The Pentium-D 820 mystery -- apparently not compatible with nF4-SLI (Intel) chip-set ....

[John Lewis]

For background see page 4 of the following article:-

http://www.xbitlabs.com/articles/cpu/display/pentiumd-820.html

There seems to be a FUNCTIONAL problem with the nF4-SLI(Intel)
chip-set if both cores of the Pentium-D 820 are enabled. Applications
crash or hang, the motherboard re-boots... However, the current
explanation given to Xbit Labs by nVidia is that " we only support the
830 and above - the 820 is too slow for enthusiasts" . Zero technical
explanation for the problem...and a bad answer anyway. Many
enthusiasts that I know buy the least expensive and overclock. None of
the nF4-SLI (Intel) motherboards that claim to support the Pentium-D
mention any exclusion of the 820 in their specs. And Asus seemed to
have been caught totally off-guard by the problem, considering their
reaction as reported by Xbit Labs.

So:-

(a) Did nVidia commit a design blunder ?

If so, I would be highly suspicious of stable performance with
the Pentium-D 830, the next one up in the family. One way to
check for a design blunder by nVidia would be to underclock
the 830, or 840, or Pentium Extreme 840 to 2.8GHz -- particularly
the 840 Extreme, since it has unlocked clock dividers--- in the
same nF4-SLI motherboard(s) in which the Pentium-D 820 fails
and see if they all remain stable. Nobody with access to these
parts has yet done this..........

OR

(b) Did Intel slip something into the P-D 820 unknown to
nVidia that would ensure that their least-expensive, highest
volume dual-core would only work on Intel chip-sets, and just
supply the faster parts to nVidia for their chip-set
qualification ??

OR

(c) Is this physical incompatibility part of the chip-set agreement
between nVidia and Intel to minimise the opportunity for
nVidia to take over the Intel dual-core compatible chip-set
market at a critical time for Intel in terms of chip-set
supply ? The Pentium-D 820 is likely to be the highest
volume dual-core part in the Intel stable.

Remember, unlike AMD, Intel dual-core requires a new
motherboard/chip-set for each dual-core processor shipped.

Maybe the cure is as simple as adding 820 support in BIOS -
since the 820 implementation is not exactly the same as the 830
and 840. However, Asus seemed to be caught unawares by the
problem and nVidia's denial of a fix.

There is no obvious difference between the 820 and its faster brothers
that would explain the failure. Yes, it uses a lower-value clock
multiplier and does not implement Speed-Step, but that is a
functionally-identical state to a P4-2.8 single-core Prescott.

Does that mean that a single-core 2.8 GHz processor would also fail in
an Intel nF4-SLI board ? Probably not, since the P-D 820 with one core
disabled works fine. ( The nF4-SLI chip-set is supposed to be
compatible with both single and dual-core LGA775 processors )

Anyway, for the moment, until this thick smoke-screen by nVidia and
Intel clears, I would highly recommend holding off purchase of any
nF4-SLI (Intel) motherboard, regardless of processor initially
installed -- single, or dual-core - if you are expecting the MB to
eventually function with a dual-core processor. I notice that
the very fine MSI P4N Diamond MB (with software-switching of
SLI, and embedded Creative 24-bit audio) is currently available
from Newegg @ $229. This board is potentially affected by the 820
problem... no comments from MSI, so far.

John Lewis

 

[Paul]

For background see page 4 of the following article:-

http://www.xbitlabs.com/articles/cpu/display/pentiumd-820.html

There seems to be a FUNCTIONAL problem with the nF4-SLI(Intel)
chip-set if both cores of the Pentium-D 820 are enabled. Applications
crash or hang, the motherboard re-boots... However, the current
explanation given to Xbit Labs by nVidia is that " we only support the
830 and above - the 820 is too slow for enthusiasts" . Zero technical
explanation for the problem...and a bad answer anyway. Many
enthusiasts that I know buy the least expensive and overclock. None of
the nF4-SLI (Intel) motherboards that claim to support the Pentium-D
mention any exclusion of the 820 in their specs. And Asus seemed to
have been caught totally off-guard by the problem, considering their
reaction as reported by Xbit Labs.

So:-

(a) Did nVidia commit a design blunder ?

If so, I would be highly suspicious of stable performance with
the Pentium-D 830, the next one up in the family. One way to
check for a design blunder by nVidia would be to underclock
the 830, or 840, or Pentium Extreme 840 to 2.8GHz -- particularly
the 840 Extreme, since it has unlocked clock dividers--- in the
same nF4-SLI motherboard(s) in which the Pentium-D 820 fails
and see if they all remain stable. Nobody with access to these
parts has yet done this..........

OR

(b) Did Intel slip something into the P-D 820 unknown to
nVidia that would ensure that their least-expensive, highest
volume dual-core would only work on Intel chip-sets, and just
supply the faster parts to nVidia for their chip-set
qualification ??

