By now we have all heard tell of Intel's new 45nm Penryn which will supposedly have a significantly lower power consumption than the current 65nm Conroes. But, exactly how much less power will these Penryns consume? Due to the Yorkfields being natively quad core, I will not be able to calculate their power consumption compared to our current Kentsfields. I can, however, calculate the power consumption of a Wolfdale at 4000MHz as it is a dual core somewhat simmilar to the current Conroes.

The reason I have chosen 4000MHz is because the largest circulating rumor surrounding the Wolfdales is that there will be a 4000MHz version. My prediction is that a Wolfdale operating at 4000MHz will have a 65W TDP. Unfortunately, none of you n00bs will ever take anyone seriously without proof, so here it is.

According to the 45nm High-K + Metal Gate parameters (Vcore, IdSat, etc) it provides 30% lower transistor power consumption at 20+% higher frequency compared to the current Conroes. Now, lets consider a 65W TDP E6700 as our base in the calculation. An E6700 operates a 2670MHz.

2670MHz * 1.2 = 3200MHz
65W * 0.7 = 45.5W

If for the 45nm core operating at 4000MHz, the Vcore remains same as for 3200MHz, it will waste 56.875W.

(4000MHz / 3200MHz) * 45.5W = 1.25
1.25 * 45.5W = 56.875W
65W / 56.875W = 1.143

This means that to 65W, there is +14.3% headroom. Or you can increase Vcore to +6.9% and still be in 65W TDP. The square root of 1.143 is 1.069.

So, seeing as how all E6700s can overclock at least 25% and remain stable at stock Vcore, and a Wolfdale can can overclock 20% higher than an E6700 at only 7% higher Vcore, we are looking at a maximum of 65W TDP on a stock 4000MHz Wolfdale. But, since Intel likes round numbers, they will rate it at 65W TDP.

Your analysis seems to provide a best-case scenario. I take it that you got the 30% power reduction and 20% increased switching speed from Intel's site.

I might have missed it, but I don't think we're looking at a 30% reduction in CPU power consumption while at the same time seeing a 20% increase in CPU frequency. At best, i would take Intel's claims to mean that you are likely to see either one or the other. Also, I don't know how well individual transistor switching parameters translate into the full integrated circuit performance.

I think 45nm is going to be a significant improvement, and certainly it should be a greater advance than the previous generations. I remember 65nm was hailed as a significant step from 90nm.

Yorkfield is a "glued" or whatever you want to call it quad like Kentsfield. Nehalem will be Intel's first "native". For whatever "native" versus "glued" is worth. See the Intel slide below, which shows Harpertown, the Xeon version of Yorkfield, is still clearly two dies.



I could convince myself to believe the 30% power reduction with 20% speed increase. At a course level, the loss in the CPU is largely related to leakage plus charging and discharging the capacitance of the FETs, and the capacitance is proportional to area, 45nm transistors have about 1/2 the area of 65nm transistors, so a net 50% improvement could be rationalized. The high-k gate is key to make sure that this simplification holds, otherwise leakage could dominate the total loss.

You need to be careful with this type of thinking though, as expecting consistent scaling results was what led Intel to think Netburst was a good idea (remember the 10GHz claims?), and the AMD 65nm die shrinks have also failed to produce the scaling benefits expected. Remember that Intel's last two die shrinks on the P4 were less than stellar - going from the pretty good for its day 130nm Northwood to the abysmal 90nm Prescott, and then a partial recovery going from the Prescott to the less horrible but still mediocre Cedar Mill core. The key there, IMO, was maturing the 65nm process so that when they released Conroe they already had experience at the node. Hopefully they won't have the same issue going from Conroe to Penryn as with the P4s, but you never know. So we'll have to see what Intel delivers, but Penryn should be nice if the scaling holds up.

Yes, I do remember the sad day when Intel shrank its P4. 10GHz NetBurst is still very probable though. You can see a plethera of Pentium Ds overclocked well past 8000MHz and I am sure that the limiting factors are cooling and voltage. With a decent Vmod and maybe a 3rd stage cascade I am sure 10GHz is possible. Unfortunately, it seems as though the Core architecture tops out at around or slightly over 6GHz. The new waste gate designs look to be the most innovative step Intel has made in a long time.


P.S. This is the most intellectual conversation I have had on Tom's Hardware in some time....