Print 17 comment(s) - last by JumpingJack.. on May 29 at 8:19 PM

(Click to Enlarge)

(Click to Enlarge)
New error correction and integrated DDR3 memory controllers are coming for some of Intel's most expensive products

Intel launched its Xeon 5500 series of server CPUs using Nehalem architecture at the end of March. Using lessons learned from the Core i7, the world's largest semiconductor manufacturer managed to achieve impressive performance gains while cutting power consumption dramatically. Sales have been impressive, and the company expects that it will account for 50% of all dual socket (2S) server sales by August.

DailyTech previously showed part of Intel's server roadmap, detailing a follow-up Xeon processor series known as Nehalem-EX. Intel is now releasing details about Nehalem-EX in order for its customers to include it in their planning.

Each Nehalem-EX processor will feature up to eight cores on a single chip, up to 24MB of shared cache, and support 16 threads with Hyper-Threading. The chip is built using Intel's P1266 45nm High-K metal gate technology, and will be comprised of 2.3 billion transistors. It will use the new Boxboro-EX chipset, and be able to support up to 16 DIMMs per socket. Eight socket systems capable of processing 128 threads simultaneously will be possible, since each chip will feature four Quick Path Interconnects.

Intel is also moving away from Fully-Buffered DIMMs to DDR3 Registered ECC DIMMs. It will use its Scalable Memory Interconnect, which has its own internal memory buffers. A new feature known as Machine Check Architecture (MCA) recovery will detect CPU, memory, and I/O errors. It is designed to work with operating systems to correct and recover from otherwise fatal system errors, thus maintaining critical uptime. Microsoft, RedHat, Novell, and VMware are already promising support in their products.

Multi-socket capable Xeon processors are typically used in large corporations, universities, and research institutions. Often, it is used to create supercomputers for High Performance Computing (HPC) applications, such as modeling the decay of nuclear waste, simulating tectonic stresses to predict earthquakes, and examining protein folding. Uptime and reliability is critical, as any errors will mean that the modeling run will have to be restarted. HPC servers are in high demand, with computer time booked months in advance.

Intel's previous champion was the Xeon 7400 series, formerly known as Dunnington. It uses DDR2 memory running at the same 1066MHz as its Front Side Bus. Compared against the Xeon 7400, Nehalem should have 9 times the memory bandwidth. Intel is also claiming 2.5 times greater database performance, 1.7 times integer throughput, and 2.2 times floating point throughput.

Nehalem-EX is codenamed Beckton within Intel, which will only say that it is scheduled for released in the second half of the year. The firm may choose to launch Beckton at the Intel Developer Forum at the end of September, with sales at the beginning of the fourth quarter as the Xeon 7500 series.

Comments     Threshold

This article is over a month old, voting and posting comments is disabled

By Moishe on 5/27/2009 11:24:01 AM , Rating: 2
Intel is kickin' arse.

The MCA thing should be a good thing for the enterprise customers as well where uptime is highly valued.

RE: wow...
By BaronMatrix on 5/27/09, Rating: -1
RE: wow...
By niva on 5/27/2009 1:21:19 PM , Rating: 5
Baron, what the hell are you talking about? I'm an AMD fanboy and as much as I like their products I have to admit that they have absolutely no answer for what Intel currently has out on the market. When this new CPU comes out AMD will have a nearly impossible job to catch up.

I too wish there was more hardware manufacturers at this level. Unfortunately it's becoming near impossible for someone to just enter the semi conductor market and compete at this level. I think we need to wait for the next generation computers, maybe when optical CPUs come out it will be more than just two companies but for now we have Intel and far behind them AMD is trailing at this stage.


RE: wow...
By EricMartello on 5/27/2009 5:17:35 PM , Rating: 3
Yeah, AMD's current CPU products are lacking by comparison to Intel's offerings but Baron is right - AMD CPUs have enjoyed the benefits of an on-die memory controller for many years and just now Intel is finally making the switch. I'd also like to point out that it has been just as many years since AMD has put that kind of innovative design into practice.

If I were AMD I'd be think about merging GPU and CPU a bit closer together (since they put all that money in to acquiring ATI). With many people looking for sub-$500 systems, especially laptops, AMD could solidify itself in the low-power, highly integrated market. Imagine a fully integrated CPU/GPU on a Mini ITX MB that has the power of a mid-range/high-end Radeon. They could also provide that tech as an OEM for media appliances like STBs and PVRs.

In any case, AMD probably won't fare too well if they try to take Intel head-on in raw performance, so they need to be innovative like they were in the past.

RE: wow...
By GodisanAtheist on 5/27/2009 6:25:27 PM , Rating: 2
How do the patent rights work for something like the x86 architecture? Intel released their first x86 chip in 1986 if I'm not wrong, therefore the patent for the x86 instruction set would have had to have been granted before or shortly after release.

Don't patents expire after 20 years? Shouldn't the x86 license be nearing its sell-by date?

Or have I really mucked things up?

RE: wow...
By EricMartello on 5/27/2009 9:45:57 PM , Rating: 2
As far as I know the x86 architecture is not "open" but I'm not sure if it needs to be licensed from Intel or not for a manufacturer to make an x86 compatible CPU...but that really isn't what's holding AMD and other competition back.

While the x86 instruction set is universal, the internal approach to processing programs written for that architecture vary, and typically companies with bigger R&D budgets have the advantage (and are able to produce CPUs with superior performance). In the past, AMD was able to gain a foothold by making CPUs that rivaled or outperformed the Intel counterparts at 40-60% of the cost.

