Paxville spanked.

  • Thread starter Thread starter gimp
  • Start date Start date
We all knew this was coming.

Intel has a long history of putting out selected test results
that it can spin in favour of its chips. This time the product
was so bad that Intel didn't even try any spin doctoring.

The performance results don't surprise me at all. Two Xeon chips
shoved together into a multi-chip package simply couldn't be
expected to perform any better than a pair of similarly clocked
single-core Xeons.

However, the power results do surprise me. If Intel wanted Xeon
server and workstation users to stay with Intel chips by using
Paxville instead of switching to Opteron, they needed to offer
those buyers a reason to do so. Since we all knew that
performance wasn't going to be the reason, Intel simply *had* to
find a miracle that would at least partly solve Xeon's long
standing power problems.

How embarrassing!

Since both performance and power are huge problems with Paxville,
I am amazed that Intel would put it on the market at all. Intel
is AMD's new best buddy - they really lent AMD a helping hand.

Heads should roll at Intel for letting Paxville escape from the
laboratory. They should have hidden behind "fab shortages",
"technical difficulties", or whatever other excuses they could
contrive until they had a *much* better product to put on the market.
How can any IT dept. justify buying this turkey?...
we'll see.:-)

As other recent threads have shown, there is no shortage of
idiots who will buy Intel regardless of the cost.

And the cost is *very* high. With the single core chips you
typically need 5 Xeon boxes to do the work of 3 similarly
configured Opteron boxes and it looks like that ratio is going to
continue now that Intel finally has something that they can
pretend is a dual-core Xeon. Factor in the need for additional
software licenses, more racks, bigger server rooms, higher power
consumption, and a lot more cooling, and the cost of using Xeons,
whether single-core or Paxville, is prohibitive.
 
Rob said:
We all knew this was coming.

Intel has a long history of putting out selected test results
that it can spin in favour of its chips. This time the product
was so bad that Intel didn't even try any spin doctoring.

The performance results don't surprise me at all. Two Xeon chips
shoved together into a multi-chip package simply couldn't be
expected to perform any better than a pair of similarly clocked
single-core Xeons.

However, the power results do surprise me. If Intel wanted Xeon
server and workstation users to stay with Intel chips by using
Paxville instead of switching to Opteron, they needed to offer
those buyers a reason to do so. Since we all knew that
performance wasn't going to be the reason, Intel simply *had* to
find a miracle that would at least partly solve Xeon's long
standing power problems.



Since both performance and power are huge problems with Paxville,
I am amazed that Intel would put it on the market at all. Intel
is AMD's new best buddy - they really lent AMD a helping hand.

Heads should roll at Intel for letting Paxville escape from the
laboratory. They should have hidden behind "fab shortages",
"technical difficulties", or whatever other excuses they could
contrive until they had a *much* better product to put on the market.


As other recent threads have shown, there is no shortage of
idiots who will buy Intel regardless of the cost.

And the cost is *very* high. With the single core chips you
typically need 5 Xeon boxes to do the work of 3 similarly
configured Opteron boxes and it looks like that ratio is going to
continue now that Intel finally has something that they can
pretend is a dual-core Xeon. Factor in the need for additional
software licenses, more racks, bigger server rooms, higher power
consumption, and a lot more cooling, and the cost of using Xeons,
whether single-core or Paxville, is prohibitive.

I wonder how Dell played in the release of this chip, if not just a
marketing ploy with a me to, to the Dell faithful.

What is really sad this won't really make a big difference in Intel's
bottom line as they are still making money hand over fist. It seems
that the old line is still true you never get fired for buying Intel. I
just wonder how many other reviews will come out that are even
balanced.

Gnu_Raiz
 
Bitstring <[email protected]>, from
the wonderful person (e-mail address removed) said
It seems
that the old line is still true you never get fired for buying Intel.

