As this year comes to a close, AMD executives won't be sorry to put 2007 behind them. Battered by rising debt, shrinking market share and mounting financial losses, AMD is counting on a comeback in 2008. But can the struggling chip maker make good on its promises to customers and investors that better times lie ahead?
Many of AMD's current troubles are tied to Barcelona, the first version of its Quad-Core Opteron processor. Repeated delays of this chip weakened AMD's finances and undermined confidence in the company's ability to deliver new products as promised. Rebuilding that lost confidence among corporate users and the financial community won't be easy and the company can ill afford further missteps.
Over the next six months, there are several milestones that AMD must hit in order to get back on track. The first, and arguably most important, involves volume shipments of Barcelona, scheduled to happen during the first quarter. AMD desperately needs this chip to shore up flagging revenue from its commercial business and to take back ground lost recently to Intel in the server market.
With a bug that was recently discovered in Barcelona fixed, AMD expects the latest version of the chip to start coming out of its manufacturing plants in January. Barcelona shipments are expected to rise throughout the first quarter, reaching "full volumes" and widespread availability of servers based on the chip during the second quarter, according to Phil Hughes, an AMD spokesman.
Another early milestone to look out for involves Shanghai, the second version of the Quad-Core Opteron chip that's set for volume production during the second half of 2008. AMD will produce Shanghai with a 45-nanometer manufacturing process that is one generation ahead of the 65-nanometer process used to make Barcelona.
The numbers used to describe chip manufacturing processes refer to the average size of the smallest features that can be created on a chip. The smaller these features can be made, the better. Making these features smaller gives chip makers the ability to shrink the overall size of a chip, thereby reducing unit manufacturing costs, or to add new features. Chips made using a more advanced process may also consume less power and run faster than other versions. But semiconductor production is as much art as it is science, and moving to a more advanced process is fraught with risks.
In the case of Shanghai, AMD will take advantage of the 45-nanometer process to triple the size of the Level 3 (L3) cache, from 2M bytes in Barcelona to 6M bytes. The L3 cache is a shared pool of memory that can be accessed by all four cores on the chip. Each core has 512K bytes of Level 2 (L2) cache. When data is moved out of the L2 cache, it moves into the L3 cache instead of main memory, meaning that the core can access this information more quickly than would otherwise be possible.
Like AMD's existing server lineup, Barcelona and Shanghai also include an on-chip memory controller that further helps to increase performance.
The memory structure used in the Quad-Core Opteron is different from Intel's server chips, which do not use an L3 cache and do not have an on-chip memory controller. Instead, Intel's quad-core Xeon 7300 server chips have 8M bytes of shared L2 cache. Since Intel processors actually pack two dual-core chips inside a single package, the L2 cache is divided into two parts, with 4M bytes on each silicon chip. Intel's processors do not use an on-chip memory controller.
Boosting the Quad-Core Opteron's L3 cache with Shanghai should give AMD's quad-core chips the opportunity to stretch their legs, but the company has to make them first. The first working samples of Shanghai are due out of AMD's factory in Dresden, Germany, in January, according to recent remarks by Mario Rivas, executive vice president of AMD's Computing Products Group. If these samples arrive late or suffer from unexpected technical problems, this could delay volume production of Shanghai to late 2008 or even early 2009. Conversely, working samples that arrive on schedule and meet or exceed expectations would give AMD an added boost.
Besides the server segment, there are a couple of key product milestones onAMD's roadmap for desktop and mobile processors that bear watching as well.
For example, the company plans a first-quarter rollout of Perseus, a package of chips aimed at PCs for the corporate market and centered on a quad-core or triple-core processor and an RS780 or SB700 chipset. A dual-core version will also be rolled out. During the second quarter, AMD plans to roll out Puma, a chip package designed for notebooks. Puma is based on a new processor, called Griffin, expected to offer better performance and power efficiency than AMD's existing line of mobile processors.
AMD plans to offer additional details of its plans and upcoming products at the Consumer Electronics Show to be held in Las Vegas during January, Hughes said.
Read up on the latest ideas and technologies from companies that sell hardware, software and services. Email Archiving Implementation: Five Costly Mistakes to Avoid
Data grids and service-oriented architecture
Controlling storage costs with Oracle database 11g
Making the Business Case for IT Consolidation
Delivering the Power of Choice with Microsoft Dynamics CRM
Achieving the impossible: Unlimited application scalability
IT Service Management Needs and Adoption Trends: An Analysis of a Global Survey of IT Executives
Taking On Demand CRM Integration to the Next Level
Zones provide focussed content from Computerworld and leading technology partners.Discover how SOA can create smarter outcomes for your business.
Attend and learn:
- How SOA is helping leading companies to become more agile
- Where you should be applying SOA processes in your company
- The top SOA implementation mistakes to avoid
Click here for more information.
- +
Computerworld Live Podcast #98: The Future of Datacentre IP 18/12/2008 10:33:00
CW Live speaks withLin Nease, Director of Emerging Business for HP ProCurve, to discuss the future of networks, including the effect of IP-based storage on datacentres, new capacity requirements generated by the use of 10Gb Ethernet, and how an efficient network design can slash energy and cooling costs, and help enterprises build a "green" image. - +
Computerworld Live Podcast #97: The Future of Enterprise Networking 25/07/2008 09:45:36
This week CW Live chats with Mark Thompson, global sales and marketing manager for HP ProCurve, on the future of the enterprise networking. Mark discusses the trends we can expect to see in the near future and how the right infrastructure can ensure your enterprise network is secure. - +
Computerworld Live Podcast #96: Security at the Edge 11/06/2008 09:22:22
CW Live speaks with Amol Mitra, HP ProCurve Director of Marketing for Asia Pacific and Japan. Today's topic: how enterprises are starting to shift away from simply controlling security via server logins, firewalls and moving to more adaptive security frameworks. - +
Data Management Edition #10: Multi-Petascale Systems 02/05/2008 09:12:33
This week we look at sustainability and the development of multicore technologies to build multi-petascale systems. - +
IT Security Edition #11: How to poison the Storm botnet 01/05/2008 08:51:55
This week CW Live presents a case study on how to poison the notorious Storm botnet . Plus we take a look at Cisco's plans for Ironport.
F-Secure Warns About a Worm Affecting Corporate Networks 2009-01-08 16:42:00+11
Research software developer appoints Susan Dart to new Business Development Director role 2009-01-08 09:08:00+11
Research software developer appoints Susan Dart to new Business Development Director role 2009-01-08 09:08:00+11
Anyware Introduce Two Powerful PCI TV Tuner Cards with S5 Power Up and Windows Media Center Remote 2009-01-07 17:30:00+11
Fortinet Cures Mobile Phone “Curse of Silence/CurseSMS” Attack 2009-01-07 16:30:00+11
Achieving the impossible: Unlimited application scalability
Learn how provide applications with significantly higher throughput and lower latency for data operations while retaining the appropriate levels of data quality with clustered caching. Read on to improve your application scalability now.





