Sun has started shipping some systems based on its UltraSparc T1 processor (see the news story) with claims of "blowing away" industry standards for performance, setting seven world record benchmarks and delivering a five-fold performance increase at one-fifth the power consumer versus Dell, HP and IBM systems. The 'T' in T1 technically stands for 'Throughput,' but from its marketing stance, Sun might as well call it the 'Terminator.'
"Sun has once again leapfrogged the competition, establishing a five year lead over any other processor architecture in the world," Jonathan Schwartz, Sun president and COO, beams in the press release.
Now that Sun's new eco-friendly T1 server systems are in play, we can determine whether the T1 is really a Terminator or a harbinger of Sun's exit from the processor business. Rivals will be forced to respond to Sun's bold claims, and will start with more aggressive pricing and different intepretations of Sun's benchmarks. For example, Stephen Shankland's news story reports that IBM isn't convinced that the T1 will gain traction beyond a small low-end market segment, and that HP launched a Niagara (UltraSparc T1) "bashing" Web site. The HP site mentions "disappointing" single-thread performance and asserts that a new software development paradigm is required to optimize software for the T1. Sun is shipping new software development tools for optimizing performance on the 32-thread T1. It boils down to a simple cost benefit issue, not a paradigm shift.
Sun created a three-dimensional rack server benchmark, SWaP (space, wattage and performance), to measure the efficiency the multithreaded, multicore processor systems. The formula--SWaP = performance / (space x power)--can be used to compare systems from various vendors.
Space: The space a server occupies can be measured by the rack unit height of the system; information found on a company's web site.
Wattage: Metrics on the server's power consumption can be obtained using data from actual benchmark runs or vendor site planning guides, recorded in watts. Estimated system power consumption from calculators or datasheets can be also used to assess wattage.
Performance: Information about the level of performance or throughput a server maintains can come from any industry-standard or ISV benchmark.
Of course, in Sun's benchmarking the T1 systems come way out on top compared to IBM's Power 5 and 5+ and Intel's Itanium 2 and Xeon EM64T.
Based on the benchmark claims, and comments from a handful of initial customers, the T1 sounds like a Terminator for a broad class of servers. In Sun's press release, Larry Lorzon, vice-president of datacenter services at EDS, said initial testing showed a 50-percent decrease in power consumption, and that the T1 systems would "lay the foundation for a tech refresh strategy focused on improving application performance while reducing space and power consumption." Sun also guarantees binary compatibility on the Solaris on the T1 Sun Fire T1000 and T2000 servers, although some applications would have to be tweaked to take full advantage of all the parallelism.
Initial customers include eBay, EDS, NEC, NTT Data and Air France. It's surprising that Google isn't on board given all the recent lovefest with Sun? Maybe Google is happy with its flavor of Linux and thousands of low-cost, x86-based racks. In that case, Sun would like to unload its AMD Opteron systems in Google's datacenters.
Sun is sweetening the pot, offering a 90-day test drive of the new systems. As I've said before, even if the claims about the processor architecture lead and benchmarking results don't fully pan out, it's hard to see how Sun doesn't get significant wins in 2006. If a strong business for UltraSparc T1 systems fails to materialize, then something is very wrong within Sun or the competition has something up its sleeve we don't know about yet...