IBM is talking today about a new concerted effort to drive HPC down into the mid-range by making it easier for non-traditional HPC users to deploy small(ish) clusters. I’m strongly in favor of anything that gets more people using HPC via the million monkeys theory(1).
This is primarily a marketing push, addings SKUs of cluster configurations prebuilt to meet the needs of specific industries and developing marketing relationships that will help its partners support the industry verticals.
But there was some interesting news aimed at crossing pretty big gaps. The first is that IBM is building benchmarking and tuning centers aimed at M$oft’s CCS 2003. These will be key to convincing new users to buy and then getting them up and going once they do buy. This is an important move.
IBM has expanded its relationship with Microsoft to offer customers, software vendors and IBM Business Partners four new benchmarking facilities in Poughkeepsie, N.Y.; Raleigh, N.C.; Beaverton, Ore.; and Montpelier, France. These new centers join a network of global Linux benchmark centers.
The second important step they are taking is that they are using their deep pockets to make systems available for users to test their applications before they buy.
Customers interested in moving to a clustered environment can quickly test-drive their HPC applications in IBM’s Deep Computing Capacity on Demand centers. With access to over 20,000 processors, customers large and small can tap into IBM Systems to help accelerate time to market and improve quality while helping to keep fixed costs to a minimum.
I think this is especially valuable to those new users in the mid range for whom a $500,000 investment in a machine is a really big deal. Now they’ll have a way to base the purchase on more than just hope that they’ll get more performance.
Also note: this effort focuses on getting users into HPC with CCS 2003, bridging them from their familiar workaday environment into HPC. I think this is smart.