backtop

Print E-mail del.icio.us 103 comment(s) - last by omnicronx.. on Apr 18 at 12:00 AM
Intel spills the rest of the beans on its new 45nm architecture, slated for launch this winter

With the launch and excitement of Intel's Silverthorne-based Atom processors behind it, Intel is turning its sights to the more salient and profitable next generation architecture for desktop, server, and mobile processors. Intel used the Spring IDF 2008 as a showcase for its Nehalem processor architecture.

Intel's Nehalem is truly a radical architecture departure from Intel thanks to its integrated memory controller that will support triple-channel DDR3-1333 memory. This won't be the only design element taken almost verbatim from AMD's playbook; Intel also plans to incorporate the new QuickPath Interface on Nehalem. QuickPath is almost identical in spirit and implementation to AMD's current interconnect technology, HyperTransport.

The first available Nehalem processors will be built on the existing 45nm manufacturing process, will incorporate SSE4 instructions, and will feature four fully integrated cores. Each core will have its own dedicated 256KB L2 cache and each core will share an 8MB of L3 cache pool.  The bulk of these 731 million transistor processors are dedicated to cache.

Event demonstrations at the Shanghai Intel Developer Forum, occurring now until the end of the week, show A1 silicon Bloomfield-based Nehalem processors at IDF at a speedy 3.2 GHz.

Like the 533 MHz variants of Intel's new Silverthorne-based Atom processors, Nehalem will also incorporate Simultaneous Multithreading (SMT) which is also known as Hyper-Threading (HT). 

Intel senior vice president Patrick Gelsinger confirmed that Xeon MP versions of Nehalem will eventually incorporate eight cores per processor, one upping the current Penryn-derived Dunnington processor the company plans to announce later this year.  These octo-core Nehalem processors will also use the newest iteration of Hyper-Threading, bringing the total count to 16 threads per chip.  And of course, these processors can be used in quad-socket configurations, bringing the processor market to 64 threads per mainboard.

Current Intel roadmaps peg the Nehalem launch date in Q4 2008, with a simulteanous rollout across servers and desktops. Since Nehalem uses a new architecture and transport bus, existing motherboards will not work with the new processors. 


Comments     Threshold


This article is over a month old, voting and posting comments is disabled
sounds familiar
By acejj26 on 4/2/2008 1:12:10 PM , Rating: 3
Didn't AMD showcase a 3.0 GHz Barcelona over a year ago? However they still haven't shipped anything close to that. What is the probability that Nehalem ships at or around this speed?




RE: sounds familiar
By KristopherKubicki (blog) on 4/2/2008 1:17:53 PM , Rating: 4
You're 100% correct on that. Last year I was criticized for mentioning that AMD's roadmap topped out at 2.6 GHz yet they demonstrated a 3.0 GHz chip. This time around, Intel's roadmap at least has a 3.2 GHz XE model on its roadmap. That will likely be the debut chip, with more mainstream chips coming in 2009.


RE: sounds familiar
By System48 on 4/2/2008 1:26:42 PM , Rating: 2
Exactly what I was going to say. I actually wish they would have a few slower models for the initial release, something that I could afford.


RE: sounds familiar
By ImSpartacus on 4/2/2008 2:59:13 PM , Rating: 2
That hurts them. They may not be up to snuff on production, so they can satisfy a low ultra-high end demand for a while until all of their fabs are cookin out chips, then they can support a larger mid-range userbase.

And they can leach $1k+ out of each sale versus 200-300$, so early adopters that may want to get a nehalem, but would normally want a ~$300 chip will get a $1000 chip (although that isn't very common).


RE: sounds familiar
By wordsworm on 4/4/2008 12:22:55 PM , Rating: 2
My guess is fabricating high end chips efficiently takes some time for serious refinement. How many bunk chips get produced at the high end for every good chip? Essentially it makes more sense to do things just the way they are.


RE: sounds familiar
By ImSpartacus on 4/6/2008 12:45:05 PM , Rating: 2
Something tells me the first high end chips that roll out aren't binned quite as high as the full production high end chips. I have no proof to back it up, but I am just assuming.


RE: sounds familiar
By batman4u on 4/2/2008 1:29:37 PM , Rating: 2
still, if they ship a 2.8Ghz CPu.... it will be rather Fast.

AMD must move quick....which sadly i doubt


RE: sounds familiar
By eye smite on 4/2/08, Rating: 0
RE: sounds familiar
By StevoLincolnite on 4/2/2008 10:01:55 PM , Rating: 5
AMD didn't invent the On-Die memory controller.


RE: sounds familiar
By eye smite on 4/2/08, Rating: 0
RE: sounds familiar
By redbone75 on 4/2/2008 11:45:40 PM , Rating: 5
quote:
No but they've had the most success with it and there's no denying that.

