Comparison between Intel dual cores and XEON processors

[Richard Cook]

Has anyone seen any benchmark comparisons out there which compare the
Intel dual core processors against the same clock speed XEON x 2?

It's like there's a conspiracy by Intel to not allow the two to be
compared. I suspect that the Pentium D 820 (2.8GHz) would be quite close
to the XEON parts. Or at least within 5% or so...

Thanks for any guidance!

 

[YKhan]

There was a recent article on Tom's Hardware which compared AMD
dual-cores against dual AMD single-cores. They mentioned a Xeon SC vs.
Xeon DC in that article too, but never seemed to do any tests of them.
It was a rather confusing article, so I'm not going to bother posting
its URL here again, you can do the search on Tom's site. But it's not
really going to tell you much about your particular question about
Xeons, but it will tell you plenty about Opterons.

Yousuf Khan

 

[George Macdonald]

Has anyone seen any benchmark comparisons out there which compare the
Intel dual core processors against the same clock speed XEON x 2?

It's like there's a conspiracy by Intel to not allow the two to be
compared. I suspect that the Pentium D 820 (2.8GHz) would be quite close
to the XEON parts. Or at least within 5% or so...

It wouldn't surprise me if the desktop chip is faster in some/many apps:
Intel usually derates the FSB on Xeons which on top of registering makes
the memory response worse, so they throw more cache into the Xeon soup. It
also depends on which chipset you want to look at - Xeon will shine with
dual sockets and dual independent bus.

You see the same thing with Athlon64 vs. Opteron - the former is almost
always faster if you can get the 1T command rate on the memory but cache is
used to boost the Opteron.

 

[daytripper]

It wouldn't surprise me if the desktop chip is faster in some/many apps:
Intel usually derates the FSB on Xeons which on top of registering makes
the memory response worse, so they throw more cache into the Xeon soup. It
also depends on which chipset you want to look at - Xeon will shine with
dual sockets and dual independent bus.

[snipped]

Unlike desktop P4s, Xeons run on multi-dropped buses. That problem goes away
with processors running behind Blackford et al. Going point-to-point should
enable 1.33 Gigawiggle FSBs...

 

[Richard Cook]

It wouldn't surprise me if the desktop chip is faster in some/many apps:
Intel usually derates the FSB on Xeons which on top of registering makes
the memory response worse, so they throw more cache into the Xeon soup. It
also depends on which chipset you want to look at - Xeon will shine with
dual sockets and dual independent bus.

Thanks for your reply, George. Assuming that the Dual XEON _is_ using a
dual independent bus, and are 2.8GHz parts, do you think the difference in
performance would exceed 10% over the dual core Pentium D 820, or would
you think it'd be somewhat less?

You see the same thing with Athlon64 vs. Opteron - the former is almost
always faster if you can get the 1T command rate on the memory but cache is
used to boost the Opteron.

I would like to go with AMD's offerings due to being the best bang for the
buck, Unfortunately, our customers are stuck on Intel. They're military
and industrial process control types - practically immovable. In fact,
some of them still insist on using NT 4.0 making driver issues a real
headache.

Thanks again!

 

[nobody]

....snip...

I would like to go with AMD's offerings due to being the best bang for the
buck, Unfortunately, our customers are stuck on Intel. They're military
and industrial process control types - practically immovable. In fact,
some of them still insist on using NT 4.0 making driver issues a real
headache.

Thanks again!

I was under impression that MS discontinued 4.0 and all support
thereof about a year ago. In fact, one of my projects of that time
was rewriting of one old system in .NET because 4.0 servers were
subject to decom by Nov.'04, and 2k/2003 server didn't play well with
some legacy technology that ran on 4.0 just fine. But then, those
guys demanding 4.0 may be invested into similar legacy software and
have no (budget, expertise, intent, other, all of the above - pick any
;-) to do a major app overhaul.
NNN

 

[Richard Cook]

...snip...

I was under impression that MS discontinued 4.0 and all support
thereof about a year ago.

I believe you are correct.

