dol@ce.chalmers.se (Fredrik Lundholm) writes:

> In article <86pthz8hh9.fsf@elrond.bloomberg.com>,
> Chris Morgan <cm@mihalis.net> wrote:
> >
> > Sun Apple
> >
> >architecture 64-bit unix 64-bit unix
> >CPU 1X1GHz 2X2GHz
> >cache 1MB on-chip 512k on chip per cpu
> >mem b/width 4.2GB/s 8GB/s per processor
>
> Stop right there. On the Sun, the memory controller is integrated on
> chip which leads shorter latencies compared to a shared off chip
> memory controller on the Apple.
>
> The mem b/with for the Apple system is 6.4GB/s shared for both CPUs.
> When the two CPU:s are competing for the shared controller performance
> will suffer. (as witnessed by the flawed unofficial apple spec-results where
> one CPU is turned off).

I think there is some confusion here, probably mine, the Apple paper
says there is 16-GBps aggregate system bandwidth in dual-cpu
systems. And anyway, we know that Blade 1500 bandwidth is 4.2GB/s
fixed, because it does not support a second cpu, so that's 4.2 vs. 6.4
from your figures.

>
> The Apple and Sun busses are otherwise pretty equal in b/width
> 128bit@200MHz (Sun) vs 32bit@1GHz (Apple)

I make this 25% more bandwidth for the Apple machine.

>
> When CPU:s are added in multiprocessor UltraSPARC IIIi-system you ADD memory
> controllers to the system and therefore increase the available total bandwidth.

You can't pull any multi-cpu US-IIIi workstations into this debate,
because there aren't any (although reading the fine print of some of
the documents that came out today reveals a "Sun Blade 2500").

But ... fine, let's get the facts straight and see what conclusions
can be drawn. Can anyone do a more thorough analysis? Hardware is not
my area.

Would you claim the Sun system performs better than the Apple? How
about bang/buck?

Thanks

Chris
--
Chris Morgan <cm at mihalis.net> http://www.mihalis.net
Temp sig. - Enquire within

Chris Morgan <cm@mihalis.net> writes:

> dol@ce.chalmers.se (Fredrik Lundholm) writes:
>
> > In article <86pthz8hh9.fsf@elrond.bloomberg.com>,
> > Chris Morgan <cm@mihalis.net> wrote:
> > >
> > > Sun Apple
> > >
> > >architecture 64-bit unix 64-bit unix
> > >CPU 1X1GHz 2X2GHz
> > >cache 1MB on-chip 512k on chip per cpu
> > >mem b/width 4.2GB/s 8GB/s per processor
> >
> > Stop right there. On the Sun, the memory controller is integrated on
> > chip which leads shorter latencies compared to a shared off chip
> > memory controller on the Apple.
> >
> > The mem b/with for the Apple system is 6.4GB/s shared for both CPUs.
> > When the two CPU:s are competing for the shared controller performance
> > will suffer. (as witnessed by the flawed unofficial apple spec-results where
> > one CPU is turned off).
>
> I think there is some confusion here, probably mine, the Apple paper
> says there is 16-GBps aggregate system bandwidth in dual-cpu
> systems. And anyway, we know that Blade 1500 bandwidth is 4.2GB/s
> fixed, because it does not support a second cpu, so that's 4.2 vs. 6.4
> from your figures.

It's kind of hard to come up with a single 'bandwith' number
for the G5. The tech overview for the G5 at

http://www.apple.com/powermac/pdf/Po..._TO_072903.pdf

has a useful diagram on page 9, which shows that a G5 has
busses all over the place. So "aggregate bandwidth" isn't a
very useful number.


> > The Apple and Sun busses are otherwise pretty equal in b/width
> > 128bit@200MHz (Sun) vs 32bit@1GHz (Apple)
>
> I make this 25% more bandwidth for the Apple machine.

Except that the IIIi's memory bus is separate from the Jbus
that carries other IO, whereas the G5's processor bus must
serve for both purposes. So I think that either system could
have the advantage in bandwith, depending on where that
bandwith is going.

> > When CPU:s are added in multiprocessor UltraSPARC IIIi-system you ADD memory
> > controllers to the system and therefore increase the available total bandwidth.
>
> You can't pull any multi-cpu US-IIIi workstations into this debate,
> because there aren't any (although reading the fine print of some of
> the documents that came out today reveals a "Sun Blade 2500").
>
> But ... fine, let's get the facts straight and see what conclusions
> can be drawn. Can anyone do a more thorough analysis? Hardware is not
> my area.
>
> Would you claim the Sun system performs better than the Apple? How
> about bang/buck?

I'd claim that it comes a lot closer than the SB150 or SB2000. :-}

The question that even us Sun diehards are asking is,
why didn't this machine ship a year ago? With features
like PC2100 DRAM, no PCI-X, and no FW800, it doesn't
exactly seem like a "fresh" design.

-SEan

On Thu, 18 Sep 2003, Sean Burke wrote:

> The question that even us Sun diehards are asking is,
> why didn't this machine ship a year ago? With features
> like PC2100 DRAM, no PCI-X, and no FW800, it doesn't
> exactly seem like a "fresh" design.

