Do You Really Need A 24-Phase VRM?
We had the opportunity to visit Intel’s Hawthorn Farm campus in Hillsboro, Oregon more than two years ago, and that trip yielded a cool picture-based piece walking us through the facility. The highlights of what we saw included radio frequency testing, acoustic measurements, shock and vibration validation, and serious work on voltage regulation circuits. If you missed that story, check out Under The Kimono: Inside Intel's Hidden R&D. It's really an interesting read.
More recently, Intel invited us back for another trip through its facility, this time giving us the opportunity to catch its engineers discussing the company’s technology on camera.
The timing isn’t exactly coincidental. As you’ve likely noticed, we do monthly roundups of the motherboards that sit at the center of your PC. Sometimes these include Intel’s own branded boards; more often, however, they don’t. And yet, Intel is eager to remind us those board designs that come out of Hawthorn Farm aren’t simply reference builds. According to Brian Forbes, the engineering director for Intel’s advanced engineering team, they’re completely different from the qualification platforms used to bring up a new processor or chipset.
And while those boards are already well-respected in business-oriented environments, Brian’s team claims to have the enthusiast market in its sights. But we didn’t hesitate to remind the Intel team that competing vendors are faster to incorporate add-on peripheral interfaces like USB 3.0 and SATA 6Gb/s. Plus, more traditional enthusiast-oriented platforms feature much more complex voltage regulator circuits, potentially enabling better overclocking results.
And that’s where we started our discussion with Mr. Forbes. Motherboard vendors make it a distinct point to tout the intricacy of their voltage regulation circuit, leading to something I call phase inflation. Boards with 12-, 18-, and 24- phase designs continually try to one-up each other. Meanwhile, Intel only uses six or eight phases. So what are the ramifications?
What Makes A VRM Good?
Brian explains that his team starts its analysis with a baseline, which is whatever the processor requires in its stock form (for a Gulftown-based chip, for example, that’s 130 watts), to determine the number of phases required just for that. Then, if the board is being groomed for the enthusiast market as some of Intel’s recent platforms have been, the team tries to figure out how far it can be pushed.
“Many times, based on the tuning of the circuitry and our component selection, we’ve been able to get upwards of 210 amps using six or eight phases,” Brian said. “Granted, that requires passive cooling with minimal airflow, which typically comes from your processor’s fan.” During that discussion, it was mentioned that, especially with the unlocked K-series and Extreme Edition parts, the team at Hawthorn Farm has to hit a point where the processor simply cannot run any faster and gets damaged—without affecting the motherboard. Despite what we’ve been programmed to believe, Intel claims that is possible to take an unlocked chip to its breaking point with six- or eight-phase power.
Then Brian got provocative. “When I look at these 12-, 14-, 18-, 24-phase boards, in many instances, it appears those designs are trying to compensate for some type of an engineering ‘gotcha’, where if I throw a little bit more at this, I should be able to get more power. And that might be true for a period. But when you start to get to extreme phases where you need those 24 phases operating in harmony…we haven’t seen it done. We’re using FLIR cameras to capture the thermal effects on each and every device. And what we’ve found 99% of the time is that the board is operating and we’re watching components degrading rapidly over a year or six months, where the thermal limits of those parts get exceeded.”
Now, that’s a significant challenge to the rest of the motherboard industry, which would have us believe that more phases are correspondingly better. Of course, without the measurement equipment Intel uses in its labs, it’s difficult to validate the power delivery of an eight-phase design compared to larger voltage regulation circuits. However, Brian reiterated over and over that his six- or eight-phase design delivers more power than anyone is able to exert on the motherboard, and it does so in a symmetrical way, whereas larger VRs tend to lack balance, causing certain switches to handle a majority of the load and heat up unevenly.
Naturally, we’re skeptics. We wanted some sort of example to show that these boards, with their less-complex VRs, can do what Intel says they can do. A member of Brian’s team pulled out his testing notes as proof. The DP67BG (a P67-based platform) was able to take a K-series chip topped with the XTS100H heat sink up to 5.1 GHz at 1.5 V stably. The same setup hit 5.4 GHz at 1.6 V and could get into Windows, but wasn’t able to run the same long-term Prime95 load without crashing. As a point of comparison, we usually don't go much past 1.35 V in our own air-cooled efforts. The fact that Intel is pushing so high is a good indicator of what its boards can take, even if you wouldn't want to operate your CPU up there for extended periods. Using more extreme measures (what I assume to be phase-change, given his -25 degree Celsius notes), the config saw 5.9 GHz at around 1.8 V before crashing. However it ran stably at 5.4 GHz.
Using the DX58SO2 board (which fixes the shortcomings of Intel’s first X58 Express-based board, including its four memory slots) and a Core i7, the same engineer jotted down stable speeds of up to 4.5 GHz at 1.425 V running Prime95 and unstable clocks of up to 5.1 GHz at 1.6 V—on air cooling. The P67 board had been pushed to memory speeds as high as 2133 MT/s, while the newer X58 platform saw 2400 MT/s using XMP profiles.
