3 ms·
I was under the impression SIMD is due to clock speed not scaling. Instruction parallelism is hard, so there's a lot to gain from just making instructions wider
by ben-schaaf 3y ago
I was under the impression SIMD is due to clock speed not scaling. Instruction parallelism is hard, so there's a lot to gain from just making instructions wider.
- mafribe 3y agoSIMD is orthogonal to clock-speed. There is indeed a lot to gain from having a single instruction trigger more complex behaviour, for example better instruction density, less instruction decoding needed, but all of this is independent of clock frequency. I think, but am not sure, that the Thinking Machines' CM-2 from 1988 was a 4096 or 8192 wide SIMD machine. Surely, at the time, clockspeeds where low.