On December 16th Analog Devices announced a new member of its high-performance fixed-/floating-point TigerSHARC family, the ADSP-TS101S. The chip runs at 300 MHz in a 0.13-micron process and is already in full production at this speed, according to ADI. The ADSP-TS101S is intended for high-performance multiprocessor applications, including telecommunications infrastructure, medical imaging, industrial instrumentation, and military electronics.
BDTI has not yet benchmarked TigerSHARC, but based on analysis of its architecture, it is likely that at 300 MHz the ADSP-TS101S is the fastest floating-point DSP available. The only real competitor among floating-point DSPs is Texas Instruments’ TMS320C67xx, which is currently in production at 167 MHz and sampling at 225 MHz. At 300 MHz, TigerSHARC’s fixed-point DSP performance is likely to be in the same league as that of TI’s flagship fixed-point architecture, the 600 MHz TMS320C64xx.
TigerSHARC’s performance doesn’t come cheap, though; the 300 MHz ADSP-TS101S is priced at $199 in 10,000-unit quantities, making it one of the most expensive DSP processors on the market. ADI believes that the chip’s hefty price tag is justified by its large on-chip memory and specialized I/O support for multiprocessor systems.
Announcing a chip that’s already available at its target clock speed is a significant (and welcome) change from ADI’s earlier policy of announcing chips far in advance of their availability. Indeed, when ADI first introduced TigerSHARC in 1998, the company announced its clock speed as 250 MHz—a speed that ADI didn’t deliver until early 2002 (see /InsideDSP/2002/05/15/Adi).
In the years that have elapsed since TigerSHARC was first introduced, the long-term outlook for high-performance floating-point DSPs has become somewhat murky. High-performance DSP-enhanced general-purpose processors—most prominently the PowerPC 74xx family—have begun to encroach on this market. It is notable that TI has not yet announced a floating-point counterpart to its top-of-the-line, fixed-point ’C64xx architecture, as the company did a few years back when it announced the floating-point variant (’C67xx) of its ’C62xx architecture. It will be interesting to see how TigerSHARC competes against the MPC74xx, and whether TI ups the ante with a new floating-point DSP architecture of its own.
Add new comment