Intel Corp (INTC)

[chipguy]

All new microarchitectures have their idiosyncracies and
best code sequences to do things. It takes a while for
assembly language coders and non-Intel programming
toolchain folks to catch on. This was most distinctly felt
for Willamette/netburst but was still important factors for
the 486, Pentium, and P6.

Probably the biggest areas for NGMA related optimizations
are in SIMD/streaming/FP type apps. A 3 GHz NGMA chip
has a peak FP performance (both cores) of 24 DP or 48
SP GFLOP/s. There is no way most current apps are coded
in a way that even begins to tap that potential. Even most
scalar integer code out there is probably far from optimal
for a four issue wide x86 chip with three integer units.