HPCwire: AMD Redraws Server Processor Roadmap

The Leading Source for Global News and Information Covering the Ecosystem of High Productivity Computing

Leading HPC
Solution Providers

HPCwire >> Blogs

Blog: From the Editor

May 08, 2008

AMD Redraws Server Processor Roadmap

On Wednesday, AMD presented its revised server processor plans for the next couple of years. The roadmap included the upcoming 45nm Shanghai chip, new six- and twelve-core Opteron processors, and the next-generation socket for DDR3 and PCIe Gen 2. AMD's new path also gives us some idea why Cray decided to play nice with Intel.

During the briefing, Randy Allen, general manager of AMD's server and workstation division, said the company's plans to move from 65nm to 45nm are on track. According to him, the first 45nm server processor, "Shanghai," is scheduled to be delivered by the end of the year. Shanghai is a shrink of the 65nm quad-core Barcelona. The smaller componentry created enough extra silicon real estate for the chip designers to increase L2 cache from 256 MB to 512 MB and L3 cache from 2 MB to 6 MB. Besides more cache, AMD says Shanghai will get a 10 percent boost in memory performance and include support for HyperTransport 3.0. The good news for OEMs is that the new part drops into existing Barcelona (Rev F) sockets with just a BIOS upgrade.

At this point the roadmap heads off into new territory. Apparently plans for the once talked about eight-core "Sandtiger" Opteron have been scrapped. Instead, AMD will develop "Istanbul," a six-core Opteron that is pin-compatible with the current Rev F sockets. The chip is planned to be introduced in the second half of 2009.

Note that even if Shanghai and Istanbul hit their dates, they're going to be about a year behind their Xeon counterparts. Intel delivered its 45 nm quad-core "Harpertown" chips back in November 2007 and plans to ship the six-core "Dunnington" in the second half of this year.

Further out, AMD has a chance to catch Intel with the third generation Opteron processors. Two new offerings, the six-core "Sao Paolo" and the twelve-core "Magny Cours," will be developed for a new platform -- what the company's calling Socket G43. Taking a cue from Intel's multipackaging approach, the twelve-core Magny Cours is actually two six-core chips jammed into the same socket. The misspelled Sao Paolo (São Paulo) chip seems jinxed already. (AMD continues to name its silicon after Formula One venues.) Spelling error aside, the platform has lots of new goodies that should really rev throughput, including DDR3 memory support, four HyperTransport 3.0 links and PCI Express 2.0 support. The third generation Opterons are due out in the first half of 2010.

Toward the end of the briefing, Allen mentioned "Suzuka," a single-socket Opteron built on the 45nm process. The Suzuka is expected to be available in the second quarter of 2009. Like the 65nm Budapest before it (which is just now becoming available), Suzuka is aimed at workstations and, coincidentally, Cray supercomputers. I say coincidentally because the new Suzuka will emphasize higher clock speeds, which is great for single-threaded workstation applications, but less great for highly parallel supercomputer workloads. At this point, AMD has no plans to move beyond four cores for single-socket platforms, a fact I'm sure Cray had knowledge of well before it became public this week.

Compared to Intel with its upcoming dynamically-scalable Nehalem processors and 80-core teraflop experiments, AMD has taken a more conservative approach to the server processor space. Because of the company's recent financial hardships, AMD needs to aim directly at the market's sweet spot and not worry about developing bleeding-edge many-core processors or other exotic silicon for the uber-HPC crowd. With the recent loss of CTO Phil Hester, the Accelerated Processing Initiative that merges CPUs and GPUs on-chip might get shoved to the back burner (although this was by no means an HPC-only project). If AMD can deliver the hardware it has staked out on its new roadmap at prices that undercut comparable Intel parts and that deliver competitive performance per watt numbers, the company will continue to maintain a following with HPC system builders. Not that Intel is going to make that easy.

Posted by Michael Feldman - May 8 @ 6:11PM

Discussion

There are 0 discussion items posted.

Join the Discussion

You must be a registered user in order to comment on HPCwire stories.

Members enjoy:

	Access to Special Content, including Podcasts, Webcasts, and Whitepapers and more!
	Ability to take an active role in the discussions that are shaping the HPC technology discipline!
	Special offers available only to HPCwire members!

Become a Registered User Today!

Registered Users Log in Here to Comment

Email Address:	Password (case sensitive)

Remember me	Forgot Password?

Michael Feldman

Michael Feldman is the editor of HPCwire.

More Michael Feldman

Feature Articles

The New Science of Visual Analytics

When Jim Thomas set out to find new ways to deal with the mountains of information our society generates, he didn't just create a new organization, he created a new science. In this article we'll take a look at how the National Visualization and Analytics Center is transforming the problem of finding needles in haystacks into an opportunity for a more secure future.
Read More...

The Week in Review

ORNL Jaguar doubles its performance; the SC08 Cluster Challenge is gearing up; the University of Central Florida uses Army dollars to purchase an IBM super; and IBM's RoadRunner prepares to break the petaflop barrier. John West recaps those stories and more in our weekly wrap-up.
Read More...

Revaluating FPGAs for 64-bit Floating-Point Calculations

We now have generally available 2.5 GHz quad-core Opterons and Virtex-5 LX330, SX95T and recently announced SX240T FPGAs. In addition to this, Xilinx is releasing a new version of their floating-point cores that reduces the amount of logic and DSP slices needed for building floating-point function units. Taken together it is time to revisit Opteron floating-point performance versus FPGA performance.
Read More...

Read more HPCwire features...

Top Headlines

Parallel Processing Calls for a Fortress Mentality

May 14 | InfoWorld | Sun Microsystems is taking the lessons learned from Java and applying them to the application development challenges of the high performance computing realm. Read more...

IBM Set to Test the Fastest Computer in the World

May 14 | Computerworld | IBM is assembling the final pieces of what they hope will soon become the world's most powerful supercomputer. Read more...

IBM Shifts Cell to 65 Nanometers

May 13 | EETimes | IBM Corp. has announced the next-generation version of its Cell processor, the first specifically geared for computer servers. Read more...

Record-Setting Simulations on Ranger Reconstruct the Reionization Era

May 12 | Texas Advanced Computing Center | In the coming months, Dr. Michael L. Norman of UCSD will use Ranger, the world's most powerful supercomputer for open-science research, to perform the largest cosmological simulation to date. Read more...

2 AMD Executives Out in Restructuring Amid Slump

May 12 | BusinessWeek | Two executives have left AMD, including the head of the slumping chip maker's microprocessor division, as the company tries to engineer a dramatic turnaround to fend off larger rival Intel Corp. Read more...

Read more headlines...

Multimedia

Podcast: Interview with Ben Bennett of ClearSpeed Technology

Today, HPC organizations are requiring substantially more floating point performance to solve real-world problems. In this podcast, Ben Bennett, ClearSpeed General Manager, discusses how acceleration technology can improve the overall performance of standard x86-based systems...