Nehalem Article @ RWT + 3.2GHz samples(?)

Thursday 03rd April 2008, 06:50:00 PM, written by Arun

David Kanter at Real World Technologies just published his usual brand of excellent architecture analysis for Intel's upcoming Nehalem processor - there aren't many changes, but it's still well worth the read. At the same time, DT says an A1 Nehalem system at IDF is running at an impressive 3.2GHz.

Given the higher per-clock performance of the architecture, as described by David Kanter, 3.2GHz+ would truly result in some truly terrifying performance numbers. This would make AMD's Shanghai unable to compete outside the low-end of the server market (and the mid-range for desktops).

We remain skeptical for now, of course; however, in the RWT discussion thread on David's article, people are pointing out that the 4 cycles access latency for the L1 Data Cache might indicate that Intel really is planning on delivering higher (rather than merely identical or lower) clocks for Nehalem compared to Penryn and its derivatives. This is despite a substantially higher core die size, so it will be interesting to see what power consumption looks like and what Intel has done to minimize the problem.

One quick note: in addition to David's article, it may be worth reading one of our previous news pieces on Nehalem's variants and sockets as we tried there to give more precise information than others have done so far.

UPDATE: Charlie @ The Inquirer claims DailyTech/Hexus were mistaken and that the 3.2GHz CPUs were Harpertowns (although we aren't sure why it'd be reported as 8 threads then? were they 2P?) and the real Nehalems were reported as 2.56GHz chips clocked at 2.13GHz.


Discuss on the forums

Tagging

intel ± nehalem, bloomfield, 45nm


Latest Thread Comments (14 total)
Posted by pjbliverpool on Thursday, 03-Apr-08 20:34:44 UTC
Quoting ShaidarHaran
You assume incorrectly. High-end = Extreme edition.
Thats exactly what I thought before reading this article. But if Bloomfield = Extreme edition then it would imply no mainstream quad core Nehalem at all. I.e, you want a Nehalem and don't want to pay ~$1000 for an Extreme edition, you have to get a dual core. That to me makes little sense and it significantly diminishes Nehalems value as a desktop CPU for none budget consumers. I mean, given the choice between a quad Penryn or a dual Nehalem (minus SMT and triple lane memory) I think many would choose the quad Penryn at the same price. Of course, if performance is good enough the dual Nehalem may still be a better option.

Posted by ShaidarHaran on Thursday, 03-Apr-08 20:47:49 UTC
Quoting pjbliverpool
Thats exactly what I thought before reading this article. But if Bloomfield = Extreme edition then it would imply no mainstream quad core Nehalem at all.

I.e, you want a Nehalem and don't want to pay ~$1000 for an Extreme edition, you have to get a dual core.

That to me makes little sense and it significantly diminishes Nehalems value as a desktop CPU for none budget consumers. I mean, given the choice between a quad Penryn or a dual Nehalem (minus SMT and triple lane memory) I think many would choose the quad Penryn at the same price.

Of course, if performance is good enough the dual Nehalem may still be a better option.
I read nothing that stated the tri-channel DDR3 MC is to be used in all quad Nehalems. You'll still be able to buy a non-extreme Nehalem minus an MC channel and some L3 cache, likely.

Posted by pjbliverpool on Thursday, 03-Apr-08 20:51:35 UTC
Quoting ShaidarHaran
I read nothing that stated the tri-channel DDR3 MC is to be used in all quad Nehalems. You'll still be able to buy a non-extreme Nehalem minus an MC channel and some L3 cache, likely.
According to that article the only other quad core aside from bloomfield and the Gainestown Xeon is a mobile chip. So on the desktop, its Bloomfield or nothing - according to the article.

Posted by ShaidarHaran on Thursday, 03-Apr-08 20:54:42 UTC
Quoting pjbliverpool
According to that article the only other quad core aside from bloomfield and the Gainestown Xeon is a mobile chip.

So on the desktop, its Bloomfield or nothing - according to the article.
Lower-end parts can have an MC channel and some cache disabled for yield and lower costs. Just because it's there doesn't mean Intel won't disable it :p

Posted by Arun on Thursday, 03-Apr-08 20:56:24 UTC
Quoting pjbliverpool
So on the desktop, its Bloomfield or nothing - according to the article.
Well, not according to me! ;) (read my news piece on the frontpage which links to this older news piece (http://www.beyond3d.com/content/news/540) with the key info)

Posted by pjbliverpool on Thursday, 03-Apr-08 21:23:16 UTC
Quoting Arun
Well, not according to me! ;) (read my news piece on the frontpage which links to this older news piece (http://www.beyond3d.com/content/news/540) with the key info)
Cheers, it seems Lynnefield was the piece of the puzzle I was missing. I thought I remembered something about a lower end quad core!

Posted by Rune on Saturday, 05-Apr-08 17:05:24 UTC
Well according to Charlie from the inq , the chip, that Hexus saw running at 3,2 Ghz, was not a Nehalem, but a Harpertown.

http://realworldtech.com/forums/index.cfm?action=detail&id=89005&threadid=88864&roomid=2

Posted by stevem on Sunday, 06-Apr-08 04:22:31 UTC
Quoting ShaidarHaran
Lower-end parts can have an MC channel and some cache disabled for yield and lower costs. Just because it's there doesn't mean Intel won't disable it :p
Yes, a lot can be done in the "uncore" without affecting the core unit(s).

The tech looks great, just wish we were at Sandy Bay already. Love the sound of AVX...

Posted by ShaidarHaran on Sunday, 06-Apr-08 14:29:18 UTC
re: Sandy Bridge - Doubling (++) fp performance is generally not a bad thing to do...

Posted by stevem on Monday, 07-Apr-08 07:29:49 UTC
Yeah, Sandy Bridge...


Add your comment in the forums

Related intel News

RWT explores Haswell's eDRAM for graphics
RWT: An Updated Look at Intel's Quick Path Interconnect
32nm sixsome over at RealWorldTech
Intel Core i3 and i5 processors launched
Analysis: Intel-TSMC announcement more complex than reported
Intel and TSMC join forces to further Atom
Fudzilla: Intel 45nm Havendale MCM replaced by 32nm+45nm MCM
Intel announce Core i7 processors, reviews show up
Intel's Aaron Coday talks to Develop about Larrabee
Larrabee to also be presented at Hot Chips