Page 1 of 1

Enlighten me regarding Haswell's power delivery

Posted: Mon May 13, 2013 1:17 pm
by Geo2160
Well, this is probably a noobish question, since I pride myself with the fact that I've been in contact with computers and overclocking for more than 10 years, but I can not figure this out.

So, we pretty much know that Haswell chips will have on-die integrated power delivery systems, yet all the new high-end Z87 boards still have huge, sometimes actively cooled radiators covering their power phases. Even the number of power phases varies. Weren't those supposed to be inside the chip itself?

Can the board manufacturers implement extra power phases in a useful way and do they work in parallel with FIVR? Or are those just for show?

Re: Enlighten me regarding Haswell's power delivery

Posted: Mon May 13, 2013 1:22 pm
by Captain Ned
RAM power details are still up to the mobo.

Re: Enlighten me regarding Haswell's power delivery

Posted: Mon May 13, 2013 1:25 pm
by Geo2160
I don't think RAM needs 10-12 power phases by itself, no matter how expensive a board is and I believe they're usually positioned near the DIMM slots themselves like here on the upper rights portion of the board.

http://cdn4.wccftech.com/wp-content/uploads/2013/05/asrock_z87_extreme4_mainreal.jpg

Re: Enlighten me regarding Haswell's power delivery

Posted: Mon May 13, 2013 1:46 pm
by just brew it!
The on-chip regulators are probably designed to work with an input voltage less than +12V. So (assuming that this is the case) for a desktop board you would still need something on the motherboard to cut the +12V from the ATX12V connector down to whatever the on-chip VRM expects.

Re: Enlighten me regarding Haswell's power delivery

Posted: Mon May 13, 2013 2:03 pm
by Geo2160
This still doesn't explain why high-end boards have a lot of extra power phases.

Re: Enlighten me regarding Haswell's power delivery

Posted: Mon May 13, 2013 2:16 pm
by just brew it!
All else being equal, more phases still allow the regulators to run cooler and deliver cleaner power. Cooler operation potentially improves reliability; cleaner power probably doesn't matter (as much) with the on-chip regulators in the loop, but it is still something they can add to the feature checklist in their marketing materials.

Re: Enlighten me regarding Haswell's power delivery

Posted: Mon May 13, 2013 6:18 pm
by DeadOfKnight
Clean power and efficiency at all levels will make it a better board for extreme overclocking, and that's how they get their endorsements from people who really stress test them. Believe it or not some people will buy a board based on extreme overclocking results even when they only plan on running on air. I have no idea why that is, but I think it's safe to assume it is a testament to their reliability. If you wanna use the same CPU & motherboard for 10 years or more you probably can.

Re: Enlighten me regarding Haswell's power delivery

Posted: Thu May 16, 2013 11:13 am
by DPete27
I agree with why VRM is still needed on Haswell boards and why manufacturers are still using phases as a marketing hype to differentiate their products from their competitors. What I don't understand and what I'd like to see in reviews is how many phases are really required. With the on-chip regulation, I have to believe fewer mobo phases are required (just need get the power to the CPU). Sure, heat will always be a factor (how much?), but now it doesn't matter how "smooth" the power delivery from the mobo is because the on-chip VRM can handle that better than any previous on-board solution could. Also, what about switching frequency differences between on-chip and on-board VRM?

Re: Enlighten me regarding Haswell's power delivery

Posted: Sat May 18, 2013 4:04 pm
by auxy
DPete27 wrote:
I agree with why VRM is still needed on Haswell boards and why manufacturers are still using phases as a marketing hype to differentiate their products from their competitors. What I don't understand and what I'd like to see in reviews is how many phases are really required. With the on-chip regulation, I have to believe fewer mobo phases are required (just need get the power to the CPU). Sure, heat will always be a factor (how much?), but now it doesn't matter how "smooth" the power delivery from the mobo is because the on-chip VRM can handle that better than any previous on-board solution could. Also, what about switching frequency differences between on-chip and on-board VRM?

You mean you don't think 35 phases are necessary? ヽ(゚Д゚)ノ

</Gigabyte>

Re: Enlighten me regarding Haswell's power delivery

Posted: Sat May 18, 2013 5:13 pm
by Wirko
Google "Haswell FIVR 2.4V". Apparently Haswell takes 2.4V and derives the voltages it needs from it. However, everyone who claims that refers to this document, which is old and not directly related to Haswell, so the actual voltage might be different.

http://www.psma.com/sites/default/files/uploads/tech-forums-nanotechnology/resources/400a-fully-integrated-silicon-voltage-regulator.pdf

Another thing is interesting here. Higher voltage means lower current, which in turn means fewer CPU pins for power. Socket 1155 has an incredible number of power pins, maybe one half of all pins (see here, all pins starting with "V"). Socket 1150 may well cut this number by half, leaving a lot more pins for the buses and other things.

Re: Enlighten me regarding Haswell's power delivery

Posted: Sun May 19, 2013 12:39 pm
by just brew it!
Wirko wrote:
Another thing is interesting here. Higher voltage means lower current, which in turn means fewer CPU pins for power. Socket 1155 has an incredible number of power pins, maybe one half of all pins (see here, all pins starting with "V"). Socket 1150 may well cut this number by half, leaving a lot more pins for the buses and other things.

This is probably part of the motivation for going down this path. With lower and lower voltage cores, the number of power (and ground!) pins required to mitigate resistive losses has become prohibitive. Cutting the number of pins reduces package size and/or allows more pins for other stuff (like additional memory channels or bringing more peripherals on-die).

I've actually wondered for a few years now why CPUs haven't switched to using larger gauge pins for power delivery, to reduce the number of pins required. But moving the last stage of VRM on-die is a more elegant solution.

Re: Enlighten me regarding Haswell's power delivery

Posted: Sun May 19, 2013 1:17 pm
by UberGerbil
FWIW, there was a discussion of this in the forums over at RWT.

Re: Enlighten me regarding Haswell's power delivery

Posted: Tue May 21, 2013 10:50 am
by Geo2160
Well, I managed to find this, which answered most of my questions:

http://sinhardware.com/index.php/blog/314-haswellfivr

Re: Enlighten me regarding Haswell's power delivery

Posted: Sun May 26, 2013 6:46 am
by DeadOfKnight
Geo2160 wrote:
Well, I managed to find this, which answered most of my questions:

http://sinhardware.com/index.php/blog/314-haswellfivr

nice find!

Re: Enlighten me regarding Haswell's power delivery

Posted: Wed Jun 05, 2013 11:22 am
by DPete27
Just looking through the Z87 mobo lineup on newegg today. Looks like the vast majority of boards are using noticeably less power phases this round. Most of the "mainstream priced" boards are using ~4 phases.

Re: Enlighten me regarding Haswell's power delivery

Posted: Wed Jun 05, 2013 6:41 pm
by Geo2160
I noticed that too. I guess it's both a good and a bad thing. Good, because it means we don't need them as much for decent overclocks and bad because the OEM dropped their BoM costs w/o dropping the final prices.