HPC in Enterprise Applications
The statistics don't actually back up Mr. Turkel's claims. According to the recently published Top 500 list highlights, The number of Itanium 2-based systems dropped further from 46 to 36 over the last six months while the number of x64 based systems rose from 136 to 199 (Top500 breaks down x64 into two categories, Intel's EM64T and AMD's Opteron processors). Neither can Ed argue that the Top500 list is only representative of more traditional HPC application segments like academic, research, and military customers. In fact, over 50% of the current Top500 list is classified as "Industry" segment which to me sounds like where one might find a few enterprise applications.
At Sun, we have long recognized that the HPC market is a breeding ground for future enterprise data center technologies. The computer industry is littered with companies now defunct or barely surviving with US bankrupcy protection because they failed to realize this. Remember the Dec Alpha processor? A great HPC CPU, it just never survived because DEC and several subsequent companies were not successful making it a commercial success. Why? Because it failed to become an industry standard. The x64 instruction set is an industry standard, and in fact, as mentioned above, one can purchase x64 processors from multiple CPU vendors. AMD just happens to have a lead today in the scalability dimension, with their Opteron processor supporting the design of up to 16 processor core servers like Sun's recently announced Sun Fire x4600. The x4600 is a great HPC server, in fact it only takes 655 of them to achieve 38.18 TFlops on the Top500 Linpack benchmark, vs the several thousand smaller two socket servers that would be required with a more traditional clustering approach.
The Sun Fire x4600 has one big advantage in running enterprise HPC applications vs the Itanium 2 servers from HP that Turkel talks about. Namely that many more enterprise applications run on the x64 instruction set vs the Itanium instruction set. Hard to do HPC unless you can actually do the "Computing" part. There is another big reason why we will see more HPC in the enterprise that Mr. Turkel failed to grasp, and it is also related to the x4600's general purpose design. That is the merger of HPC and enterprise applications running on the same cluster. Lets say I'm the CIO of an oil company and recently purchased a 10,000 processor core cluster of x4600's. Next week my VP of Public Relations walks into the office and hands me a high priority request for 300 new Windows desktops to handle the rapidly increasing call volume from customers complaining about the recent spike of gas prices in the US. I could go buy a couple dozen more vmware servers to handle my call center desktops, or I could just carve out a few dozen cores from my HPC cluster to handle the peak call periods. That is the beauty of industry standards and general purpose servers, they can run all sorts of different software, from pure HPC, to enterprise HPC, to pure enterprise.
Of course there is one thing better than industry standards, and that is open standards. Maybe that helps to explain why Sun's new UltraSPARC T1 based family of servers was our fastest product line ever to reach $100M in sales as was announced earlier this week in our quarterly earnings call. In case you forgot, the UltraSPARC T1 powered T1000 and T2000 servers are based on one of the world's first open source CPU architectures, OpenSPARC. Of course it doesn't hurt that some of the fastest growing open source software runs on the UltraSPARC T1. Open source software for open source hardware. What a cool idea. And if you are concerned that the UltraSPARC T1 is not focused on HPC performance, just wait for the encore.