3 ms·
They're x86 processors, but rather slow ones. The high FLOPS figures come from the custom vector instruction units, which are 4x as wide as SSE and much more f
by modeless 15y ago
They're x86 processors, but rather slow ones. The high FLOPS figures come from the custom vector instruction units, which are 4x as wide as SSE and much more flexible. You're going to need excellent SIMD support in your language to use this chip.
- palish 15y ago4x as wide as SIMD. Wow. (SIMD is four 32-bit floats wide, i.e. you can multiply four floats AAAA with four other floats BBBB in a single instruction = AB AB AB AB. So 4x that is 16 pairs of floats in a single instruction!)