Zibb

Brian DipertEDN Senior Technical Editor Brian Dipert exposes, analyzes and
opines on diverse topics in technology. Follow the Brian's Brain Twitter feed at www.twitter.com/BrianzBrain.



   Advertisement

Profile

RSS Feed

  • Add this blog to your RSS newsreader!

Recent Posts

Recent Comments

Most Commented On

Archives

By Category

Consumer Electronics Design Articles

Blog

Tuesday, June 2, 2009

AMD vs Intel: Computex Round Three, Intel Makes Core i7 Both Faster and Less Costly

Jun 2 2009 10:40PM | Permalink |Comments (4) |


AMD isn't the only x86 CPU company focused on cost-reducing its high-end devices, although strictly speaking, none of the information I'm about to pass along to you is (yet) official. Don't worry, I'm not breaking any embargoes; Intel hasn't (yet) briefed me on its Core i7 proliferation plans, nor has it (yet) publicly released the potential products I'm about to share with you (as I write this at 9PM PT on Tuesday night). But innumerable leaks, coupled with product samples already mounted on functional system boards, give a pretty good indication of where the company's headed.

Intel's Core i7 product line, the first fruits of the company's latest-and-greatest Nehalem microarchitecture, is a product without x86 peer from a raw performance standpoint, as innumerable benchmark test results to date have confirmed. Intel has historically used its Core microarchitecture- and 45 nm lithography-based Penryn microprocessor, in single-die (dual-core) and dual-die (quad-core) configurations, to match up against AMD's 65 nm Athlon X2 and Phenom CPUs. However, now that AMD's also up and running on 45 nm, in a limited-production fashion with Athlon II and Phenom II, Intel's going to need to also up the ante to remain competitive in these more conventional product categories.

To maintain its Core i7 leadership, numerous partner and customer leaks suggest that Intel's poised to release two new high-end product variants running at 3.06 and 3.33 GHz clock frequencies. Here's how they stack up against the three current Core i7 CPUs (with the two new processors in italics):

Product

Clock Speed

Number of cores/threads (the latter due to HyperThreading Support)

Price (quantity)

Core i7-965 Extreme

3.2 GHz (unlocked)

4/8

$999 (1,000)

Core i7-940

2.93 GHz

4/8

$562 (1,000)

Core i7-920

2.66 GHz

4/8

 $284 (1,000)

Core i7-975 Extreme

3.33 GHz (unlocked)

4/8

$1,079.87 (1)

Core i7-950

3.06 GHz

4/8

$649 (1)

And to keep pace with mainstream Athlon II and Phenom II, Anand Lal Shimpi (who I confess with a bit of chagrin appears to have better contacts at Taiwanese motherboard manufacturers than I do) recently published an Intel-unsanctioned preview of the company's upcoming Lynnfield processors. Unlike with AMD's recent Phenom II-to-Athlon II die chop, Lynnfield will have identical core counts and cache allocations to today's Core i7 CPUs. However, Lynnfield removes one of the three SDRAM controllers (and consequent external memory buses) integrated in Core i7. Here's what today's Core i7 system configuration looks like:

Using the pincount and silicon real estate freed up by the DRAM controller-and-bus amputation, Intel has reportedly instead embedded 16 lanes' worth of PCI Express v2 transceivers along with the corresponding controller logic within the CPU. This integration move effectively eliminates the need for the separate IOH (I/O Hub) chip used with Core i7, and it also simplifies and speed-reduces (from QPI to DMI) the interconnect bus between the Lynnfield CPU and the remaining piece of core logic, the PCH (an acronym rename of the prior ICH, i.e. I/O Controller Hub).

What will Intel use as the marketing moniker for these new CPUs, assuming Anand's right and they actually exist? Here's my theory; I think that the '7' in the current Core i7 CPUs is the sum of the physical CPU cores (4) and DRAM controllers (3)...4+3=7. Extrapolating this naming convention to Lynnfield, with one less DRAM controller, results in 'Core i6'. We shall see...

How will AMD respond? So far in this writeup, I've focused on performance metrics, but cost is an equally important characteristic. As such, AMD can continue its practice of recent years in slashing prices to maintain a competitive price/performance combination versus its much larger nemesis. However, the profitability of AMD's strategy is by no means guaranteed, particularly when you consider that Intel ramped its 45 nm process into production more than a year ahead of its key competitor and appears to have a comparable (if not larger) current lead in the race to 32 nm.

And what's Intel's next integration step? Consider that the most common use of the 16-lane PCI-E bus will be to direct-connect the CPU to a GPU (graphics processor). Therefore, although I have at this point absolutely no hard data to back up my prognosis, I suspect that lower-end Nehalem-derived products (both at the 45 nm and follow-on 32 nm process nodes) will trade off CPU cores for graphics cores...first those found today in the company's core logic chips, with a legacy stretching back to the i740, and later those derived from the upcoming and x86-derived Larrabee GPU.

Followup: Intel's two new high-end Core i7 CPUs are now public. I've been promised review samples of each soon. Here's updated pricing, along with a product reordering versus the prior table:

Product

Clock Speed

Number of cores/threads (the latter due to HyperThreading Support)

Price (quantity)

Core i7-975 Extreme

3.33 GHz (unlocked)

4/8

$999 (1,000)

Core i7-965 Extreme

3.2 GHz (unlocked)

4/8

$999 (1,000)

Core i7-950

3.06 GHz

4/8

$562 (1,000)

Core i7-940

2.93 GHz

4/8

$562 (1,000)

Core i7-920

2.66 GHz

4/8

 $284 (1,000)

Perhaps obviously, judging from the prices, the Core i7-975 Extreme replaces the Core i7-965 Extreme, and the Core i7-950 replaces the Core i7-940.


Reader Comments



at 6/3/2009 1:29:00 PM, Stiggle said:

Yet another socket change I suppose. Might as well make the chip fixed to the Motherboard and have the whole motherboard as a quick change solution. Seems like Intel is following the Microsoft multiple iterations and upgrade paths model. I am wondering how long it will be before Intel or AMD have FPGA logic on the CPU for loading new patches etc.




at 6/3/2009 1:33:55 PM, Brian Dipert said:
Dear Stiggle, the chips are already heavily microcode-based, therefore updateable via 'firmware upgrades'. I'll write more about this in a day or few, but Via recently reported to me (at ESC West) that Nano (which I tested for an early January cover story: www.edn.com/article/CA6625435.html) got a notable performance boost/power consumption reduction via such a microcode tweak. That's why we're only now seeing production systems based on Nano; the OEMs were waiting to qualify this fix



at 6/3/2009 2:55:16 PM, John said:

This looks a lot like a product road map that I saw about 2 years ago when I was workng in the server validation lab at Intel. They realized that CSI (QPI) was a non-starter for comsumer level boards (the layer count required was too much), and they started looking at SOC options to combine the processor and ICH. This may be the first step in that direction.




at 6/4/2009 3:22:28 AM, AlexAxe said:
Hello,
Can i take a one small pic from your blog?
Thanks
AlexAxe

Post a comment



Display Name

Change Image
Before submitting this form, please type the characters displayed above.
Note the letters are NOT case sensitive.


ADVERTISEMENT

©1997-2009 Reed Business Information, a division of Reed Elsevier Inc. All rights reserved.
Use of this Web site is subject to its Terms of Use | Privacy Policy

Please visit these other Reed Business sites