Is true but eventually (~100 years) the companies that compute with AMD
are going to outgrow the companies that compute with Intel, all else
being equal (which it won't be .. but I'd assume that stupidity over IT
supplier would also extend into stupidity over other business
decisions). Of course, that assumes AMD can =stay= ahead for the 100
years ...
 
Rob said:
We all knew this was coming.

Intel has a long history of putting out selected test results that it
can spin in favour of its chips. This time the product was so bad that
Intel didn't even try any spin doctoring.

Today I was watching that pro-Intel business channel, otherwise known as
CNBC, and even they couldn't hide Intel's problems. In fact, the host
Larry Kudlow seemed downright hostile at excuse-making on behalf of Intel.

Yousuf Khan
 
I wonder how Dell played in the release of this chip, if not just a
marketing ploy with a me to, to the Dell faithful.

What is really sad this won't really make a big difference in Intel's
bottom line as they are still making money hand over fist. It seems
that the old line is still true you never get fired for buying Intel. I
just wonder how many other reviews will come out that are even
balanced.

Gnu_Raiz

The procurement decisions are usually made in corner offices, and
these are not the places where you would expect enough of sense.

NNN
 
How embarrassing! How can any IT dept. justify buying this turkey?...
we'll see.:-)

It's going to take a bit more than a review on "GamePC" to convince
most IT dept. managers to change their views on things!

That being said, you are of course right. Someone would have to have
rocks in their head to even consider buying a Xeon server these days!
 
I wonder how Dell played in the release of this chip, if not just a
marketing ploy with a me to, to the Dell faithful.

If the early release of Paxville was intended to steal some of
AMD's thunder over the opening of Fab 36, it backfired badly. It
dramatically emphasizes that AMD is ready to ramp up Opteron
production at a time when Intel, yet again, has nothing that can
compete.

Hence, I would imagine that even though their public face is
still smiling, behind closed doors Dell is extremely angry with
Intel. Paxville has got to be a huge let down for Dell.
What is really sad this won't really make a big difference in Intel's
bottom line as they are still making money hand over fist. It seems
that the old line is still true you never get fired for buying Intel.
I
just wonder how many other reviews will come out that are even
balanced.

I'm sure quite a few people have Paxville boxes but are still
under NDA. I am eager to see what they eventually are allowed to
say about Paxville.
 
We all knew this was coming.

Yes we did but this is a slaughter.
Intel has a long history of putting out selected test results
that it can spin in favour of its chips. This time the product
was so bad that Intel didn't even try any spin doctoring.

The performance results don't surprise me at all. Two Xeon chips
shoved together into a multi-chip package simply couldn't be
expected to perform any better than a pair of similarly clocked
single-core Xeons.

However, the power results do surprise me. If Intel wanted Xeon
server and workstation users to stay with Intel chips by using
Paxville instead of switching to Opteron, they needed to offer
those buyers a reason to do so. Since we all knew that
performance wasn't going to be the reason, Intel simply *had* to
find a miracle that would at least partly solve Xeon's long
standing power problems.

Oh-oh I see something which is way off in the benchmarks: The Sandra Memory
Bandwidth performance is very bad on the Xeon system: it's about half what
one would expect for a 800MHz FSB dual channel memory system, even a P4
uniprocessor system of a year ago or so. Either the mbrd is not doing its
thing right or GamePC screwed up on configuration - even a shared FSB does
not account for the anomaly here. The Opteron systems are about right for
registered memory and an Athlon64 with unbuffered DIMMs at 1T timing would
add about 1.3GB/s IME.
Since both performance and power are huge problems with Paxville,
I am amazed that Intel would put it on the market at all. Intel
is AMD's new best buddy - they really lent AMD a helping hand.

Heads should roll at Intel for letting Paxville escape from the
laboratory. They should have hidden behind "fab shortages",
"technical difficulties", or whatever other excuses they could
contrive until they had a *much* better product to put on the market.


As other recent threads have shown, there is no shortage of
idiots who will buy Intel regardless of the cost.