That was when AMD was motivated...but now it's like they're in so much debt from buying out ATI that they killed their R&D budget for CPUs. Meanwhile, Intel as been owning the processor market ever since it's C2D line of processors emerged. They beat AMD across the board and even though they cost more, they offer better price:performance, especially in games which is what sells a lot of high end hardware.

RE: wow...
By AstroCreep on 5/27/2009 11:50:17 PM , Rating: 2
...I'm not sure if it needs to be licensed from Intel...

My understanding is "Yes", because recently when AMD announced that it was spinning off the fab-unit as a separate business (GlobalFoundries) Intel cried foul and said that act "Violates" the license AMD bought to design/manufacture x86-compatible CPUs.
Whether or not if the claim by Intel about the AMD split is valid is yet to be seen.

I also seem to remember reading some speculation in the last year about NVIDIA possibly getting a license to make their own x86 CPUs (speculation that was heavily fueled by the Ion platform), but speculation was simply all it was.

RE: wow...
By EricMartello on 5/28/2009 12:30:33 AM , Rating: 2
It's too bad that we are so rooted in would be nice to move away from CISC to RISC which is an inherently more efficient processor architecture, and it would have to be an OPEN standard.

RE: wow...
By Regs on 5/28/2009 11:23:38 PM , Rating: 2
They're actually becoming more inheritably alike. Like economics of scale and the possibilities curve, there's always a trade off. Software companies aren't always going to use the most resources to program for risc, and not every PC manufacture is going to use CISC because they're expensive, run hotter, and sometimes slower.

RE: wow...
By JumpingJack on 5/29/2009 8:18:32 PM , Rating: 2
Intel released the 8086, hence the derived reference 'x86', in 1978.

Patents have a 20 year term from the initial filing date or 17 years from the grant date. However, since 1978 much more technology has gone into microprocessors, such as extensions, new instructions, yada yada, much of the modern x86 processor is still protected under a bevy of patents.

For example, the Pentium Pro was launched in 1995, and it was the first intel super scalar x86 CPU. It is not clear to me if Intel had patented various architectural features from the Pentium Pro since some were also being implemented in NexGEN and Cyrix processors.

The general progression of the x86 CPU has made IP from both AMD and Intel extremely intertwined.

RE: wow...
By JumpingJack on 5/29/2009 8:19:40 PM , Rating: 2
Yeah, poor Baron -- he has very little understanding of CPUs. He once recommended to someone that they should store their processor in styrofoam.

RE: wow...
By bruce24 on 5/27/2009 1:25:31 PM , Rating: 1
Anand's own recent tests show that real-world usage doesn't favor Nehalem near as much. Maybe all the customers will want to run VMMark all day.

No what it shows is that with a different benchmark, vApus Mark1, the current Opteron gets closer but is still behind. Also note they felt the need to say "However, we are well aware that the current vApus Mark I has its limitations."

AMD should have the 4P crown with Istanbul as it goes against Dunnington, then in 7 or 8 months Nehalem-EX should have no problem blowing past Istanbul and it will be very interesting to see if it also tops Magny-Cours.

RE: wow...
By Justin Time on 5/28/2009 12:04:58 AM , Rating: 1
Tech Leader? Depends on what you call "leading".

Most of AMD's so-called "innovations" show a remarkable similarity to those from the DEC/Compaq/HP Alpha... which was eventually passed on to Intel.

RE: wow...
By Pryde on 5/28/2009 1:54:49 AM , Rating: 3
I agree but don't forget about Timna, if it had reached the market then Intel would of been first with a IMC but failed because it was a RAMBUS controller.

AMD has been nothing but a copycat to other designs, this is why they will never beat Intel in the long run. Intel had a bad run with P4 but before long Core 2 was out and now we have Nehalem. AMD doesn't have anything new in the pipeline until after 2010 and then if Intel keeps to their current road map they will be facing 32nm Sandy Bridge.

RE: wow...
By winterspan on 5/28/2009 6:56:02 AM , Rating: 1
That virtualization test is far from conclusive, not least because they didn't even use the 4.0 version of the software which supports Intel's EPT --- whereas it does support AMD's version of EPT. This makes an enormous difference in performance (25% or more) alone, and there are many other factors involved as well.

Just wait and see for the final benchmarks tested using the newest releases that finally support all of Nehalem's virtualization features.

Also, AMD's 6-core Instanbul is the only thing coming out anytime soon, and Nehalem-EX will completely blow it away (unfortunately in my opinion).

I love AMD and competition as much as anyone, but you have to be realistic. They are still in dire circumstances and really need to make some stuff happen over the next 24-36 months.

"We are going to continue to work with them to make sure they understand the reality of the Internet.  A lot of these people don't have Ph.Ds, and they don't have a degree in computer science." -- RIM co-CEO Michael Lazaridis

Most Popular ArticlesSmartphone Screen Protectors – What To Look For
September 21, 2016, 9:33 AM
UN Meeting to Tackle Antimicrobial Resistance
September 21, 2016, 9:52 AM
Walmart may get "Robot Shopping Carts?"
September 17, 2016, 6:01 AM
5 Cases for iPhone 7 and 7 iPhone Plus
September 18, 2016, 10:08 AM
Update: Problem-Free Galaxy Note7s CPSC Approved
September 22, 2016, 5:30 AM

Copyright 2016 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki