SEO

Hi all,

I did a trial build of a bunch of software I look after (mostly C++)
on a new Sun Blade 2500 with two 1.28 GHz cpus, and for fun compared
the elapsed time to an older Blade 1000 with dual 750MHz cpus.

Here are the result :-

SB2500 SB1000

real 8m47.836s 15m22.264s
user 16m15.640s 26m38.170s
sys 0m55.280s 1m34.190s


Yow! I like it.

Both machines used Forte 6 update 2 and have 2 GB RAM. gmake was
used to drive, with the -j3 flag to ensure the CPUs are kept busy at
all times. These compiles are very cpu bound, I don't think much
else really affects the readings more than a few % (linking is a few
seconds for each of a handful of shared libs, which are all that
gets built).

Now, the disclaimers :-

1. I'm not an expert benchmark engineer
2. The older machine was doing other stuff, but nothing intensive.
3. The newer machine is at a higher patchlevel
4. The newer machine was only booted up on Friday, the older machine
three weeks ago

Even so, I think I can say the new machine is (by the standards of
Sun SPARC Solaris machines that I am personally familiar with)
"fast"

Chris

--
Chris Morgan
"Post posting of policy changes by the boss will result in
real rule revisions that are irreversible"

- anonymous correspondent

Remember the SB1000 a totally differnt animal than the SB2500, the sun blade
1000 is FCAL based with an ultra sparc IIICu processor with 8mb external
cache, compared to the blade 2500 which is IDE based with IIIi and 2mb
external cache so depending on the applications running on each of these
machines your benchmarks will vary.

Most likely forte is optomized to used the CPU's external cache which I
think is the biggest bottleneck on the SB2500.


"Chris Morgan" <cm@mihalis.net> wrote in message
news:867jyvlwtr.fsf@elrond.bloomberg.com...
>
> Hi all,
>
> I did a trial build of a bunch of software I look after (mostly C++)
> on a new Sun Blade 2500 with two 1.28 GHz cpus, and for fun compared
> the elapsed time to an older Blade 1000 with dual 750MHz cpus.
>
> Here are the result :-
>
> SB2500 SB1000
>
> real 8m47.836s 15m22.264s
> user 16m15.640s 26m38.170s
> sys 0m55.280s 1m34.190s
>
>
> Yow! I like it.
>
> Both machines used Forte 6 update 2 and have 2 GB RAM. gmake was
> used to drive, with the -j3 flag to ensure the CPUs are kept busy at
> all times. These compiles are very cpu bound, I don't think much
> else really affects the readings more than a few % (linking is a few
> seconds for each of a handful of shared libs, which are all that
> gets built).
>
> Now, the disclaimers :-
>
> 1. I'm not an expert benchmark engineer
> 2. The older machine was doing other stuff, but nothing intensive.
> 3. The newer machine is at a higher patchlevel
> 4. The newer machine was only booted up on Friday, the older machine
> three weeks ago
>
> Even so, I think I can say the new machine is (by the standards of
> Sun SPARC Solaris machines that I am personally familiar with)
> "fast"
>
> Chris
>
> --
> Chris Morgan
> "Post posting of policy changes by the boss will result in
> real rule revisions that are irreversible"
>
> - anonymous correspondent

Actually ignore my other post I read this benchmark wrong

One last hint also not only is the IIIi clocked at a higher rate but its
using a faster cache 2MB L2 vs 8MB L3 on the IIICu


"Chris Morgan" <cm@mihalis.net> wrote in message
news:867jyvlwtr.fsf@elrond.bloomberg.com...
>
> Hi all,
>
> I did a trial build of a bunch of software I look after (mostly C++)
> on a new Sun Blade 2500 with two 1.28 GHz cpus, and for fun compared
> the elapsed time to an older Blade 1000 with dual 750MHz cpus.
>
> Here are the result :-
>
> SB2500 SB1000
>
> real 8m47.836s 15m22.264s
> user 16m15.640s 26m38.170s
> sys 0m55.280s 1m34.190s
>
>
> Yow! I like it.
>
> Both machines used Forte 6 update 2 and have 2 GB RAM. gmake was
> used to drive, with the -j3 flag to ensure the CPUs are kept busy at
> all times. These compiles are very cpu bound, I don't think much
> else really affects the readings more than a few % (linking is a few
> seconds for each of a handful of shared libs, which are all that
> gets built).
>
> Now, the disclaimers :-
>
> 1. I'm not an expert benchmark engineer
> 2. The older machine was doing other stuff, but nothing intensive.
> 3. The newer machine is at a higher patchlevel
> 4. The newer machine was only booted up on Friday, the older machine
> three weeks ago
>
> Even so, I think I can say the new machine is (by the standards of
> Sun SPARC Solaris machines that I am personally familiar with)
> "fast"
>
> Chris
>
> --
> Chris Morgan
> "Post posting of policy changes by the boss will result in
> real rule revisions that are irreversible"
>
> - anonymous correspondent

