather you not know about Core 2 Duo are that it has been (1) tweaked for benchmarks rather than for real code, and that at 2.93 Ghz it is (2)exceeding its thermal limits on the 65nm process."

(1)”Tweaked for benchmark" - Is he suggesting that the Intel engineers at the early stage of designing a new architecture which was about 5 years ago imagined what the benchmark tests will be like today and made sure that "high benchmark scores" was on the top of the design priority list? I don't think so.

He makes a case as to why the benchmarks do not represent the ‘real code’ of the real world application. Of course it isn’t. Not because you don’t wish it to be unrepresentative of real world apps but because there is no practical way to quantify what a ‘real world application’ is. Just like the American constitution for example, it's not perfect but its the best of what we have.

By definition a benchmark is a standard which you can use to measure one against another. A 3D mark score doesn’t mean anything unless you use that score to compare another score. That is what a benchmark is and for the sake of practicality you can only sample a few real world scenarios to make comparisons. The bottomline is that Intel never defined what the benchmarks are and both AMD and Intel are measured with the same stick.

Complaining about how the benchmarks doesn’t represent the real world is analogous to saying that the 100 meter dash Olympic race doesn’t represent how fast the runner is in real world situations like for example running across the Siberian Thundra. Of course it isn’t. The whole point of benchmarking is not to show performance against the real world but to show how one is better than the other.


(2) “exceeding thermal limits” – Scientia made his case by using Intel’s temperature guides which suggests that TAT should be at 55°C and the extreme editions are running above the limits.

Absolutely incorrect. The thermal dissipation rating of the CPU packaging is a function of the design of heat spreader (IHS), thermal interface material (TIM) and the hot zoning design of the chip. Obviously the temperature limits defined by intel were derived from the capability of the IHS and TIM to dissipate heat at 55°C. This limit has nothing to do with the silicon thermal limit. Nothing is stopping Intel from slapping a thicker IHS and a more expensive TIM just to increase heat transfer (along with clock speed) while maintaining that 55°C thermal requirement.

5 comments:

Anonymous said...

1. All review sites are evil boo!
2. AMD 6000+ is exceeding thermals of Intel's 2.93 quad.
3. Fanboy.

Anonymous said...

Here is the next thing to report on AMD REPORTS $611 MIllion Loss.

http://www.tgdaily.com/index.php?option=com_content&task=view&id=31704&Itemid=118

Let the excuses by scientia roll in. I wish he owns upto some of the BS he has been putting up on his blog and finally put some responsibility on AMD instead of blaming everyone but the company responsible for its own demise.

Anonymous said...

It's so sad. I actually think he believes the shit he sputters.

Anonymous said...

he's ignoring your thermal limit argument. it is what he gets when he talks about things he does not know about.

Scientia from AMDZone said...

roborat

"Nothing is stopping Intel from slapping a thicker IHS and a more expensive TIM just to increase heat transfer (along with clock speed) while maintaining that 55°C thermal requirement."

You do have a remarkable talent for the Strawman argument. You made that statement as though you are contradicting me when in reality you are simply agreeing with what I said. I stated that the temperature on that chip would be fine with premium cooling (as clearly shown in the tests). In fact, not only fine at stock speeds but overclocked as well. With a high end air cooler those chips run remarkablly cool even with large OC's.

Instead of making red herring remarks why don't you instead address what my real point was. If Intel is busting the thermal limits with a stock HSF then this chip is a special edition and not a normal chip. This also refutes the grandious claims that Intel had all kinds of thermal margin to increase clock. This argument is clearly false if you have to have premium cooling to even run at 3.0Ghz.

Now, why don't you admit the truth? Intel has hit the thermal ceiling with C2D on 65nm at less than 3.0Ghz. Intel could easily exceed this if they required premium cooling but this immediately takes the chip out of the normal category. Clearly, Intel is stalled at 3.0Ghz until they shift to 45nm. I'm guessing 45nm will bump that thermal ceiling up a bit and Intel will be off and running again.

I'm not picking on Intel here. Do you recall back in 2005 (I believe) when Sun sold an Opteron system that was overclocked 200Mhz but with premium cooling? Didn't Intel just do the same thing with a 3.2Ghz system? I think when you have to go to premium cooling that is a clear sign of a thermal limit. But, again, I'm sure Intel will slip by that quite easily with 45nm.