In fact, one of my projects of that time
was rewriting of one old system in .NET because 4.0 servers were
subject to decom by Nov.'04, and 2k/2003 server didn't play well with
some legacy technology that ran on 4.0 just fine. But then, those
guys demanding 4.0 may be invested into similar legacy software and
have no (budget, expertise, intent, other, all of the above - pick any
;-) to do a major app overhaul.

Actually, we're using NT 4.0 embedded licenses for our customers, which
gets us around the availability issue.

 

[George Macdonald]

It wouldn't surprise me if the desktop chip is faster in some/many apps:
Intel usually derates the FSB on Xeons which on top of registering makes
the memory response worse, so they throw more cache into the Xeon soup. It
also depends on which chipset you want to look at - Xeon will shine with
dual sockets and dual independent bus.

[snipped]

Unlike desktop P4s, Xeons run on multi-dropped buses. That problem goes away
with processors running behind Blackford et al. Going point-to-point should
enable 1.33 Gigawiggle FSBs...

All those code names - so confusing.

 

[HenryNettles]

Has anyone seen any benchmark comparisons out there which compare the
Intel dual core processors against the same clock speed XEON x 2?

It's like there's a conspiracy by Intel to not allow the two to be
compared. I suspect that the Pentium D 820 (2.8GHz) would be quite close
to the XEON parts. Or at least within 5% or so...

It wouldn't surprise me if the desktop chip is faster in some/many apps:
Intel usually derates the FSB on Xeons which on top of registering makes
the memory response worse, so they throw more cache into the Xeon soup. It
also depends on which chipset you want to look at - Xeon will shine with
dual sockets and dual independent bus.

[snipped]

Unlike desktop P4s, Xeons run on multi-dropped buses. That problem goes away
with processors running behind Blackford et al. Going point-to-point should
enable 1.33 Gigawiggle FSBs...

All those code names - so confusing.

Are you sure there is such a word as "gigawiggle"? My first thought on
seeing this was that it must be a made-up word, like someone saying
"bazillion" when they mean a really large number.

A quick search on google finds one hit, something to do with antennas and
feed horns.

 

[HenryNettles]

...snip...

I was under impression that MS discontinued 4.0 and all support
thereof about a year ago. In fact, one of my projects of that time
was rewriting of one old system in .NET because 4.0 servers were
subject to decom by Nov.'04, and 2k/2003 server didn't play well with
some legacy technology that ran on 4.0 just fine. But then, those
guys demanding 4.0 may be invested into similar legacy software and
have no (budget, expertise, intent, other, all of the above - pick any
;-) to do a major app overhaul.
NNN

The last time I had to work with NT4, the drivers were a real problem.
Among other things, NT4 had (has) no support for modern, fast IDE
interfaces, which left a lot to be desired performance wise. I suppose you
could work around this by using SCSI drives, but the cost differential for
SCSI is huge these days.

Still, if they have ancient software which only works on NT4, and it would
cost tens or hundreds of thousands to rework the software, then....

Who was it that said the hardware is ALWAYS cheaper than the software? I
seem to recall reading that in Jerry Pournelle's column in Byte magazine 15
years ago, but I bet it goes back much further than that.

 

[George Macdonald]

Thanks for your reply, George. Assuming that the Dual XEON _is_ using a
dual independent bus, and are 2.8GHz parts, do you think the difference in
performance would exceed 10% over the dual core Pentium D 820, or would
you think it'd be somewhat less?

I find Intel's roadmaps here a bit confusing so I'm not sure what can be
plugged into what or with which socket. AIUI dual independent bus will be
the Bensley platform which will have the Dempsey dual-core CPUs as its
primary CPU - dunno whether a new single core, up-clocked over the
dual-core part, will play there but nobody seems to be talking about such
an animal. This new platform is supposed to ship 1Q06 and with FB-DIMMs
will very likely beat an equivalent Pentium D, assuming there is in fact an
"equivalent" with single socket.

For currently available Xeons, I'm pretty sure a Pentium D is going to beat
a dual socket Nocona system handily and be very close to a Paxville single
socket system... depending on how cache friendly the applications are. I'm
sorry I don't have any experience with Intel CPUs to put a number on it but
if Daytripper responds here he will have a better handle on things than I.

