Intel's socket plans?

Status
Not open for further replies.

metallifux

Distinguished
Apr 26, 2010
224
0
18,690
Can someone give me the simplified plan for intels sockets this year? I've read multiple reports that Ivy Bridge will be coming by 3rd Q this year and will be a 22nm shrink of Sandy which will give huge power reductions but what socket will it use? If we set up a sandy system now will we be able to drop in an Ivy CPU next Xmas?
 
Solution


LGA2011 appears to be the successor to LGA1567, which also has a quad-channel memory controller. LGA1567 works only with Xeon 6500s and 7500s, which are relatively low-clocked chips with enormous dies and enormous price tags. You wouldn't want one for a gaming system since the least-expensive LGA1567 chip is the Xeon E6510, which is a 1.73 GHz, 105-watt quad-core with HyperThreading, 12 MB L3 cache, no Turbo Boost, and its IMC capped...

metallifux

Distinguished
Apr 26, 2010
224
0
18,690
Wikipedia has just been updated...

Ivy Bridge is the codename given to the 22 nm die shrink of the Sandy Bridge architecture. According to the keynote speech presented by Paul Otellini during the 2010 Intel Developer Forum (IDF), Ivy Bridge processors may be introduced as early as the second half of 2011.[37] However, it now seems more likely that Ivy Bridge will be launched a year after Sandy Bridge, probably during CES 2012. Ivy Bridge will replace dual-core processors with quad-core processors at the entry level segment, while delivering eight-core processors for mainstream and higher-end level segments.[38]

Ivy Bridge processors will employ a maximum of 24 EUs graphics sub-system (double that of Sandy Bridge)[39] and also provide DirectX 11 support, rather than the older DirectX 10.1 and OpenGL 3 graphics technology that will be featured in the first batch of Sandy Bridge processors, while retaining its LGA 1155 socket compatibility
 

metallifux

Distinguished
Apr 26, 2010
224
0
18,690



well if your keen on having the latest then its not a bad deal as it looks like ivy will work LGA 1155 so you can get a Sandy set up now then drop in an Ivy later. However if you want quad channel RAM and possibly native USB 3.0, you will need to wait till Q3 this year for the LGA 2011 socket.

:(

I myself was going to wait until bulldozer, but since that is only going to be coming out in April and will at best be comparable to Sandy since it will be built on 32nm i'm now reconsidering intel.
 


I dunno about the benefits of quad-channel for desktop - IIRC the Nehalem tests on some review site where they benched with all 3 channels and then again with one bank pulled showed the tri-channel vs. dual-channel difference to be something like 1% in most apps. But quad would be a great way to double your memory using the cheaper 2-gig modules, if you need a lot of RAM.

 

willmalcom

Distinguished
Mar 1, 2010
366
0
18,810


Thanks for the find!

I had some of these same questions, and this is a big help!
 


What we know:

1. LGA1156 Lynnfield/Clarkdale -> LGA1155 for 2- and 4-core Sandy Bridge(-DT.) Done.
2. LGA1366 Bloomfield/Gainestown/Gulftown/Westmere -> LGA1356 for 6 and 8-core desktops and 1-2P Xeons with 2-8 cores. Scheduled for Q3 2011.
3. LGA1567 Beckton/Westmere-EX -> LGA2011 for 4-8 core Sandy Bridge-EP 1-2P server and 8-core Sandy-Bridge-EX 4P servers. -EP is scheduled for Q3 2011 and -EX for Q4 2011.

Ivy Bridge is supposed to be a drop-in upgrade for Sandy Bridge in LGA1155 sockets and I bet that it would be for LGA1356 and LGA2011 as well. Intel has done some sockets that only fit one revision of a chip (PPGA370, Socket 423, PAC418) but generally they support at least a new chip and its die shrink.
 


LGA2011 appears to be the successor to LGA1567, which also has a quad-channel memory controller. LGA1567 works only with Xeon 6500s and 7500s, which are relatively low-clocked chips with enormous dies and enormous price tags. You wouldn't want one for a gaming system since the least-expensive LGA1567 chip is the Xeon E6510, which is a 1.73 GHz, 105-watt quad-core with HyperThreading, 12 MB L3 cache, no Turbo Boost, and its IMC capped at DDR3-800 speeds, yet it costs nearly as much as a Core i7 970 ($744.) Good luck finding a board for that too, as the only one that I can find is a massive quad-socket Supermicro unit that costs a fuzz over a grand.

What you want is one of the LGA1356 units that have triple-channel memory and 8-core CPUs with Turbo Boost and the all of the rest of the bells and whistles and a considerably lower price.

I myself was going to wait until bulldozer, but since that is only going to be coming out in April and will at best be comparable to Sandy since it will be built on 32nm i'm now reconsidering intel.

I'd wait for Bulldozer to ship before automatically writing it off as "at best be comparable to Sandy Bridge." AMD has a history of pretty significant performance increases with new or heavily-reworked microarchitectures- think of K7 (all-new) and K8 (heavily reworked.) Anyway, process node doesn't mean squat for performance. If you follow that line of thinking, Intel's 65 nm Pentium D Preslers should have been far faster than AMD's 90 nm X2 Manchesters and Toledos. We all know how that one turned out. Microarchitecture >> process node for determining processor performance.



4 GB DDR3 modules are pretty much exactly double the price of 2 GB DDR3 modules right now, so there's really no reason to not go for 4 GB modules if you need a lot of RAM. If you need more than 24 GB of RAM, get a multiple-CPU workstation with support for registered memory and two CPUs' worth of DIMM slots.
 
Solution


Hmm, haven't checked memory prices lately but I see that some of the GSkill ripjaws are about equal in price for 2 x 4GB modules vs. 4 x 2GB, same CAS, on Newegg anyway.

 


Testing a 32-bit app, even in a 64-bit version of windows, is automatically an invalid test, as all 32-bit apps, even running in 64-bit windows, are bound by the 2GB Address Space (3GB if the LARGEADDRESSAWARE switch is toggled) due to the way the program is compiled. As such, adding more RAM, or even more channels, won't improve performance.

I view most every one of these memory tests as fundamentally flawed for that very reason. Test with a 64-bit executable on win64, and then see where performance actually falls.
 
Status
Not open for further replies.