Haswell will move the VRM circuitry on the CPU package, so the MB will supply the cpu directly with +12V, instead of stepping down the current down to ~1V as is done with the present-day cpus. This means that it would be essentially impossible to support Haswell and Ivy Bridge on the same MB.
This change is also positive, meaning that there is a good reason to break backwards compatibility, unlike the case was with 1156/1155.
Interestingly, after the VRM is on package, there should be absolutely no cpu performance- or overclockability-affecting parts left on the MB. So, how will the MB makers sell their expensive editions in the next gen?
How is that supposed to work? VRMs contain big, bulky components that usually require a good bit of cooling.
How is that supposed to work? VRMs contain big, bulky components that usually require a good bit of cooling.
This. How is Intel going to pull that off? Somehow, I have my doubts.
How is that supposed to work? VRMs contain big, bulky components that usually require a good bit of cooling.
If I get Ivy Bridge next month. Can I get a motherboard with it that will also run with Haswell in 2013? So that if I choose to upgrade some time in the next year or two, I don't have to buy another motherboard.
short answer: no
long answer: noooooooooooooooooooooooooooooooooooo
Have you ever looked at a recent laptop motherboard? There's nothing bulky about the VRMs and they typically don't even have a heatsink. Moving them on-chip might also be a necessity to achieve the ultra-low standby power consumption.How is that supposed to work? VRMs contain big, bulky components that usually require a good bit of cooling.
That may just depend on the motherboard manufacturer. The chip may have the integrated VRM, but the motherboard manufacturer may decide it needs the bigger VRMs to handle power better.
How is that supposed to work? VRMs contain big, bulky components that usually require a good bit of cooling.
Also the slide seems to indicate on-die VRM rather than being on package. Interesting how it will turn out.
I'd fully expect that none of the 1150 parts will be capable of taking in anything but 12V. Part of the advantages of moving the circuitry on-die is to get rid of huge lines needed to shift the massive currents needed for 70W+ at 1V. Making it possible for the MB to supply low voltages would entirely eliminate this advantage.That may just depend on the motherboard manufacturer. The chip may have the integrated VRM, but the motherboard manufacturer may decide it needs the bigger VRMs to handle power better. Of course in most if not all laptops you'll only have them on the CPU.
I'd be quite surprised if there was an option to bypass the integrated VRM. Such an option would require both design overhead on the integrated VRM and an adequate number of pins on the socket for off-chip VRM power delivery. One marked advantage of having an on-chip VRM is that the number of power supply pins can be reduced as the input voltage rises.
It's not just about the pins. Moving power laterally on the chip and package gets harder each time the voltage goes down. 80A is a lot of current for something as thin as a cpu.
How do they make an on-die version handle it when its running at 6-7GHz @ 1.5-1.6V using 300W or so?