Kira (kirsc.delete@this.aeterna.ru) on August 5, 2012 5:45 pm wrote:
> Michael S (already5chosen.delete@this.yahoo.com) on August 5, 2012 2:55 pm
> wrote:
> > Kira (kirsc.delete@this.aeterna.ru) on August 5, 2012 2:26 pm
> wrote:
> > > mpx
> > (mpx.delete@this.nomail.pl) on August 1, 2012 11:19
> pm wrote:
> > > >
> > >
> > defderdar
> (derderdar.delete@this.mailinator.com) on August 1, 2012 2:52 pm
> >
> >
> >
> > > > wrote:
> > > > > won't Oracle try to cripple
> some
> > performance on some
> > > level
> > > >
> > >
> >
> > > > Itanium
> > does it by itself, there's no need for
>
> > > external intervention.
> > > >
> >
> > > >
>
> > > >
> > >
> > > So if Itanium support should be
> >
> >
> > killed because of its supposedly "crippling" performance, why on
> earth should
> >
> > > Oracle keep supporting the (far slower) M-series
> SPARC?
> >
> > I'd expect
> > fully loaded M9000 (64-socket SPARC64
> VII+) to be approximately twice faster
> > than fully loaded Superdome2
> (32-socket Tukwila) in majority of commercial
> > applications.
> > Do you
> have a hard data that could contradict my
> > expectations?
> >
>
> SPEC
> indicates that Tukwila is significantly faster than S64 VII (albeit the slightly
> older 2.66GHz
> model):
>
> http://www.spec.org/cpu2006/results/res2010q4/cpu2006-20101206-13883.
> html
> http://www.spec.org/cpu2006/results/res2010q2/cpu2006-20100426-10756.html
>
>

And TCP-H @1000GB indicates that SPARC64 VII+ is significantly faster than Tukwila:
http://www.tpc.org/tpch/results/tpch_result_detail.asp?id=111060301
http://www.tpc.org/tpch/results/tpch_result_detail.asp?id=110042601

There is simply not enough of public data points to decisively conclude which of two chips is faster socket for socket. But, IMO, even the little data we have is sufficient to conclude that throughput-wise 32 Tukwila sockets can't match 64 SPARC64 VII+ sockets.

> There's also the fact that it's using DDR3, rather than S64's DDR2, and
> doesn't seem to be depending on a shared bus, which S64 VII seems to still be
> doing. I would be extremely surprised if a 64s M9K came out twice as fast as a
> 32s SD2 in typical commercial applications.

I also would be extremely surprised if M9k is twice, or even 1.5 times faster *over the board*. After all, most applications don't scale linearly with # of sockets.
However, I would be even more surprised if HP manages to find just one popular commercial throughput benchmark in which (Tukwila-based) SD32 is faster than top M9K.