What exactly is having voltage regulation on die supposed to buy as far as CPU performance is concerned? Are the small core voltages getting so small and current draws so large that the the pin resistance is getting to high to ensure a consistent on die voltage from offboard regulators? Why not just add more pins? It seems like quite the thermal trade off as onboard regulators will be limited to the inefficient linear type if my understanding serves me.
Pinout. Rather than having a bunch of 1.something volt connections, they can use a couple of higher voltage pins.
Problem is, vregs generate a lot of heat. It's also tricky trying to filter the output, on the die itself.
Lower power use and lower heat? I'd pay for that. I hope AMD implements it too. And I hope they both do it <i>soon</i>. I like quiet PCs.
The <b><font color=red>Devil</font color=red></b> is in my <b><font color=red>'98 Mercury Sable</font color=red></b>!
<b>@ 201K miles!</b>
Lower heat? That depends on whether you are running an app where it makes sense to cut the CPU power from time to time. In new games etc where CPU is busy 100% of the time everything will end up getting hotter due to the regulator inefficiencies which are somewhere around 90-95% for even the highest quality switchers. I believe however that they will be forced to use a linear type regulator since there is no way to implement large inductances or capacitances on die reasonably which bars them from using a high efficiency switcher. Using a linear can only result in the inefficiency and heat generation being even worse.
If they were to do this for multiple CPU sections independently of one another I can see how it starts to make more sense, ie cut power to the parts of the CPU that aren't going to be needed during your game. Maybe that is what this is and I just didn't get the details. That would be kinda cool.
there is no way to implement large inductances or capacitances on die reasonably which bars them from using a high efficiency switcher
That was my thinking, but having seen it in print, I'm not so sure that the pi filter is even needed. With the vreg on chip, it could be responsive enough to supply the exact voltage required. If you do something that makes use of 100% of cpu all the time, you would see a 5 to 10% increase in TDP, but for "home users" the voltage control would have it's pluses.. Intel could at least claim that the ODVC makes chips run cooler 19 times out of 20.
I agree with your point about how many users would see benefits most of the time...surfin the web, emailing grandma etc.
Sorry, I probably wasn't clear about what I meant. The L's and C's I was referring to wouldn't necessarily be used in filters per se. They are the backbone of the different switcher regulator circuit architectures. For example: buck type switchers (very common in PC power supplies)use switched inductors as an energy storage component. Of course there will be some filtering required on the output of these types of circuits to keep ripple to acceptable limits, but that is more of a downstream effect.
Linear regulators don't require these external passive components and may not even require output filtering as you have pointed out. The tradeoff is that linears are typically much less efficient depending on the circumstance. Most likely these linear regulators is what designers will be forced to use since creating sizable passives on die ie the types needed for switchers isn't practical. Unfortunately lack of efficiency means heat.
It should be interesting to see how this pulled off in practice.