The x86 fashion statement

What Sun isn't going to be hyping, however, is that even Niagara Ibeats Sun's own AMD x86 line on measures of cost/performance outsidethe floating point intensive arena.

As I've said repeatedly, I believe that the next few years will see more change in IT than anything we've seen before with IBM building an ecosystem around Linux on Cell, Microsoft moving to network computing on PowerPC, and Sun leading the charge to next generation Unix with CMT/SMP on SPARC.

Of these, Microsoft and Sun appear to be first in the field although Microsoft's nexen X360 will be limited to the gaming and home entertainment segments for a few more years because they have neither the software nor the market preparation in place to move faster.

In contrast Sun has Solaris 10 firmly established in the market and is expected to announce commercial availability for their first CMT products - the "Erie" and "Ontario" rack mount servers- early next week.

I had expected these to top out at 1.4Ghz, but apparently I was wrong about this with the high end coming in at 1.2Ghz. On the other hand, there are some eye-popping performance numbers coming - things like five times the SAP transaction completion rate over Dell/Xeon and new world records for the 100GB and 300GB TPC/C business intelligence processing tests.

What Sun isn't going to be hyping, however, is that even Niagara I beats Sun's own AMD x86 line on measures of cost/performance outside the floating point intensive arena.

Compare Sun's "becky boxes" with AMD CPUs to Xeon and they look pretty good - offering significantly better performance at lower total cost. Compare them, however, against those same boxes with Niagara CPUs, and what you see is that the AMDs are faster on single threaded processes, faster on floating point, and capable of running Microsoft's operating systems and applications, but cost more to build, more to sell, and more to use.

Since CMT CPUs will gain both floating point and single thread execution speed much faster than the AMD CPus will what we're seeing here is the beginning of a losing race for x86 - and with Intel nowhere in the overall competition this reduces new investment in x86 to little more than either a continuation of things past or a fashion statement