True - I know of someone who recently bought a Dell dual Xeon system for
the office (at a firesale price:-)) and is very impressed by it. His
comment on AMD: "I decided to take a chance on an AMD system for home and
it actually works quite well - no compatibility issues at all... I'm
surprised how well it works".;-)
 
It's going to take a bit more than a review on "GamePC" to convince
most IT dept. managers to change their views on things!

True and we'll see what others make of it. As noted elsewhere, the Sandra
Memory Bandwidth is all to hell on the Xeon system - about half what it
should be so either the new mbrd is flawed or GamePC screwed up on config.
 
---> skip

G> True - I know of someone who recently bought a Dell dual Xeon
G> system for the office (at a firesale price:-)) and is very
G> impressed by it. His comment on AMD: "I decided to take a chance
G> on an AMD system for home and it actually works quite well - no
G> compatibility issues at all... I'm surprised how well it
G> works".;-)

G> -- Rgds, George Macdonald

Amazing statement. I have heard similiar stuff in the past. I guess
when you have no respect it is hard to get any respect.

Another point dell does unload hardware at some amazingly
Amazing statement. I have heard similiar statments in the past.

My friend's son worked at dell in sales, he said that dell always
unloads computers at the end of the quarter in order to boost sales
reports. I would suspect if you get the right price the duel xeon
systems are competitive.

Whatever.
 


I'm starting to loose respect for Intel .

Apache Bench - 10,000 Users w/ Concurrency Of 2
------------------------------------------------

Upgrade from Intel 2.4 GHz to 2.8 GHz.
You Gain: 9.1 percent increase.

Upgrade from Opteron 1.8 GHz to Opteron 2.4 GHz.
You Gain: 33.6 percent increase.

Upgrade from Intel 2.4 GHz to Opteron 1.8 GHz.
You Gain: 75.2 percent increase.

Upgrade from Intel 2.8 GHz to Opteron 2.4 GHz.
You Gain: 114.6 percent increase.


Cheers,
Ed
 
EdG said:
I'm starting to loose respect for Intel .

Apache Bench - 10,000 Users w/ Concurrency Of 2
------------------------------------------------

Upgrade from Intel 2.4 GHz to 2.8 GHz.
You Gain: 9.1 percent increase.

Upgrade from Opteron 1.8 GHz to Opteron 2.4 GHz.
You Gain: 33.6 percent increase.

Upgrade from Intel 2.4 GHz to Opteron 1.8 GHz.
You Gain: 75.2 percent increase.

Upgrade from Intel 2.8 GHz to Opteron 2.4 GHz.
You Gain: 114.6 percent increase.

I know someone who had 24 Opty 240 dualies running Apache.

He was told be by his boss to come up with a proposal to replace
them with something newer that would at least double their web
serving capacity.

HP suggested either 18 to 20 Opty 270 2P boxes or 11 or 12 Opty
870 4P boxes.

Dell suggested *52* Xeon 3.0 GHz / 2 MB 2P boxes.

They went with 20 Opty 270 boxes.
 
YKhan said:
Well, apparently this is what Dell (the man, but probably also the
company) thinks about it:

http://www.theinquirer.net/?article=27122

Yousuf Khan

Did you mean to link to this one instead:
http://www.theinquirer.net/?article=27109

That one says Dell thinks Intel will be "supercompetitive" next
year - which implies that Dell is specifically ignoring Paxville
and waiting for Intel's next generation of processors.

I have yet to see *anything* where Dell comments specifically on
Paxville.
 
So what happened to the 24 Opty boxes, why not just upgrade the cpu's,
seems like the 270's should be a pop in upgrade.

Or is this how all those used Opterons show up on ebay, I keep seeing
matched pairs showing up for sale. Only problem is the motherboards
cost an arm and a leg, as I really do not see to many of them for sale,
for some odd reason.

Gnu_Raiz
 
So what happened to the 24 Opty boxes, why not just upgrade the cpu's,
seems like the 270's should be a pop in upgrade.

