Oracle Demos New SPARC T4, New Chip Significant Departure From Predecessors

Oracle is publicly demonstrating its new T4 processor today and is shipping beta test systems to selected partners. The new T4 chip is a major departure from previous designs. Sun's T1 processor, codenamed Niagara and introduced in 2005, rejected a conventional focus on single-thread performance in favor of an aggressively multi-threaded, multi-core approach. CPU clock speeds were purposefully kept low to minimize power consumption.

The Niagara T1, introduced in 2005

The T1, which was introduced in 2005, ran at a maximum of 1.4GHz and offered a maximum of eight cores, with each core capable of handling four threads for a total of 32 threads. The T3, released last year, is clocked at up to 1.67GHz, offers a maximum of 16 cores per physical CPU, and can handle eight threads per core for a maximum of 128 threads. One of the commonalities between the two designs is their simplicity--both architectures are in-order designs built to maximize throughput.

The T3. Fun fact:  Both the T1 and T3 dedicated ~30 percent of the total die to the CPU (not including cache)

The T4 doesn't abandon Niagara's multi-threaded focus, but it's a considerable step back from the T3's implementation. The T4 offers a maximum of eight cores per physical chip and keeps the T3's eight-threads-per-core limitation. The T4 compensates for this decline in maximum theoretical throughput in several ways. First, the T4 is an out-of-order processor with an enhanced branch predictor. Its maximum speed is said to be at least 3GHz, nearly double that of the 1.67GHz T3. Oracle claims the chip's single-threaded performance has been significantly boosted, and expects T4 to deliver a 2x-7x speed increase in single-threaded workloads compared to T3.

The shift makes sense for multiple reasons. Sun's decision to focus nearly exclusively on multi-threaded performance may have made Niagara a uniquely fabulous choice for certain types of workloads, but weak single-threaded performance meant it was increasingly pigeon-holed by its own success. The x86 competitive market has also evolved. In 2005, the T1 offered up to 32 threads per CPU socket at 1.4GHz. Dual-core Opterons didn't start shipping until 2006. Niagara's clock speeds lagged far behind AMD's first dual-core Opterons, but the T1 had a unique angle no x86 chip could match.

Today, both Intel and AMD offer a mixture of 6-12 core products with 16-core iterations on the way from AMD. The threading ratio advantage of Niagara has fallen from 32x at the family's introduction to as low as 8x. Continuing to focus on multi-threaded performance would've only made it more difficult for Oracle to find workloads in which the chip was able to shine.

If the company's redesign pays off, it could make the T4 the most relevant CPU Sun has built in years. Oracle has already adjusted its pricing ratios to reflect what it feels is T4's improved performance. All Oracle software is priced by seat count, with a T3 chip considered to be 0.25 of a seat. (Both Itanium and x86 chips are priced at 1x).  T4 chips will carry a 0.5x ratio, meaning Oracle believes the T4 will be at least 2x as fast as the T3, across the board. The clock speed boosts make this likely--and given Intel's current domination of the entire server industry, renewed competition from Sun would be welcome news.  likely.