IBM just realized that Power has a power problem

In this interview, Frank Soltis, a key person in designing the IBM Power processor, identified heat as a major problem for the Power. Followed the footsteps of AMD Opteron, IBM improved Power4 and added embedded memory controller to Power5, with impressive results. However, last time I checked, the Power5 generates over 150 watts of heat, in the same league as INTEL's Itanium and Xeons, much hotter than 72 watts for the 8-core UltraSparc T1 and 95 watts for the dual core Opteron 880.

With perfect hindsight, Mr. Soltis pointed out that both DEC Alpha and INTEL Itanium died or will die because of lack of backward compatibility with existing software.

Looking forward 10 years, I think there will be three major architectures left: AMD64, Power and Sparc.

AMD64 with its coming 32P glueless ccNUMA capability and multi-core DCA architecture will rule the high performance server, desktop, mobile and handheld markets.

Power will find its survival in game consoles. It will be out of the mainstream enterprise server market because Opteron will be more powerful and SUN will not port Solaris 10 over to Power. AIX is dead. Linux will not move fast enough to halt Solaris 10.

SUN will continue to shine with a Solaris 10 Sparc.

But SUN needs to focus on Solaris 10, Sparc and Opteron. There is one thing SUN can add to kill all Linux: a Solaris 10 GUI control panel that is as easy as Windows server 2003.

SUN should stop wasting money on application servers and other non profitable software. In the software industry, it's a winner take all game, the best makes 90% of the money. So if you can't be the best, you quit. BEA is very good, even JBOSS can do a better job than SUN, so why bother with a SUN J2EE server? SUN only needs some top guys writing J2EE standards and let others do it. If SUN can bundle Oracle, it can also bundle WebLogic, at a discount. Other stuff such as free SUN studio tools should also be cut, why not let Borland do it? If there is money to be made there, it's much cheaper just to buy a stake in Borland. It might be painful to reduce some workforce, but it's for the greater good of the network computing era SUN hopes to define.

I got a feeling SUN is still quite wasteful in using its money. I heard a lot of PMs just sit in their remote homes and forward emails. If you go to Microsoft, there is no such waste of resources.