The bulk of HPC purchases has evolved into clusters of 2s
nodes, a configuration which is right in the middle of the
x86 SHV server heartland. Expecting IPF based 2s systems
to beat x86 in performance/price with a process lag is IMO
absurd.
What will happen when both Xeon and IPF are at 32
nm at nearly the same time? I'm pretty damn sure the IPF
chip will easily beat the best x86 chip in absolute FP/HPC
performance but I am also pretty damn sure that the IPF
chip will be priced 2x to 5x higher than Xeon DP because
of its greater system scalability and RAS features.
Why ?
Would Tukwilla beat a Clovertown with IMC/CSI ? I seriously doubt it.
Not only won't IPF beat x86 at the same process size , but all its extra features ( RAS ) would be moved to x86 by the time 32nm time.
Poulson will feature 6-10 cores and a load of L3 at 32nm.Nehalem EX Beckton will have 8 cores , 24MB L3 at 45nm.
I'm 90% sure Nehalem > Poulson in whatever app you want.
Market Data 
Markets 
AI