Dunno about the others, but the lack of PCI-X might be
due to Sun's keeping an eye on Infiniband. Why change
HW busses more than you need to?

--
Rich Teer, SCNA, SCSA

President,
Rite Online Inc.

Voice: +1 (250) 979-1638
URL: http://www.rite-online.net

On 17 Sep 2003 17:58:17 -0400
Chris Morgan <cm@mihalis.net> wrote:

> hoh@invalid.invalid (Goran Larsson) writes:
>
> > For many users it is important. I can not understand why Apple
> > ignored error correction. I can accept the lack of ECC in my iBook,
> > but for my servers (SPARC/Solaris) and workstation (SPARC/Solaris)
> > that has to be up 24/7 the lack of ECC is completely unacceptable. I
> > gladly trade CPU speed for stability.
>
> I don't see why workstations need to be up 24/7. To me that's what
> servers are for. Anyway, you are well within your rights to insist on
> ECC, but most of my users choose a PC without ECC over a SPARC
> workstation or other machine with that protection more than 99.9% of
> the time.

The first argument is that PC are VERY CHEAPER and are very stable under
linux.
--
@+ Yann Marigo
Le mirroir officiel de obsolyte.com : http://obsolyte.oldmachines.org

> The first argument is that PC are VERY CHEAPER and are very stable under
> linux.

They are?

That's news to me.

Cheap as long as your time is worthless.

--
Steven Hill

``Video barbam et pallium; philosophum nondum video''

Dans article <Pine.LNX.4.44.0309181543470.2424-
100000@kerouac.projectcolo.org.uk>, sjh@waroffice.net disait...
>
> Cheap as long as your time is worthless.
>

Old legend. Stupid windows newbies complain hoooow Linux is difficult and
arcane. However, for a skilled unix administrator it's a breeze. Current
distributions install in less than an hour and just work. They recognize
out of the box SCSI cards, Fibre Channel cards, Giga Ethernet cards, RAID
cards and autoconfigure everything. Linux is WAY EASIER to install than
Solaris, IRIX or whatver Unix you like.
Another reason for that old legend is RedHat (and rpm). RPM sucks rocks
badly, and RedHat sucks even more. Unfortunately, for most Linux==Redhat,
which is bad. However, Mandrake, Debian and others made nice tools to
resolve automatically dependancies between application packages and it
all goes fine without pain nor headache.

--
Quis, quid, ubi, quibus auxiliis, cur, quomodo, quando?

Yann Marigo <yann.marigoNOFUCKINSPAM@laposte.net> writes:

> The first argument is that PC are VERY CHEAPER and are very stable
> under linux.

Unless they decide to change the VM system in the middle of a
"stable" kernel series.

Now, FreeBSD on the other hand... ;>

--
David Magda <dmagda at ee.ryerson.ca>, http://www.magda.ca/
Because the innovator has for enemies all those who have done well under
the old conditions, and lukewarm defenders in those who may do well
under the new. -- Niccolo Machiavelli, _The Prince_, Chapter VI

> distributions install in less than an hour and just work.

Unless you have anything non standard or new.

> They recognize out of the box SCSI cards, Fibre Channel cards, Giga
> Ethernet cards, RAID cards and autoconfigure everything. Linux is WAY
> EASIER to install than Solaris, IRIX or whatver Unix you like. Another
> reason for that old legend is RedHat (and rpm).

Yer arse.

I like predicatbility and reliability. Call me old fashioned. You do not
yet get this from Linux. You do from the well known and supported hardware
of Sun, Hewlett Packard and IBM.

--
Steven Hill

This is not the signature that you are looking for

Dans article <Pine.LNX.4.44.0309182233110.10042-
100000@kerouac.projectcolo.org.uk>, sjh@waroffice.net disait...
>
> Unless you have anything non standard or new.
>

Non standard and new is for the PC gamerz crowd. For servers,
workstations, i e serious job everybody use validated, serious hardware.

> I like predicatbility and reliability. Call me old fashioned. You do not
> yet get this from Linux.

I do, every day.

> You do from the well known and supported hardware
> of Sun, Hewlett Packard and IBM.
>

Sure, for a hefty price. BTW, E12000 PCI sucks wind. We've got better
performance with PC (high end grade) hardware. That's the sad truth.

--
Quis, quid, ubi, quibus auxiliis, cur, quomodo, quando?

"Steven Hill" <sjh@waroffice.net> wrote in message
news:Pine.LNX.4.44.0309181543470.2424-100000@kerouac.projectcolo.org.uk...
> > The first argument is that PC are VERY CHEAPER and are very stable under
> > linux.
>
> They are?
>
> That's news to me.
>
> Cheap as long as your time is worthless.

I've been running petroleum related technical computing applications
on a 3Ghz HP XW6000 running RedHat 7.3 for the last three
months.

It is indeed stable. I haven't had an OS or X Server crash since it was
installed. To be honest I didn't expect the graphics drivers to be
bullet proof, but in practice the NVidia drivers for the Quadro4 cards
have been a lot less troublesome than our prior experiences with the
Expert3D cards from Sun.

As far as benchmarks go, running 2D gridding operations in one
or our geological mapping applications is about 3 times faster on the
HP box under Linux than a 900Mhz Blade 2000 running Solaris
2.8.

-cjs