Didn't you mean to say "they've had their greatest success thus far" with it? There's no denying the benefit the on-die controller has had for AMD's architecture, but it tends to be a moot point considering the success of the Core architecture which doesn't have an on-die memory controller.

Yes, Intel is moving to an on-die controller, but they've already proven they can best AMD without copying them. To say anything otherwise smacks of fanboyism.


RE: sounds familiar
By vignyan on 4/3/2008 1:08:46 AM , Rating: 2
Point well made. Adding to that, i think the usage of same technology means little. its like saying that Daily tech copied Inquirer. Both are tech blogs and by far Daily tech is much better than Inq. even though Inq had used it first with quite some success.. :)

Not all processors are made equal.. And On-die memory controller does have a lot of limitations (processor upgrades to support newer memories, socket stability) which are commonly seen as problems with AMD (ignoring the under-performance issue)... Think! :P


RE: sounds familiar
By JumpingJack on 4/3/2008 3:23:52 AM , Rating: 3
Yes, it was a good point... it takes some study and reading to understand the memory hierarchy, mem+bus+cache work together to server one purpose, bring data and instructions to the core and keep it fed.

There are differnt design and implementation choices at each juncture, which is better a fast mem bus + small cache or slow mem bus + large cache. It depends really, the size and detail of the cache is dictated by performance goals at design, account for mem access latencies and speed.

The IMC allows AMD to implement smaller cache footprints while by suffering less far call penalties going to main memory compared to the current Intel approach, however, a large enough cache reduces misses such that much fewer (costly) memory accesses are required reducing the need for a fast mem bus.

This works for Intel really well for small memory footprints, and small working sets (i.e. DT/mobile like applications), their cache can be overwhelmed and performance held back though in some server workloads as shown in SpecFP_rate measurments and HPC applications.

For general purpose, Intel has demonstrated that an IMC is not the end all beat all that AMD would want you to think it is.


RE: sounds familiar
By eye smite on 4/3/08, Rating: 0
RE: sounds familiar
By DOSGuy on 4/3/2008 1:37:36 AM , Rating: 4
You're kidding, right? The Core 2 has not hit a bottleneck; it's one of the most overclockable architectures ever produced. Intel doesn't need to copy AMD to keep that going. They could sell much faster Core 2 processors, but they haven't needed to.

What happens next is where things get interesting. Intel already has a superior architecture that can clock to 4 GHz and beyond, and they're following that up with a processor that has a built-in memory controller and adding a second thread per core. They're moving much faster than they need to with the current lack of competition. If I didn't know better, I might think they were trying to run AMD out of business. Naturally they don't want to have a monopoly for legal reasons, but they definitely seem to be trying to grab as much of the pie as they can get away with.


RE: sounds familiar
By NullSubroutine on 4/3/2008 4:14:24 AM , Rating: 3
I think some people on these tech sites forget alot about how the CPU business world works.

AMD had the fastest chips around for quite a few years, did they reign supreme as far as dominating the market? No, of course not.

Having the fastest chip does not equal business success, all it succeeds in doing is getting a bunch of tech nerds all exicted about how many FPS they can get in their games.

I like to get the fastest computer gear as much as the next guy, but in terms of business AMD having or not having the fastest chip does not mean they will win or lose.

What matters is how many chips you sell and the most chips you sell are in the OEM cheap/mid range desktops/laptops like Dell, HP, etc. The majority of people buying PCs cannot tell the difference between a 2ghz dual core vs a 4ghz quad core. What they do understand is a computer that costs $500 vs that costs $2000.

AMD doesn't need the fastest processors to make fanboys of either side happy(or the 'fanboys of "whatever is fastest"') to survive. They need to produce products that allows them to remain competitive while getting enough market penetration into OEM's to sell volumes of chips.

As far as Nelahem (however you spell it I forgot), it doesnt have anything to do with 'maintaining supriority over desktop computing' it has to do with producing a chip that scales as good as the Opteron does when more core are introduced into the system, where the Opteron still has an advantage.

The reason why you see Intel not producing faster binned processors is because there is no competition and for them to produce faster binned processors (even though they have head room) means they will have lower yields. Unless their market % is threatened they will not lower yields (and lose money).


RE: sounds familiar
By NullSubroutine on 4/3/2008 7:27:10 AM , Rating: 2
http://www.fudzilla.com/index.php?option=com_conte...

96.6% of all CPUs sold are below $200. I rest my case!


RE: sounds familiar
By Etsp on 4/3/2008 11:10:44 AM , Rating: 2
If I were you, I wouldn't rest any case based solely on a link to fudzilla... If I remember correctly, fudzilla was started by the ATI fanboy that used to write for the Inq a while back...