SEO

hi there,

does anyone have any performance data comparing E420Rs (4 x 450Mhz,
4GB) to modern V480 (4 x 1.0Ghz) or T2000 servers.

basically a V480 is equivalent to how many E420Rs?
I'm not so concerned over IO or RAM but mainly raw CPU power.

I cannot find this sort of performance comparison data on SUN site and
their support people do not know either - (they know how to try sell me
servers though...)


Also is it a fair comparison to add up CPU cyles when making
comparions?
eg 4 way E420R == 4 x 450Mhz == 1.8Ghz of CPU

therefore it is similar to a 2 x 1.0Ghz machine?

Or how would you normally compare servers like this?

I understand the chips are different too....

thanks in advance,

robert

> Also is it a fair comparison to add up CPU cyles when making
> comparions?
> eg 4 way E420R == 4 x 450Mhz == 1.8Ghz of CPU
>
> therefore it is similar to a 2 x 1.0Ghz machine?
>

If you look at the SPEC benchmarks you see that in the integer domain
the speed is linear to the cpu frequenz. So i would say yes. For
floating point there was a significant jump in the USIII.

But why the hell are you talking about pure CPU power with a sparc cpu
?
If you want integer cpu power get an ix86 system and if you want
floating point get an itanium.

llothar wrote:
> > Also is it a fair comparison to add up CPU cyles when making
> > comparions?
> > eg 4 way E420R == 4 x 450Mhz == 1.8Ghz of CPU
> >
> > therefore it is similar to a 2 x 1.0Ghz machine?
> >
>
> If you look at the SPEC benchmarks you see that in the integer domain
> the speed is linear to the cpu frequenz. So i would say yes. For
> floating point there was a significant jump in the USIII.
>
> But why the hell are you talking about pure CPU power with a sparc cpu
> ?
> If you want integer cpu power get an ix86 system and if you want
> floating point get an itanium.

I am going to assume you are being sarcastic about the itanium as they
are true dogs in terms of performance. Although SPEC benchmarks tell
alot about a systems performance, the only true measure of performance
is a test run with your specific load. Having said that, the loads that
I deal with the Sparc floating point and integer performance is far
better than any of the Intel processors. Only an AMD Opteron provides
any competition and then only at 3x the clock speed. Running MM5,
HYSPLIT4 and SLU-DRAS is an exercise in patience and frustration with
an x86 processor. Random crashes of the system and programs are what
you can expect from x86.

Words to the wise, rpasken@eas.slu.edu wrote:

>Having said that, the loads that
>I deal with the Sparc floating point and integer performance is far
>better than any of the Intel processors. Only an AMD Opteron provides
>any competition and then only at 3x the clock speed. Running MM5,
>HYSPLIT4 and SLU-DRAS is an exercise in patience and frustration with
>an x86 processor. Random crashes of the system and programs are what
>you can expect from x86.

Out of curiousity, what do you think how the new T1 cpu would perform
in your line of work?
--
Claus Dragon <clauskick@mpsahotmail.com>
=(UDIC)=
d++ e++ T--
K1!2!3!456!7!S a26
"Coffee is a mocker. So, I am going to mock."

- Me, lately.

thanks for the info around the comparisons.

For the most part, large organizations want stability and performance.
X86 architecture doesn't give the stability that SPARC does. (as
pointed out earlier)

Its getting better, but still SPARC is the hardware of choice for
discerning customers....

i've heard from customers that a T2000 server is "twice as fast" as a
V480....so you can read from that what you will.

-robert

Claus Dragon wrote:
> Words to the wise, rpasken@eas.slu.edu wrote:
>
> >Having said that, the loads that
> >I deal with the Sparc floating point and integer performance is far
> >better than any of the Intel processors. Only an AMD Opteron provides
> >any competition and then only at 3x the clock speed. Running MM5,
> >HYSPLIT4 and SLU-DRAS is an exercise in patience and frustration with
> >an x86 processor. Random crashes of the system and programs are what
> >you can expect from x86.
>
> Out of curiousity, what do you think how the new T1 cpu would perform
> in your line of work?
> --
> Claus Dragon <clauskick@mpsahotmail.com>
> =(UDIC)=
> d++ e++ T--
> K1!2!3!456!7!S a26
> "Coffee is a mocker. So, I am going to mock."
>
> - Me, lately.

In article <1149809644.479590.19110@j55g2000cwa.googlegroups. com>,
<robert.dowd@cp.net> wrote:
>hi there,
>
>does anyone have any performance data comparing E420Rs (4 x 450Mhz,
>4GB) to modern V480 (4 x 1.0Ghz) or T2000 servers.

The V480 is retired, why are you comparing V480 and
not V490?

>basically a V480 is equivalent to how many E420Rs?
>I'm not so concerned over IO or RAM but mainly raw CPU power.

I would look at the SPEC int rate numbers.
a 4CPU 1.5GHZ USIV+ gives apx CINT2000 Rate 78/87.
a 4CPU E420R gives less than CINT2000 Rate 10/11.

So a V490 equals eight 420R in throughput.
Using single threaded applications I would assume a V490 is 4x
as fast.

Regards
Fredrik

--
Fredrik Lundholm
dol @ ce.chalmers.se

* rpasken@eas.slu.edu:

>> But why the hell are you talking about pure CPU power with a sparc cpu
>> ?
>> If you want integer cpu power get an ix86 system and if you want
>> floating point get an itanium.
>
> I am going to assume you are being sarcastic about the itanium

why should that be sarcastic?

