320 power phases should be enough for anybody

[Mark R]

Just wondering if this kind of logic still applies for the Haswell on-module VRM?

This VRM is interesting because of the use of on-die transformers/inductors, and ultra-high phase count, even though it's power efficiency is poor compared to traditional regulator design.

Any views on whether this is good or bad for overclocking?

In terms of capability, it looks pretty over-engineered. The regulator will ship with 16 blocks of 20 phases each, with 3 blocks alone sufficient to run a quad core sandy-bridge xeon CPU stable in Linpack.

---
Meant to post this in the CPU forum. Dear mods, please move.

Done
-ViRGE

 

[Ayah]

Depends on the current carrying capacity. They could be 0.5A phases (hopefully not likely).

 

[ViRGE]

If power delivery is as stable as they're saying (ripple <2mA) then I can't see how this wouldn't be good for overclocking. Though I suspect Haswell the CPU will give out long before its VRMs.

 

[Haserath]

Without anything other than the number of phases, we really can't conclude anything.

It's like saying the 5870 is faster than the 480 simply because it had 1600 shaders vs 448.

 

[IntelUser2000]

That Chinese site slide shows an experimental iVR, not the one they'll be using it in Haswell.

 

[Phynaz]

He's talking about Haswell's VRM

http://www.expreview.com/25414.html

 

[Haserath]

This slide points out that each regulator is thermally constrained to a lower amperage, but we don't know what.

With 20 power cells powering a 84W chip, then it must be constrained to single digit amps, or it might be yields and other ICs taking up separate cells.

 

[IntelUser2000]

He's talking about Haswell's VRM

http://www.expreview.com/25414.html

Yes, so am I. That one is an experimental chip.

 

[Enigmoid]

320 power phases should be enough for anybody

Still rocking your 640 KB?