Kira (kirsc@aeterna.ru) on 4/20/12 wrote:
>
>What about Power4 through Power6, which were also fairly
>picky about optimization? Were they "broken" uarches?

I think POWER6 was indeed fairly broken, but it damn well
made up for it with high frequencies and ultimately good
performance. So it did the whole P4 thing, but without
quite as much fragility.

But yes, I definitely think POWER7 is the better uarch.

POWER5 I thought was quite reasonable, although iirc it
still had that odd 2-cycle integer latency from POWER4.

In the end, some amount of fragility in performance
is quite acceptable. There are always going to be things
that perform better on one uarch over another. And it is
possible that the benchmarks that phoronix picked are ones
that show up certain unimportant special cases more than
they really should matter.

And if the performance fragility is of the type where you
generally perform quite well, but then have a few cases
where you are truly outstanding, nobody is going to'
complain.

So in the end, performance matters most. If you are the
unquestioned top performer, and the fragility just sometimes
means that it's not as obvious than at other times, I'd
be crazy to complain.

But if it's a matter of "pretty ho-hum performance, but
then sometimes if things line up just right you look quite
good", that's a bad, bad, bad thing. I'd much rather
take something reliably fast, than something that I have
to waste my time tweaking for.

Because then my binary may run really well on one uarch,
but two years down the line I will have wasted all my time
and need to tweak all over again for the next fragile piece
of crap.

Linus