> as they
> are true dogs in terms of performance. Although SPEC benchmarks tell
> alot about a systems performance, the only true measure of performance
> is a test run with your specific load. Having said that, the loads that
> I deal with the Sparc floating point and integer performance is far
> better than any of the Intel processors.

Fine. And for the tasks my employee does (mostly numerical simulations)
the Itanium has proven to be the fastest solution when we bought our HP
Superdomes (around between 1-2 years ago). The SPARC numbers were
somewhere in the lower range so that it simply didn't cut it.

> Only an AMD Opteron provides
> any competition and then only at 3x the clock speed. Running MM5,
> HYSPLIT4 and SLU-DRAS is an exercise in patience and frustration with
> an x86 processor. Random crashes of the system and programs are what
> you can expect from x86.

Only of you choose crap hardware or if you make other mistakes (which
also would lead to a crashing SPARC system). We run shitloads of x64 and
x64 systems at work, most of thenm 24hrs/day and under heavy load, and
we still have to see the "random crashes" you're talking about...

Benjamin

Benjamin Gawert wrote:
> * rpasken@eas.slu.edu:
>
> >> But why the hell are you talking about pure CPU power with a sparc cpu
> >> ?
> >> If you want integer cpu power get an ix86 system and if you want
> >> floating point get an itanium.
> >
> > I am going to assume you are being sarcastic about the itanium
>
> why should that be sarcastic?
>
> > as they
> > are true dogs in terms of performance. Although SPEC benchmarks tell
> > alot about a systems performance, the only true measure of performance
> > is a test run with your specific load. Having said that, the loads that
> > I deal with the Sparc floating point and integer performance is far
> > better than any of the Intel processors.
>
> Fine. And for the tasks my employee does (mostly numerical simulations)
> the Itanium has proven to be the fastest solution when we bought our HP
> Superdomes (around between 1-2 years ago). The SPARC numbers were
> somewhere in the lower range so that it simply didn't cut it.
>
> > Only an AMD Opteron provides
> > any competition and then only at 3x the clock speed. Running MM5,
> > HYSPLIT4 and SLU-DRAS is an exercise in patience and frustration with
> > an x86 processor. Random crashes of the system and programs are what
> > you can expect from x86.
>
> Only of you choose crap hardware or if you make other mistakes (which
> also would lead to a crashing SPARC system). We run shitloads of x64 and
> x64 systems at work, most of thenm 24hrs/day and under heavy load, and
> we still have to see the "random crashes" you're talking about...
>
> Benjamin

I should have explain what MM5, HYPSLIT4 and SLU-DRAS are. MM5 is a
very fine scale (1km/1min resolution) numerical weather prediction
model It does very little I/O and is very floating point intensive. It
reads raw weather data at the begining of the run and then the only I/O
is when it spits out a specific forecast typically every 30 minutes of
forecast time. HYSPLIT4 is an atmospheric dispersion model. Using MM5
output it computes the concentrations of chemicals as they are
dispersed. It handles chemical transformations as well as dispersion.
SLU-DRAS is a real-time Doppler radar/lidar analysis system. It takes
live multiple Doppler velocity and reflectivity data and converts it
into the 3-D wind field, temperatures and pressures. All three systems
are well documented software packages that run on a wide range of
hardware. Only nuclear bomb testing software is more floating point
intensive.

I used Compaq/HP x86 hardware (ML-350 dual 1.4GHZ P3's)and a Sun (Blade
1000 dual 750GHZ) running both Solaris and Linux for the benchmarks. No
benchmark of any the software ran less than 5 days. Only about half my
benchmarks completed on the compaq hardware due to sig seg violations
at random points in the code. According to dmesg they were associated
with memory errors.

* rpasken@eas.slu.edu:

> I should have explain what MM5, HYPSLIT4 and SLU-DRAS are.

not necessary...

> MM5 is a
> very fine scale (1km/1min resolution) numerical weather prediction
> model It does very little I/O and is very floating point intensive. It
> reads raw weather data at the begining of the run and then the only I/O
> is when it spits out a specific forecast typically every 30 minutes of
> forecast time. HYSPLIT4 is an atmospheric dispersion model. Using MM5
> output it computes the concentrations of chemicals as they are
> dispersed. It handles chemical transformations as well as dispersion.
> SLU-DRAS is a real-time Doppler radar/lidar analysis system. It takes
> live multiple Doppler velocity and reflectivity data and converts it
> into the 3-D wind field, temperatures and pressures. All three systems
> are well documented software packages that run on a wide range of
> hardware.

Right.

> Only nuclear bomb testing software is more floating point
> intensive.

Definitely not.

> I used Compaq/HP x86 hardware (ML-350 dual 1.4GHZ P3's)and a Sun (Blade
> 1000 dual 750GHZ) running both Solaris and Linux for the benchmarks. No
> benchmark of any the software ran less than 5 days. Only about half my
> benchmarks completed on the compaq hardware due to sig seg violations
> at random points in the code. According to dmesg they were associated
> with memory errors.

So you have run these programs on an ancient Compaq ML350 P3 server with
defective memory. Ok. And how does that relate to your statement that
about Itanium which you obviously never ever used?

Benjamin