IBM’s Watson-Jeopardy challenge proved to be entertaining theater but it left hanging the question of how a business could capitalize on Watson technology. IBM almost immediately began suggesting workloads that would apply in healthcare, finance, and elsewhere. BottmlineIT discussed some of these back in early March.
The scale of the hardware Watson required for Jeopardy, however, went beyond what most businesses could or would acquire. Not many businesses are likely to configure 90 tightly integrated IBM Power 750 servers containing 2880 POWER7 processor cores as well as the 15TB of onboard memory for a single workload as was the configuration for Watson when it won Jeopardy.
This week we got the answer. IBM introduced new optimized POWER7 products, both upgraded servers and new blades, capable of running Watson-like workloads. These products, according to IBM, actually provide a performance kick beyond the Power 750 used with Watson.
To do what Watson did—process complex natural language queries and come up with the right answer extremely fast—is not a job for wimpy servers. It still requires high end servers, not commodity x86 Wintel machines.
IBM leads the high end UNIX server market, growing revenue in that segment by 12%, according to IDC,. The researcher’s overall discussion of the Q4 2010 worldwide server market is a little more nuanced vis-à-vis HP than IBM presents but IDC still declares IBM the leader in the worldwide server systems market with 37.4% market share in factory revenue for 4Q10.
But the real questions revolve around when and how commercial businesses are going to deploy Watson technology. Along with announcing the new POWER7 servers, IBM introduced an early adopter tapping the new POWER7 technology, if not exactly the Watson capabilities.
That business, RPM Technologies, provides wealth management software to some of the largest banks and financial services companies in Canada. In terms of the new POWER7 technology, “POWER7 chips along with AIX 6.1 provided a big boost to the batch and threading speed of our products,” said RPM’s chief architect. With POWER7 chips, batch job runtimes improved by upwards of 35% and used fewer resources, he reported. As part of the upgrade, RPM also moved to a fully virtualized environment across two POWER7 16-core P750 machines, which reduced the time and effort to manage the boxes.
Another early adopter, the University of Massachusetts-Dartmouth, may be more on track to tap Watson-like capabilities. The school’s researchers are using two IBM POWER7 blades to study the effect of cosmic disturbances, called gravitational waves, on black holes in space.
“We are running billions of intense calculations on the POWER7 blades… able to get results as much as eight times faster than running the same calculations on an Intel Xeon processor. Calculations that used to take a month to run are now finished in less than a week”, reported Gaurav Khanna, professor of physics at UMass-Dartmouth. Not fast enough to win Jeopardy but impressive nonetheless.
The new POWER7 products include the following enhancements:
- Enhanced IBM Power 750 Express—the same system that powers Watson—further optimized with a faster POWER7 processor delivering more than three times the performance of comparable 32-core servers.
- 16-core, single-wide IBM BladeCenter PS703 and 32-core, double-wide IBM BladeCenter PS704 blade servers, which provide an alternative to sprawling racks.
- Enhanced IBM Power 755, a high-performance computing cluster node with 32 POWER7 cores and a faster processor.
Along with the servers, IBM announced new switches closely integrated with its Power servers to support workloads such as cloud computing, financial services, Web 2.0, streaming video, medical and scientific research, and business analytics. According to a recent report by The Tolly Group, the new IBM switches demonstrated an average of 55% better price/performance over comparable switches.
So, everyone is still waiting for a business user that actually is tapping Watson-like capabilities to address a business problem. It will happen. As you know, it takes time to get systems implemented, tested, and put into production. Stay tuned.