Aaron Spink (aaronspink@notearthlink.net) on 7/19/10 wrote:
---------------------------

>And these applications do just fine on OOOE processors and >generally run better and faster on OOOE processors.

Sure. But it's a trade-off of silicon and power. An
OoOE core is the Swiss Army Knife of computing - it does
many things very well. But it's a bit big and bulky for
any particular use. And some fairly important functions
can be done cheaper or faster or at lower power using
in-order cores (or many in-order cores). So yeah, once
you've built your PC with some in-order cores for the 3D
graphics pipeline and some in-order cores for video
decoding and some in-order cores for encryption/decryption
and maybe network packet checksumming and maybe an in-order
core for RAID ECC, then for sure, what's left as the
workload for the "real" CPU is stuff that's not very
well-suited for in-order execution.

Really, I'm a huge fan of OoOE cpu's, and the stuff I
write (a big EDA app traversing 10s of GB of highly-
interlinked data structures) would just suck on an
in-order cpu. But that doesn't blind me to the fact
that there's an awful lot of in-order cores lurking in
an awful lot of hardware and doing a lot of useful stuff
very efficiently, in terms of both silicon area and power.
The argument that those gizmos aren't "real" cpu's seems
to be little more than quibbling over terminology.

>Here's another way to think about it, why did CELL fail so badly? Its pretty much
>the ideal: in-order, software controlled data store, etc. Shouldn't it be blowing everything else out of the water?

As far as I can see, CELL didn't do anything well: the
lousy DP in the early versions made it almost useless for
HPC, the need to manage the memory hierarchy explicitly made
it almost useless for complex apps written in HLLs,
and the rapid advances in Nvidia and ATI's more
specialized 3D chips made it uncompetitive for 3D graphics.

If you want to do something new, it has to provide a
compelling advantage over the competition in some
decent-sized niche. There are lots of ways to fail ...