PDA

View Full Version : Intel 45nm CPUs and X38 chipset - what (and what not) to expect


freecableguy
September 24th, 2007, 12:36 AM
http://www.thetechrepository.com/attachment.php?attachmentid=346&stc=1&d=1168997789


IDF 2007 gave the world a small taste of what the expect from the upcoming 45nm line of CPUs from Intel as well as some juicy screenshots of some rather amazing systems based on the newly-arriving X38 chipset. Although these results are nice to have they really don't provide any real information in the way of overclocking potential. Sure, the systems were overclocked to some degree, but without some rather specific details it's rather hard to tell exactly what the future has in store for the Intel enthusiast. Since direct comparisions are usually the best way to explain a performance increase from a previous product generation/revision, we'll do exactly that. At that same time, we hope to dispel a few of the rumors floating around as well as support a few that we think you should be paying attention to. In the end you should be able to walk away with a pretty solid understanding of what Penryn/Woldfale/Yorkfield/Harpertown have to offer (and what they don't). If nothing else, this should be a good preview of things to come.


Chipset Performance

Now, there are some out there that think that X38 is going to be a "monster chipset." While I tend to agree with this camp at the same time I think there are a few stipulations which should be made available before you decide to accept this claim at face value. The first argument I want to make is this - heat (or power if you will). The X38 chipset is a lot more hungry for juice than any chipset we've see from Intel to date. Those that believe that the X38 chipset is a derivative of the G/P35 chipset are surely wrong - in fact separate design teams were charged with exclusive development of the P35 and X38 chipsets. As such they actually share very little in common save general chipset knowledge passed down from the gods. ;)

The addition of the IHS to the X38 chipset is merely for the mortal purpose of safely allowing stock mounted cooling systems to provide adequate clamping pressure . (Those that think the addition was at the bequest of overclockers looking for a more stable base to rest their DI pot on will surely be disappointed to hear otherwise.) While it is true that the addition of PCI-Express 2.0 lane control logic (with higher transfer speeds) did contibute to the overall increase in thermal output the real reason is the use memory switching logic that has come to look more and more like a high-level data buffer. It should also come at no surprise that the silicon used in the new chipset has been specifically designed for hot, high-speed operations.

In the end, although the X38 will be faster clock-for-clock than G/P35, the difference will not be immediately noticable. Preliminary benchmarks show the performance increase to be only on the order of 5-10% for stock speeds. Thankfully we live in a world where stock speed benchmarks don't account for anything! The real power of the X38 chipset will be seen during overclocking, specifically the overclocking of Extreme CPUs. What do I mean by that? Great question - Intel has already made it public knowledge that they believe the future of computing to be core-centric, quad-core to be exact. And as such all future "Extreme" CPUs will be quad-core CPUs begining with the release of the Yorkfield QX9650 on or about 12 November, 2007. Therefore, whenever I say "Extreme CPU" you would be best to read this as "quad-core CPU." :)

My explorations with P35 have been a difficult journey. We've been introducted to a lot of new concepts and BIOS settings. GTL reference voltage adjustments (http://www.thetechrepository.com/showthread.php?t=87), introducted to the overclocking community by us back in January of this year, should be a part of your overclocking forté when dealing with a quad core CPU - if not you're behind the power curve. Individual phase adjustments for memory channels with respect to the "performance level" (also known as Read Delay), CPU PLL voltages, 1N vs. 2N, Northbridge/MCH straps...there's a lot to know these days. Getting into overclocking is simultaneously both easier and harder than ever. There is one thing that I have learned regarding the P35 chipset and quad-core overclocking though and that's this - assuming you have applied all know tuning knowledge correctly thus far, the FSB limitation we are now seeing has become a matter of silicon limitations and nothing more. Enter the X38 chipset - who would have thought that an "Exteme" chipset (X38) would be best paired with an "Extreme" (quad-core) CPU!? (Did a little lightbulb just go off in your head?)

In terms of performance tuning the DFI P35-T2R has brought us just about as far as we can expect to go. My best result on CPU watercooling and stock cooling on all other motherboard components (NB, SB, MOSFETS, memory, etc.):


http://www.thetechrepository.com/attachment.php?attachmentid=613&stc=1&d=1190607551


I have to say, getting this system to 475MHz FSB stable on the 266 strap is about a good as it gets. Using an 8x multiplier, I've set my CPU at an impressive 3.8GHz while utilizing a 4:5 memory divider for about DDR-1190 memory speeds. Stability at these speeds is a sight to behold, especially considering the near-silent watercooling in use and ~50C, quad-P95, full-load core temps. Will FSB go higher? The simple answer...no, not really. Not without loosening some rather major settings which provide nearly all the performance gain found by overclocking this high. That's what the X38 chipset is for - look for some big gains here. Soon, 500-550MHz FSB with a quad core won't be so difficult to achieve. :D


Processor Performance

You want the quick and dirty when it comes to overclocking 45nm? Really? OK, why the hell not. Here it is. Nominal VID will be about 1.150 - 1.250v. Similar overclocks from 65nm variants will be on the order of 0.1v less than you've experienced thus far. For example, if your QX6850 requires 1.45v for 3.6GHz stable operation then expect that the QX9650 will reach these speeds at around 1.35v. (Note: I am in no way suggesting that the QX9650 will require 1.35v for 3.6GHz operation - this is only an example of math at work.) Although most people will find 3.6GHz to be a rather good sweet spot for continued stable operation, it won't be uncommon to see overclocks in the range of 3.8 to 4.0Ghz (on air or water) when using a next-generation quad-core. Some more experienced users will find that stable speeds as high as about 4.2GHz are possible with off the shelf water cooling and high-end aircooling.

Combined with the right chipset, and a motherboard capable of providing good, clean power required for such high-speed operations, the 45nm CPU from Intel will be a force to be reckoned with. Even though I am a big fan of AMD (imagine that!), I don't expect them to catch-up with Intel this round. For those that are asking the question, 'should I wait or should I upgrade now?'; my answer is simple - update now and then later. :rolleyes:


-FCG