wielander :
Don't get me wrong; I think it would be awesome. I just don't think it would make sense from a marketing standpoint.
There are LGA1366 32nm Xeons because in multi-CPU server setups AMD can actually compete with the performance of hexacore i7s. That simply isn't the case with desktop processors. The need for efficiency also isn't as great.
Intel couldn't really counter AMD on 45 nm since LGA1366 can only realistically handle one die per socket. Intel is also realistically limited to four cores per socket on LGA1366 on the 45 nm process since their 45 nm Nehalem-EX six-core units are clocked lower than AMD's 12-core units for the same power envelope (105-watt six-core Xeon E7540 @ 2.00/2.26 GHz vs. 115-watt 12-core Opteron 6174 @ 2.20 GHz.) There's no way they are going to win that battle as the Opterons would be considerably faster and less expensive due to smaller dies (the Nehalem-EX's die is about 650 mm^2 compared to Magny-Cours dies being ~350 mm^2 each.) However, going to 32 nm allowed Intel to make a 95-watt Xeon X5670 @ 2.933/3.333 GHz with a die size under 300 mm^2 that
can rival the Magny-Cours.
Desktop is a different story as desktop workloads are not nearly as heavily threaded as server workloads are, with quite a few desktop workloads still being single-threaded. Few desktop programs benefit very much from more than three cores today, so desktop performance pretty much depends on how fast an individual core is. Power usage isn't nearly as much of an issue as you have fewer cores in desktop machines and can afford to have more complex cores and clock them high. Intel's Nehalem is somewhat faster core-for-core and clock-for-clock than AMD's K10. Intel can sit on 45 nm on the desktop until either programs on the desktop become more multithreaded and AMD's inexpensive six-cores start to notably outperform Intel's quads, AMD rolls out 32 nm and gooses clock speeds enough that Intel needs to move to 32 nm to keep up, or AMD's new Bulldozer core is faster core-for-core than Nehalem and Intel needs more clock speed and cache than they can get with 45 nm to beat it.
We haven't seen any samples. If these chips were coming before Sandy Bridge, you would think that we would have heard something.
Sure we have, the i7 970 and 980X. Gulftown is what's going to replace Nehalem on LGA1366, if Intel decides to replace Nehalem rather than just EOLing LGA1366 without releasing anything on 32 nm for it besides Xeon 5600s and a few expensive six-core i7s. You can bet the quad-core units on LGA1366 are going to be just like the quad-core Xeon 5600s- a six-core die with two cores shut off. I doubt Intel would go through the trouble of making a unique die mask for a native 32 nm quad-core for LGA1366 when they know Sandy Bridge is right around the corner and LGA1366 will be EOLed.
I would be surprised if Intel is short of things to produce at 32nm yet (based on the reports of shortages of Arrandale CPUs that lasted into late April). Once Sandy Bridge arrives early Q1 they should be swamped again.
I heard Intel had trouble with 32 nm in the beginning, which is why they limited it to some very small dual-core CPUs and some very high-dollar server CPUs. I bet now the lack of 32 nm mainstream parts is because they're starting to make some Sandy Bridge-based parts. The LGA1155-based parts are supposed to ship Q4 of this year and it's Q3 right now, so they better have at least some cooking in the fabs if they are going to have anything except a paper launch.
Sandy Bridge is basically going to make Bloomfield redundant. Why introduce a inferior, competing product?
The LGA1155-based Sandy Bridge units are a replacement for LGA1156-based systems. They won't make Bloomfield/Gulftown on LGA1366 any more redundant than the current LGA1156 i7s make the LGA1366 i7s redundant. The replacement for LGA1366 is LGA2011, which is supposed to be introduced in Q3-Q4 2011, so there is plenty of time to introduce 32 nm Gulftown derivatives on LGA1366.
Bulldozer isn't scheduled to arrive until the middle of 2011, maybe a few or a couple months before LGA2011. Intel and the market won't have very much time to react.
Just because Intel is washing itself in a bathtub filled with Benjamins doesn't mean that they are going to get generous. AMD doesn't want to get into a serious price war either because they still aren't in the greatest situation financially. I think we are going to see more of a standoff than a sell-off.
It will all depend on how fast Bulldozer is in comparison to Sandy Bridge. I believe Intel has already committed themselves to what Sandy Bridge is at this point as I doubt they have enough time to revise much of the chips between when Bulldozer probably will launch and when the high-end Sandy Bridge units will launch. About the only things they'll be able to change will be pricing. It seems like what you see with the LGA1155 units, you'll see with the LGA2011 units, but just more of it. Sandy Bridge is also a "tick," so Intel will be about two years away from being able to really give a good answer to Bulldozer if Bulldozer is grossly faster than Sandy Bridge. Thus Intel must be hoping Sandy Bridge matches up well with Bulldozer, else they're in for a couple of bumpy years until their big uarch redesign in Haswell comes out in 2013.
Facebook
Twitter
Instagram