I would like to go with AMD's offerings due to being the best bang for the
buck, Unfortunately, our customers are stuck on Intel. They're military
and industrial process control types - practically immovable. In fact,
some of them still insist on using NT 4.0 making driver issues a real
headache.

Do those people know they could end up having to do a serious upgrade on
their A/C systems to cope with Intel's current CPUs? This has already
happened. It's just amazing that AMD can't get any respect here, given how
far ahead it has been for the past 2 years or so. I guess they just have
big pocket books -- we know the military does -- for things like NT4
special support.

 

[Rob Stow]

Are you sure there is such a word as "gigawiggle"? My first thought on
seeing this was that it must be a made-up word, like someone saying
"bazillion" when they mean a really large number.

A White House staffer tells George Bush that 3 Brazilian soldiers
were killed by terrorists. Bush immediately says "That's
terrible !", and then after thinking about it for a minute asks
"How many is a brazillion? "

 

[George Macdonald]

On Wed, 16 Nov 2005 15:21:33 -0500, George Macdonald


Has anyone seen any benchmark comparisons out there which compare the
Intel dual core processors against the same clock speed XEON x 2?

It's like there's a conspiracy by Intel to not allow the two to be
compared. I suspect that the Pentium D 820 (2.8GHz) would be quite close
to the XEON parts. Or at least within 5% or so...

It wouldn't surprise me if the desktop chip is faster in some/many apps:
Intel usually derates the FSB on Xeons which on top of registering makes
the memory response worse, so they throw more cache into the Xeon soup. It
also depends on which chipset you want to look at - Xeon will shine with
dual sockets and dual independent bus.
[snipped]

Unlike desktop P4s, Xeons run on multi-dropped buses. That problem goes away
with processors running behind Blackford et al. Going point-to-point should
enable 1.33 Gigawiggle FSBs...

All those code names - so confusing.

Are you sure there is such a word as "gigawiggle"? My first thought on
seeing this was that it must be a made-up word, like someone saying
"bazillion" when they mean a really large number.

It's a good made-up word though. Intel uses GHz for its FSB transfer
data rates when there are clearly no GHz in sight; it's been a sore point
here in the past when people quote Intel's marketroids' err units... when
clearly what is meant is GT/s, Giga-Transfers per second.

 

[Keith]

On Wed, 16 Nov 2005 15:21:33 -0500, George Macdonald


Has anyone seen any benchmark comparisons out there which compare the
Intel dual core processors against the same clock speed XEON x 2?

It's like there's a conspiracy by Intel to not allow the two to be
compared. I suspect that the Pentium D 820 (2.8GHz) would be quite close
to the XEON parts. Or at least within 5% or so...

It wouldn't surprise me if the desktop chip is faster in some/many apps:
Intel usually derates the FSB on Xeons which on top of registering makes
the memory response worse, so they throw more cache into the Xeon soup. It
also depends on which chipset you want to look at - Xeon will shine with
dual sockets and dual independent bus.
[snipped]

Unlike desktop P4s, Xeons run on multi-dropped buses. That problem goes away
with processors running behind Blackford et al. Going point-to-point should
enable 1.33 Gigawiggle FSBs...

All those code names - so confusing.

Are you sure there is such a word as "gigawiggle"? My first thought on
seeing this was that it must be a made-up word, like someone saying
"bazillion" when they mean a really large number.

It's a good made-up word though. Intel uses GHz for its FSB transfer
data rates when there are clearly no GHz in sight; it's been a sore point
here in the past when people quote Intel's marketroids' err units... when
clearly what is meant is GT/s, Giga-Transfers per second.

Oh, it's en even better term than that. I implies both data
transfer rate and MHz, without specifying either. It makes a lot
of sense, and is somewhat humerous at the same time. It might just
catch on!

 

[Richard Cook]

Do those people know they could end up having to do a serious upgrade on
their A/C systems to cope with Intel's current CPUs?