Or is this how all those used Opterons show up on ebay, I keep seeing
matched pairs showing up for sale. Only problem is the motherboards
cost an arm and a leg, as I really do not see to many of them for sale,
for some odd reason.

Gnu_Raiz

It happened that some motherboard makers (particularly MSI)
underestimated the current requirements of (back then upcoming) 90nm
chips. Even though the wattage is the same (or even somewhat less)
compared to 130nm, the newer chips run on lesser voltage, so the
current they draw is greater. Wouldn't be too bad if it were not
greater than VRs can handle reliably without the risk of getting fried
too soon. As an old electrician used to say, "It ain't voltage that's
gonna kill you son, it's amps."
In short, no dual core or even single core upgrade beyond 250 for my
master2-far :-(
The dual socket 940 motherboards are not a big problem. A quick check
on Pricewatch: ASUS K8N-DL at Newegg for $225.98 delivered. More
expensive than most single socket boards, but duallies were always
priced higher.

NNN
 
So what happened to the 24 Opty boxes, why not just upgrade the cpu's,
seems like the 270's should be a pop in upgrade.

The Opty 240 boxes were leased, not purchased, and they are going
back to the manufacturer.

The terms of the lease specified that all parts and labour for
hardware upgrades were to be provided by the manufacturer. At
that manufacturer's ridiculously inflated pricing, replacing the
CPU's and replacing the 1 GB DIMMs with 2 GB DIMMs cost more than
buying brand new boxes from HP.

An option was to buy some of the leased machines and then upgrade
them. That was cheaper than paying the manufacturer to upgrade
them, but still more expensive than simply buying new machines
from HP.

If anyone asks, I will decline to name the manufacturer of the
leased machines. Apparently things involved with the termination
of the lease and the return of the Opty 240 boxes is not going
well and the lawyers have become involved. Sigh.

Or is this how all those used Opterons show up on ebay, I keep seeing
matched pairs showing up for sale. Only problem is the motherboards
cost an arm and a leg, as I really do not see to many of them for sale,
for some odd reason.

You hinted at the reason for the used processor abundance but
motherboard scarcity yourself: the drop-in Opty CPU single core
to dual core CPU replacements are so easy to do. And when the
motherboard doesn't support dual-core, the whole box is sold, not
just the parts.

When I do see a used Opty server motherboards for sale, the price
is usually so high that it wouldn't make sense to not spend just
a little and buy a brand new motherboard.

If you don't need motherboards with SCSI on-board, look for
workstation boards instead of server boards. If you have a SCSI
card on hand, the non-SCSI variant of Tyan S2885, for example,
makes a nice server board and I have seen them going for as
little as $USD 250 used, versus $USD 425 new.

(There is one S2885 auction on E-Bay now, currently at $249.)

Note also that except for the S2881 and some steppings of the
s2885, the Tyan boards in the S288x and S287x series are not
dual-core compatible. For that reason you have a lot of
dickering power with vendors trying to unload their stocks of
those boards. Base prices for the S2880 and S2882 are 70% of
what they were are year ago and you can dicker down from there.
I'm sure the same thing applies to other brands of older
dual-socket 940 motherboards.

If you do need dual-core capability, the SuperMicro boards seem
to have good prices and features. *Please* be my guinea pig and
try one of these for me:
http://www.lynncomp.com/index.html?loadfile=item89050.html
 
Rob said:
Did you mean to link to this one instead:
http://www.theinquirer.net/?article=27109

That one says Dell thinks Intel will be "supercompetitive" next year -
which implies that Dell is specifically ignoring Paxville and waiting
for Intel's next generation of processors.

Either link is fine, I guess. Related articles, I'd say.
I have yet to see *anything* where Dell comments specifically on Paxville.

Well, I don't know, about Paxville specifically, but he was generically
talking about Intel Dual Core Xeons leading AMD Dual Core Opterons by
2006. Paxville could make it into 2006, I suppose. Maybe he expects
frequencies of Paxville to increase high enough for it to be competitive
with Opteron?

Yousuf Khan
 
Back
Top