Sun must replace hot air with firm chip detail - Gartner
Need for speeds and feeds
There is a pretty big gap in Sun Microsystems' processor roadmap, and analyst firm Gartner has called on Sun to produce some filling.
In a research note issued last week, Gartner charged that Sun's cancellation of the Gemini and UltraSPARC V processors, first reported here, has conjured up an air of ambiguity around the company's plans. The main cause for concern, according to Gartner, stems from Sun's lengthy dependence of the UltraSPARC IV processor. Customers have been urged to question Sun on exactly what they are supposed to do in 2006/7 when UltraSPARC IV is well past its prime.
"Now that the UltraSPARC V/Millennium project has been eliminated, the current architecture, UltraSPARC IV, will become the placeholder for high-performance systems until Rock appears in late 2006 or 2007," Gartner said. "However, lack of specifications on Rock and the delay before its probable appearance create uncertainty around Sun's high-performance processor road map. Sun will offer UltraSparc IV+ in late 2005 as its high-performance product, expecting its performance advances to last through the release of Rock."
On the low-end, Sun customers appear well taken care of. Sun plans to release the Niagara family of multicore chips about 18 months from now. That means customers can expect to see the processors by the first quarter of 2006, if things progress as planned. The Niagara chips are aimed one to four processor servers, which often handle tasks such as Web or application serving. The Niagara chips should boost Sun's overall server performance on these types of tasks by 15x.
Along with Niagara-based systems, Sun will have a line of Opteron servers, stretching from the workstation on up to an 8-way box. Pucker!
As Gartner points out, however, the midrange to high-end of Sun's server line is haunted by many more questions. The UltraSPARC IV just rolled out and is making its way through Sun's 8, 16, 32, 64 and more than 100 processor systems. Sun typically gives new generations of UltraSPARCs two years to live. By mid-2006, Sun customers will be anxiously looking for an UltraSPARC IV replacement, which is where UltraSPARC V was once meant to step in.
Instead, Sun is now telling customers to wait for the new Rock family of chips. These will likely have fewer but more powerful processor cores than the Niagara chips. The design sounds intriguing, but Sun is not really saying when the chips will arrive or exactly what they will look like. Gartner has picked up on this point.
"Sun's current customers should demand details from Sun regarding the company's product road map, as well as further elaboration on its plans for implementing throughput computing," Gartner said. "They should evaluate Sun's chip multithreading, as manifested by Niagara, as a technology for selected network-facing applications, but should be aware that the product will not necessarily deliver similar improvements for computing-intensive loads, as envisioned for the yet-to-be-released Rock. Customers should not assume Rock will fully substitute for the highly pipelined, out-of-order execution design of Millennium."
We doubt Sun will be able to deliver the Rock processors by 2006. The company has serious problems bringing processors out on time, and it appears reluctant to provide even a vague date for the product at this time. So, expect 2007 to be when the first Rock sightings happen, which would mean Sun has anywhere from a 6 to 12 month gap where it could have serious trouble matching UltraSPARC IV against rival product from IBM and Intel.
To be fair, however, IBM and Intel are not being much more forthcoming with their 2007 products. IBM has been very shy about describing Power6. Our sources believe the chip may only have two processor cores. Similarly, Intel refuses to say exactly how many cores the Tukwila version of Itanium will have, although the best indications point to four. What IBM and Intel have on their side is already strong processor performance. UltraSPARC IV is having trouble keeping up as it is, and by 2006, it will be looking more nultra than ultra, we suspect.
It's safe to assume Sun will play up the strength of Solaris to try and convince users that Rock is worth the wait. Sun easily has the best software story to tell of all the major vendors, allowing customers to run numerous applications on a single server and on a single instance of Solaris. In addition, Sun has been dead focused on boosting its multithreaded software attack for a long while, and this type of code is meant to fly 30 times faster on Rock systems than current UltraSPARC III kit.
While Sun's strategy has its iffy points, the company has managed to impress a number of industry insiders we talked to this week, including some at Sun competitors. The insiders largely agree that Sun will go through a rough spell but that it could be poised for a major comeback in 2007, when its software/hardware pairing looks to be very competitive.
The big question is whether customers will stick with Sun until it can reinvent itself once again. ®
HP rides Hondo to super-sized Itanium servers
Sun's Opteron box defies European power supply standards
Sun slashes Solaris x86 price for big buyers
Sun tops off poor Q3 by axing execs
For Sun chip chief, an empty roadmap is a clean roadmap
Intel to merge Xeon, Itanium chipsets
Sun shelves UltraSPARC V in favor of the great unknown