Most likely, not - I don't think power consumption is a issue with these
folks anyway. To give you an idea of how without clue these people are, we
have a particular customer (military) who is *still* buying slot-1 BX
based dual P3 PICMG motherboards with 700MHz processors from us. Well,
we've pretty much dried up the planets supply of 700MHz P3 CPU chips, yet
they keep insisting that we find more, like we have some pull at Intel to
re-open an old mask, or something. Last I heard, they were going after the
manufacturer of the processor board to find out why they still can't get
these CPU chips! (rolls eyes)

Thanks again for your help - looks like the Pentium D is the better way to
go for now.

 

[Del Cecchi]

A White House staffer tells George Bush that 3 Brazilian soldiers were
killed by terrorists. Bush immediately says "That's terrible !", and
then after thinking about it for a minute asks "How many is a brazillion? "

Does this mean it is open season on political jokes and other off topic
crap? If so, I can come up with some.

 

[Del Cecchi]

Most likely, not - I don't think power consumption is a issue with these
folks anyway. To give you an idea of how without clue these people are, we
have a particular customer (military) who is *still* buying slot-1 BX
based dual P3 PICMG motherboards with 700MHz processors from us. Well,
we've pretty much dried up the planets supply of 700MHz P3 CPU chips, yet
they keep insisting that we find more, like we have some pull at Intel to
re-open an old mask, or something. Last I heard, they were going after the
manufacturer of the processor board to find out why they still can't get
these CPU chips! (rolls eyes)

Thanks again for your help - looks like the Pentium D is the better way to
go for now.

To show how without a clue the poster is, consider what the military
might be doing with those boards.

Would you be able to certify that an expensive piece of equipment
complete with expensive software will work properly if you substitute
some other motherboard? How much would it cost to perform the necessary
testing? How much would it cost to fix the software? How much to deal
with the maintenance channels now that there are two different flavors
of boards and software?

How much to deal with the accounts in the NY Times about the "50,000
dollar pcs" because those charges get rolled into the hardware purchase?

 

[Tony Hill]

To show how without a clue the poster is, consider what the military
might be doing with those boards.

Would you be able to certify that an expensive piece of equipment
complete with expensive software will work properly if you substitute
some other motherboard? How much would it cost to perform the necessary
testing? How much would it cost to fix the software? How much to deal
with the maintenance channels now that there are two different flavors
of boards and software?

How much to deal with the accounts in the NY Times about the "50,000
dollar pcs" because those charges get rolled into the hardware purchase?

The military people buying the boards are still completely clueless,
or at best EXTREMELY shortsighted in this case. If you're designing
an application that is going to be that difficulty to certify in the
first place then you had absolutely MUST make sure that you'll have
adequate sources for parts to support that application through it's
lifetime.

There are good reasons why big companies like HPaq and IBM offer
long-term service contracts in such situations.

 

[Del Cecchi]

The military people buying the boards are still completely clueless,
or at best EXTREMELY shortsighted in this case. If you're designing
an application that is going to be that difficulty to certify in the
first place then you had absolutely MUST make sure that you'll have
adequate sources for parts to support that application through it's
lifetime.

There are good reasons why big companies like HPaq and IBM offer
long-term service contracts in such situations.

In defense of the military, who would have thought that they would still
be flying B-52s in 2005? And I can recall significant scrambling around
various IBM facilities to keep those ATC triplex 360/65s running so that
folks could continue to ride airplanes around. Something about the
planned replacement project crashing and burning so that the 360s had to
be maintained for another while. I wonder if the replacement 3081s are
still there?

Internal and lab machines were cannibalized to get parts. Finding SLT
modules in 1980 or so was a good trick.

And how about that IRS? What was the lifetime of that system again?

del cecchi

 

[Keith]

In defense of the military, who would have thought that they would still
be flying B-52s in 2005?

....and planning to keep them active until 2030 or beyond.
Everything electrical/electronic has been replaced a number of
times already though.

And I can recall significant scrambling around
various IBM facilities to keep those ATC triplex 360/65s running so that
folks could continue to ride airplanes around. Something about the
planned replacement project crashing and burning so that the 360s had to
be maintained for another while. I wonder if the replacement 3081s are
still there?

I remember having to redesign Germanium transistors and dodes out
of several circuits to keep /360s running. It seems someone found
that the only company still making Ge was a garage outfit; scared
the hell out of the PHBs!