Intel Core i7-3960X (Sandy Bridge-E) And X79 Platform Preview

Interesting to see a major site in hardware-land post actual numbers on SB-E. The conspiracy theorist in me says Intel allowed performance numbers to leak to anticipate AMD's Bulldozer launch.

We’re at least a month or two away from Sandy Bridge-E’s launch, and a lot is expected to happen in that time. There’s AMD’s anticipated Bulldozer architecture, to start.

Although we don’t yet have a die shot or block diagram of Sandy Bridge-E, it’s pretty clearly an amalgam of Sandy Bridge’s architecture and scalable cache structure with the same core count that previously gave Gulftown an advantage in well-threaded applications.

Of course, in the segment it was designed to address, Intel moves PCI Express control from X58 to the Sandy Bridge-E die itself, adding a fourth 64-bit memory channel able to run at higher data rates. The result is a simpler two-chip platform than X58 better able to service the server apps dependent on memory bandwidth. Decidedly, consumer apps see little, if any, benefit from the more complex memory controller.

Cumulatively, the impact of Sandy Bridge-E over Core i7-990X is felt in both single- and multi-threaded apps, topping out in the 30% range in a benchmark like Blender. If you count yourself amongst the workstation users justified in spending $1000 on a six-core processor due to the productivity gains it provides, Core i7-3960X looks to be a substantial upgrade as a result of its Sandy Bridge roots.

Also, by the time you read this, we’ll be on the way to IDF in San Francisco, where we’re scheduled to sit in on several briefings about Ivy Bridge, its 22 nm tri-gate transistors, improvements to the architecture’s media functionality, and Windows 8.

Though Sandy Bridge-E promises notable gains in the server world, it’s destined to be less influential on the desktop, if only because the number of folks willing to pay a steep premium for two additional cores and an otherwise-similar platform is small. Sandy Bridge spoiled us, so a high-end part just doesn't have the impact on enthusiasts that Bloomfield had back in 2008.

Ivy Bridge is sure to make a bigger splash, so stay tuned for more information from Intel as it flows out of IDF.


Belgium Massman says:

More importantly:

There is a silver lining, though. Sandy Bridge-E is expected to be overclockable in ways Sandy Bridge is not.

Two of the three SKUs purportedly planned for launch will be multiplier-unlocked, simplifying overclocking quite a bit. Both of those models should support ratios as high as 57x (just like the Sandy Bridge-based K-series SKUs were), easily taking care of the air-cooling community. The third, Core i7-3820, will get a number of accessible bins on top of the highest Turbo Boost multiplier, similar to Core i5-2500 and Core i7-2600.

But whereas P67 and Z68 employed internal clock generators, X79’s BCLK comes from a CK505 embedded clock fed through a buffer (a chip that follows Intel’s DB1200GS Differential Buffer Specification) responsible for “gearing” the frequency.

With Sandy Bridge, if you didn’t have an unlocked K-series SKU, or a partially unlocked Core i5/i7, then you were pretty much stuck. Changing the BCLK frequency directly affected other buses, quickly affecting stability. We’ve been able to push single-digit increases, but anything more than 9 or 10 MHz is asking for trouble. Of course, Intel's intention there wasn't nefarious. By integrating the clock, it cut power use and cost on the 25 and 14 MHz crystals. Inflexible scaling just turned out to be a side-effect.

Sandy Bridge-E should alleviate this somewhat by using the buffer chip to apply one of three different ratios to the BCLK. These will modify the PCI Express bus and DMI, creating a greater range of viable frequency settings. I emphasize the word should because I wasn’t able to get the mechanism working on our lab system. Increasing the BCLK and dropping our -3960X’s multiplier simply kept the platform from POSTing, no matter what combination of settings I used.

At least at first, the question to answer is going to be: who cares, anyway? When you consider the Core i7-2600K is completely multiplier-unlocked, who is going to want to fight a partially-unlocked quad-core -3820 on an expensive new X79-based motherboard? The extra 2 MB of L3 cache, two additional memory channels, and PCI Express-rich processor probably won’t inspire many upgrades. The real candidates for greatness are the six-core models, both of which Intel thankfully leaves unlocked. Specifically, I’m excited to see what the Core i7-3930K can do.

Belgium Massman says:

So now we know. The X79 chipset will still have the CK505 clock generator embedded and a gearing buffer will be used to 'shift' frequencies up and down. The gearing is pretty much acting like BCLK straps ... pretty similar to what we have been doing for years now (remember BSEL mods?). There are a couple of rumours stating that behind-closed-doors, Intel has admitted to making a mistake with Sandy Bridge by locking the BCLK to 100MHz and not allowing other straps to be used. Whether it's really a mistake or just a matter of keeping things as 'simple' as possible, we will never know of course. But we do know that any CK505-compliant clockgen is capable of delivering 100/133/166/200/266/333/400 MHz. With PCIe/Sata locked at 100MHz. Ivy Bridge is said to open up the BCLK 'straps' 100 and 133 MHz, which could point to the absence of logic to support other straps on Sandy Bridge. Maybe we can also lay the connection between the additional BCLK option and the [IMG]upgrade path problem[/IMG]. If we assume the rom to be flashed is not the BIOS, but rather the Intel Management Engine firmware (which would explain why the procedure is said to be impossible for end-user), it kind of makes sense. As we know, the ME is in charge of setting the frequencies; hence why you need to install the MEI driver before you can increase the BCLK in OS. If the current firmware of the ME does not support multiple bclk configurations (which is doesn't need at the moment), it will need an update before it can fully support Ivy Bridge's 133 MHz strap. I'm speculating a bit, though :p.

United States Hondacity says:

it seems like another bios implementation wars... oh yeah you forgot the foxcon QF in your list on the other thread.

United States Hondacity says:

oh thats the 1800$ cpu, nice :D

Belgium leeghoofd says:

Love the tight timings :) rams cost 1800 dollars too :p

Belgium Teemto says:

Who needs Sandy Bridge-E when you've got the FX :-)! (in dutch but with pictures).

Please log in or register to comment.

Leave a Reply: (BBCODE allowed: [B], [QUOTE], [I], [URL], [IMG],...)