@-D-
A DSP is not the same as a CPU, how it's math results are measured do not come up equal. Let's dig up the AT&T DSP3210 performance specs here for a moment:
The AT&T DSP3210 is capable of 12.5 million instructions per section, 25 million floating point operations per second according to Dave Haynie. Yes folk, that's SIMD for ya. Each instruction does 2 math operations. All of these, 32-bit floating-point. No intermix of 8 or 16-bit here, simplified the design. But real world tests reveal the same DSP getting 16.7 MIPS / 33MFLOPS. The DSP3210 was introduced in 1991, back when 386's were still speed-kings. And a 386 can't even do a single MFLOP.
Now, the 68030 does 1.3 MFLOPS @ 50Mhz. The 68040-25 did about twice that, at 2.4 MFLOPS The Pentium 2 gets 9 MFLOPS @ 450Mhz. So looks like I am right here.
As for audio, the DSP3210 was not limited to audio, nor was it ever intended to be limited to audio. They had an audio chip that was to run alongside it (from Crystal Semiconductor or Analog Devices) for audio.
As well, the DSP3210 continued to be evolved, with 55Mhz, 66Mhz, and even 80Mhz models arriving. But it ended it's life in 1999, along with all of Lucent's 32-bit DSP solutions. Sad really. It could have given the Amiga a real shot in the arm.
