IBM’s eX5 Proprietary Memory Gambit

IBM announced three new servers this week under the family name ‘eX5’.  This announcement was IBM’s thinly veiled effort to get in front of Dell and HP in the introduction of the latest generation of technology from Intel, Nehalem EX.  Since all vendors are under an embargo as it relates to the introduction of these processors through March 30, IBM focused on their hardware architecture and never specifically mentions the new Intel chips.

No pricing, performance specifications, or availability was included in the announcement. 

These new servers stay the course that IBM originally plotted for their X Architecture products; proprietary implementations of industry standard technology designed to lock in customers.  IBM proudly proclaims they spent $800 million developing these new servers!  Someone has to pay for that development effort – IBM customers.  Let’s break this down.

Proprietary implementation of Intel’s processors — The X Architecture has always included a proprietary, IBM only, CPU implementation.  IBM claims this is a ‘value add’ and significantly increases the performance of their platforms.  When analyzed this proprietary implementation provides a marginal performance boost at best, hardly offset by the cost of customer lock-in.

Proprietary memory architectures — IBM has created proprietary memory modules that allow the IBM server to host more memory than was intended by Intel.  Their claim is that customers can run larger databases, more VMs and bigger VMs on fewer servers.  Again, these claims aren’t backed up by any specific facts or test results in IBM’s announcement.  We’ve seen this before from IBM and Cisco.  These memory subsystems substantially drive up cost, power and cooling requirements and complexity.  Additionally, straying from the Intel architecture raises long term questions of compatibility, performance and reliability.  Paul McDougall of Information Week highlights some of these concerns in his blog post yesterday “I find that typically if the memory requirements grow, so have the overall system requirements. Even if you could swap the CPUs as well, would you not want new power supplies, faster bus, etc…”  So, more memory (above and beyond what Intel intended in designing the chip) doesn’t necessarily mean better anything.

IBM’s approach is to provide more memory density on these servers but it comes at a reduced overall computing performance. This doesn’t result in a balanced architecture.  We just don’t purchase a server for just memory but overall performance, value, etc. Plus, IBM will need to recoup their $800 million investment which customers will be on the hook for.

What’s Dell’s response?  We will launch industry standard, non-proprietary server platforms based on Intel’s Nehalem EX processors very soon.  Our solutions will help drive down cost and complexity, while delivering industry-leading performance, thermals and availability.  Our adherence to standards remains a critical component of driving down TCO and delivering the Efficient Enterprise to our customers.  Stay tuned for the launch!

About the Author: Andre Fuochi