On Fri, 13 Feb 2004, Rodrick Brown wrote:

>
> Remember the SB1000 a totally differnt animal than the SB2500, the sun blade
> 1000 is FCAL based with an ultra sparc IIICu processor with 8mb external
> cache, compared to the blade 2500 which is IDE based with IIIi and 2mb

The SB 2500 does NOT use IDE hard drives; the SB 1500 does though.

--
Rich Teer, SCNA, SCSA

President,
Rite Online Inc.

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

On 9 Feb 2004, Chris Morgan wrote:

>
> Hi all,
>
> I did a trial build of a bunch of software I look after (mostly C++)
> on a new Sun Blade 2500 with two 1.28 GHz cpus, and for fun compared
> the elapsed time to an older Blade 1000 with dual 750MHz cpus.
>
> Here are the result :-
>
> SB2500 SB1000
>
> real 8m47.836s 15m22.264s
> user 16m15.640s 26m38.170s
> sys 0m55.280s 1m34.190s
>
>
> Yow! I like it.
>
> Both machines used Forte 6 update 2 and have 2 GB RAM. gmake was
> used to drive, with the -j3 flag to ensure the CPUs are kept busy at
> all times. These compiles are very cpu bound, I don't think much
> else really affects the readings more than a few % (linking is a few
> seconds for each of a handful of shared libs, which are all that
> gets built).
>
snip
>
> Even so, I think I can say the new machine is (by the standards of
> Sun SPARC Solaris machines that I am personally familiar with)
> "fast"

Nice... Thanks for posting the information.

Now for the fun part - Sun has announced the US-IV that "should" drop into
the SB2000 - the 1.2 GHz version of that part should substantially
outperform your 750 MHz SB1000 (and possibly breathe a bit of life into
the SB2000). If that wasn't enough, the US-IIIi+ (US-IIIi done on the 90
nm process) has 4 MB of on chip cache! This is supposed to have twice the
performance of the current US-IIIi.

I also have a question for you (asking for your best guess and not
anything that might be considered proprietary) - with Sun soon to support
the full AMD-64 ISA with Solaris (and presumably with Sun's C compiler) -
do you think that will have a positive impact on application avialability
for Solaris??

I'm guessing that porting Solaris to AMD-64 may end up helping Sparc sales
more than hurting - figuring that the more boxes sold that run some form
of 64 bit Solaris will encourage more people to write Solaris
applications.

Erik

Rodrick Brown <rbrown@[remove]doitt.nyc.gov> wrote:
> Actually ignore my other post I read this benchmark wrong
>
> One last hint also not only is the IIIi clocked at a higher rate but its
> using a faster cache 2MB L2 vs 8MB L3 on the IIICu

The Cache on the US-III isn't an L3 cache, it's L2.
The difference is that on the IIIi the cache runs at the same speed as the
CPU, whereas on the III it runs at half the speed. Which is better here
will depend on the workload - bigger but slower on the US-III against
smaller but faster on the US-IIIi.

Scott

"Rodrick Brown" <rbrown@[remove]doitt.nyc.gov> writes:

> Remember the SB1000 a totally differnt animal than the SB2500, the sun blade
> 1000 is FCAL based with an ultra sparc IIICu processor with 8mb external
> cache, compared to the blade 2500 which is IDE based with IIIi and 2mb
> external cache so depending on the applications running on each of these
> machines your benchmarks will vary.

Hmmm, my blade 1000 is not Cu, my SB2500 is not IDE based, and IIRC
the _on-chip_ cache is 1MB with no external. I do agree that they are
very different beasts though.

Chris
--
Chris Morgan
"Post posting of policy changes by the boss will result in
real rule revisions that are irreversible"

- anonymous correspondent

Erik Magnuson <erik@cts.com> writes:

> Nice... Thanks for posting the information.

My pleasure, I hope to post some more benchmark results. I might get
forte 8 soon.

