Advanced Micro Devices Inc (AMD)

[neye_eve]

@sgolds, m.moy, comb - SIMD

it seems to be the thinking that SSE.* doesn't do columnar operations, and m.moy has posited that there doesn't seem to be much optimization left in this regard.

So, my question for y'all is, how is it that the AltiVec unit(s) in the G4(5)s do so well at the MD part of SIMD compared to the x86 vectorization units?

I have a friend doing postgrad biochem at an ivy league school and they looked into opteron, xeon, and G5 solutions, and were able to get more than 2x the performance in a lot of their tasks due to the superior nature of the G5 vector units.

Is it just a matter of having started with SSE, and it would be impractical to go to a solution that really works and have all that software to rewrite/recompile?

thanks in advance,
neye