Sun Microsystems has excised the UltraSPARC VI processor from its roadmap, deciding to go instead with the "Rock."
Sun is looking for the freshly announced "Rock" family of chips to combine the best features of its standard UltraSPARC processor line and its low-end multicore processor line. The Rock processors will be stacked with numerous processor cores and be able to perform application specific functions. Need a Rock to handle packet processing, compute processing or encryption? They're all there.
"This gives us the best of both worlds," said David Yen, Sun's processor chief, at a conference here today.
Sun will continue to advance its UltraSPARC IV and V processors along with the multicore Gemini and Niagara chips. The standard UltraSPARC products tend to have better overall performance and are designed to crunch single software threads. By contrast, the Niagara line runs at a lower frequency but spreads software threads across numerous cores. The standard chips are typically used for databases and other business software, while the Niagara line is aimed at Web and application servers.
When the Rock family arrives some day far in the future, Sun hopes to find a pleasant middle ground. Details are fairly thin at this point, but Yen did say the Rock processors could be tuned to handle specific functions. Sun would ship compute chips along with ones designed to run Java, TCP/IP traffic or security functions.
Why take this approach?
Well, first off, Sun seems to be making a pretty clear business decision. It often takes grief for continuing with UltraSPARC production given the heavy research and development cost this requires. When Sun first showed two, divergent processor lines, analysts shuddered even more.
With this approach Sun can spread its investment along a wider array of similar products.
Also, don't forger those Fujitsu rumors from last year. Sun could very well team with Fujitsu on a fast, single thread chip, if need be. Sun might chip in some money to create a variant of Fujitsu's SPARC64 processor but would not have to do the total design. We hear such a scenario may be announced sooner rather than later.
Secondly, Sun thinks this strategy fits in well with where "the network" is heading. Millions upon millions of cell phones, thin clients and RFID tag readers will be feeding massive amounts of data to the network, Sun argues. This will require processors able to handle a variety of small tasks well.
Sun thinks both Intel and IBM will be limited in the way they can respond to these new workloads. Intel because hyperthreading will not allow different types of software threads to be split up on one processor core, and IBM because its Power chips do not have enough cores for specialization.
Analysts, however, warn that moving away from the general purpose realm is a risky approach. IBM tried in the past to tune its chips for databases and other software, but often a chip is left idle and cannot be used for other tasks.
In any case, this answers a nagging question about how Sun would continue on with UltraSPARC and whether it was worth the effort. At least the company has a plan, and, as always, it's different from rivals. ®