>
> Now for the fun part - Sun has announced the US-IV that "should" drop into
> the SB2000 - the 1.2 GHz version of that part should substantially
> outperform your 750 MHz SB1000 (and possibly breathe a bit of life into
> the SB2000). If that wasn't enough, the US-IIIi+ (US-IIIi done on the 90
> nm process) has 4 MB of on chip cache! This is supposed to have twice the
> performance of the current US-IIIi.

probably not supported in SB1000 even if I could justify the expenditure

>
> I also have a question for you (asking for your best guess and not
> anything that might be considered proprietary) - with Sun soon to support
> the full AMD-64 ISA with Solaris (and presumably with Sun's C compiler) -
> do you think that will have a positive impact on application avialability
> for Solaris??
>
> I'm guessing that porting Solaris to AMD-64 may end up helping Sparc sales
> more than hurting - figuring that the more boxes sold that run some form
> of 64 bit Solaris will encourage more people to write Solaris
> applications.

I think that there are grounds for optimism for all of Opteron, SPARC
and Solaris. That's why I own shares in Sun. My personal involvement
centers on the SPARC desktop market which is in a pretty bad way, but
for the server market I am greatly enjoying the current Itanium
"discomfiture". I truly think AMD64 has potential as a "disruptive
technology". It is capable of teaching Intel the "Intel lesson" (aka
"it's the software stupid").

Chris
--
Chris Morgan
"Post posting of policy changes by the boss will result in
real rule revisions that are irreversible"

- anonymous correspondent

On 13 Feb 2004, Chris Morgan wrote:

> Erik Magnuson <erik@cts.com> writes:
>
> > Nice... Thanks for posting the information.
>
> My pleasure, I hope to post some more benchmark results. I might get
> forte 8 soon.

Will look forward to seeing them. I'd also be curious as to how well the
power saving via frequency switching works on the SB-2500. IIRC, the
US-IIIi is supposed to switch between full, 1/2 and 1/32 speed.

-snip- reference to US-IV in SB2000
>
> probably not supported in SB1000 even if I could justify the expenditure

I'm not sure that putting US-IV modules in an SB1000 would make sense
either. I do wonder if Sun is keeping the SB2000 around so that it can be
upgraded with the US-IV (would make it a lot more competitive).

Another tidbit, Sun has dropped the price on V210 and V240's with dual
1 GHz US-IIIi's...

>
> I think that there are grounds for optimism for all of Opteron, SPARC
> and Solaris. That's why I own shares in Sun. My personal involvement
> centers on the SPARC desktop market which is in a pretty bad way, but
> for the server market I am greatly enjoying the current Itanium
> "discomfiture". I truly think AMD64 has potential as a "disruptive
> technology". It is capable of teaching Intel the "Intel lesson" (aka
> "it's the software stupid").

I think it would be neat if Solaris on the Opteron put new life in the
Solaris desktop and have that reflected in the Sparc desktop - lot of
little things can help, e.g. better USB support.

The was a letter on either El Reg or the Inquirer about the importance of
HP's PA-RISC 8800. The correspondent stated that not much HP-UX software
has been ported to the Itanium (and I wonder how many ISV's are still
interested in doing an Itanium port). Doesn't sound good for the future of
HP-UX (but does sound good for Solaris).

2004 should be a very interesting year in the computer industry -
especially for Intel. They're getting whacked by AMD - then they have
IBM's G5 that performs as well as their top end P4 at one quarter of the
power (and indications that Intel's strained silicon process is inferior
to IBM's process).

Wished I had followed through on an urge to buy Sun stock a year ago...

Erik

On Sat, 14 Feb 2004 07:52:02 GMT Erik Magnuson <erik@cts.com> wrote:
> Wished I had followed through on an urge to buy Sun stock a year ago...

I'm not so sure. Sun will have problems executing. There's many a
slip twixt cup and lip.

Consider that the clunky and unreliable Linux has stolen a lot of
Sun's mindshare (IMHO worse than the actual dollars they've already
given up in the low end market). That should never have happened and
personally I don't see any good strategies from Sun on getting it
back. Sun still thinks they are in the hardware business -- they've
released an Opteron server with no software!

Maybe it's not really for sale yet, has anyone tried to order one?
The specs page says Solaris x86 will be available in April, yet the
order page says:

The Solaris Operating System (x86 Platform Edition) is
pre-installed on the server in 32-bit ...

however you can't select Solaris in the menu.

It is at a pretty good price, though, which is both surprising and exciting.

/fc