Ah... the details.

Know your hardware if you want to succeed at overclocking. No, I mean it, really know your hardware.

The first PC I ever built completely from scratch used an Asus P5W DH Deluxe Wi-Fi, a motherboard that won a Custom PC Approved award back in Issue 39’s (December 2006) labs test. I was keen to start learning how to overclock and wanted to get one that would allow me to satisfy my growing urge for tinkering with my CPU.

However, at the time I ordered all my kit, I didn’t have my copy of Custom PC to hand and couldn’t quite remember the suffix. I knew it was an Asus board with ‘P5’ in the title, so punched the digits into the eTailer website and searched away.

Two boards came up; the P5W DH Deluxe and the P5B Deluxe. Which one was it? I wanted to get it right, so investigated the specs a little. The P5W was based on Intel’s 975X chipset, while the P5B was based on Intel’s P965 chipset. Surely then the P5W DH with its higher end chipset was the one I wanted?

Click.

Wrong. It was the P5B that was the zomgwtf overclocker. Sure, the P5W DH was okay and it got me started and lasted for a few years. But it got me thinking - why wasn’t the board with high-end chipset the better one? Despite the fact the 975X even had an ‘X’ for eXtreme, the Asus Commando and P5B with their P965 chipsets were the boards holding world records.

...

You’d think that when motherboard manufacturers released their uber high-end motherboards that it would be those that were the overclocking kings. You'd think the £260 high-end model would overclock further than the £150 budget model, but as with the two examples listed above, that just often isn’t the case.

Having talked about this to various members of the team, we're still undecided as to why this is. Is there a degree of luck involved at the design stage, arising from choosing all the myriad components that make a motherboard? Is it because there are different design teams at most motherboard companies, and some are more accomplished than others? Or is there just a touch of magic involved when it comes to separating the great from the good?

This is a good excerpt from a blog post by Mark Mackay at bit-tech.net and the answer to this particular P5B or P5W is actually quite simple: the 975X was released months earlier than the P965 and required more voltage to achieve similar FSB overclocks. This was due to the 975X being manufactured at 130nm while the new P965 was released on the newer 90nm process.

The only reason for the 975X to be marketed as better was because it was the one that officially supported CrossFire in 8x/8x configurations, while the P965 needed some BIOS tweaking and only had x16/x4, therefore hurting performance in CrossFire, most of the time. Throw overclocking in the mix and the P965 overulled the 975X.
Had Intel done right, they would've toss the 975X out, rework the P965 with CrossFire support and call them 985X, or something like that, but they didn't.
For instance, you can buy a C0 revision of the Core i7 965 and you'll probably get a lower overclock than if you manage to grab a D0 Core i7 920 - you won't have unlocked multipliers but the newer revision can do more.
Same goes for the HD4830 vs HD4770, the HD4770 is faster on average, consumes less power and overclocks a lot more - that's just the way we play the overclocking game: we must really know the details.

As for the mild fiasco that the 975X was, the worse came later: most P965 boards supported 45nm Core 2 Quads while the 975X supports some, the P5W only officially supports them underclocked at 1066MHz FSB. Sometimes, you just don't get what you pay for.