OR

(c) Is this physical incompatibility part of the chip-set agreement
between nVidia and Intel to minimise the opportunity for
nVidia to take over the Intel dual-core compatible chip-set
market at a critical time for Intel in terms of chip-set
supply ? The Pentium-D 820 is likely to be the highest
volume dual-core part in the Intel stable.

Remember, unlike AMD, Intel dual-core requires a new
motherboard/chip-set for each dual-core processor shipped.

Maybe the cure is as simple as adding 820 support in BIOS -
since the 820 implementation is not exactly the same as the 830
and 840. However, Asus seemed to be caught unawares by the
problem and nVidia's denial of a fix.

There is no obvious difference between the 820 and its faster brothers
that would explain the failure. Yes, it uses a lower-value clock
multiplier and does not implement Speed-Step, but that is a
functionally-identical state to a P4-2.8 single-core Prescott.

Does that mean that a single-core 2.8 GHz processor would also fail in
an Intel nF4-SLI board ? Probably not, since the P-D 820 with one core
disabled works fine. ( The nF4-SLI chip-set is supposed to be
compatible with both single and dual-core LGA775 processors )

Anyway, for the moment, until this thick smoke-screen by nVidia and
Intel clears, I would highly recommend holding off purchase of any
nF4-SLI (Intel) motherboard, regardless of processor initially
installed -- single, or dual-core - if you are expecting the MB to
eventually function with a dual-core processor. I notice that
the very fine MSI P4N Diamond MB (with software-switching of
SLI, and embedded Creative 24-bit audio) is currently available
from Newegg @ $229. This board is potentially affected by the 820
problem... no comments from MSI, so far.

John Lewis

As for Asus, they get a "weasel" rating, in terms of letting
the limitation of the P5ND2 be known. The motherboard page with
specs for the board is here:

http://usa.asus.com/prog/spec.asp?m=P5ND2-SLI Deluxe&langs=09

"Supports Enhanced Intel SteepStep Technology(EIST) Intel
dual-core new generation CPUs enabling ( subject to Intel CPU
availability and NVIDIA chipset supportability )"

Would you buy a used car from the person who wrote that ?
Are we supposed to guess at just what "supportability" means ?

The Asus FAQ page has an entry for the 820 vs P5ND2 issue. The
downloadable Asus motherboard manual also has a statement
in the specification section, that the 820 runs in single
core mode only. So, Asus has "confessed" in as invisible
a way as they can manage (no one visits the FAQ page because
it is so crappy and pointless). The CPU support page is completely
silent on the subject of dual core processors for the P5ND2,
which is also peculiar (only the P4-670 3.8GHz processor is
listed?). I don't think the Asus support page is ready for
duals yet (the P5WD2 has no processors listed).

As for the differences between chips, the 820 doesn't have
EIST (it probably does have it, but it is useless, as the
idle speed and full-power speed would both be 2.8GHz, which
means no change between states). Another difference on the
820, is it is the only 05A power spec processor, and the
PRB (processor requirements bit) in the MSR is zero for this
processor. Which should be a "no sweat" issue (BIOS is supposed
to read this bit, and only has to respond in some unexplained
way, when PRB=1).

While there are a pile of errata in the Intel datasheet, there
is nothing in there that is suggestive of a failure mechanism
(doc 30683203):

http://www.intel.com/design/pentiumxe/specupdt/306832.htm

Perhaps you could drop Wesley Fink at Anandtech a line, and
see if he knows whats up.

Paul

 

[John Lewis]

As for Asus, they get a "weasel" rating, in terms of letting
the limitation of the P5ND2 be known. The motherboard page with
specs for the board is here:

http://usa.asus.com/prog/spec.asp?m=P5ND2-SLI Deluxe&langs=09

"Supports Enhanced Intel SteepStep Technology(EIST) Intel
dual-core new generation CPUs enabling ( subject to Intel CPU
availability and NVIDIA chipset supportability )"

Would you buy a used car from the person who wrote that ?
Are we supposed to guess at just what "supportability" means ?

The Asus FAQ page has an entry for the 820 vs P5ND2 issue. The
downloadable Asus motherboard manual also has a statement
in the specification section, that the 820 runs in single
core mode only. So, Asus has "confessed" in as invisible
a way as they can manage (no one visits the FAQ page because
it is so crappy and pointless). The CPU support page is completely
silent on the subject of dual core processors for the P5ND2,
which is also peculiar (only the P4-670 3.8GHz processor is
listed?). I don't think the Asus support page is ready for
duals yet (the P5WD2 has no processors listed).

As for the differences between chips, the 820 doesn't have
EIST (it probably does have it, but it is useless, as the
idle speed and full-power speed would both be 2.8GHz, which
means no change between states). Another difference on the
820, is it is the only 05A power spec processor, and the
PRB (processor requirements bit) in the MSR is zero for this
processor. Which should be a "no sweat" issue (BIOS is supposed
to read this bit, and only has to respond in some unexplained
way, when PRB=1).

While there are a pile of errata in the Intel datasheet, there
is nothing in there that is suggestive of a failure mechanism
(doc 30683203):

http://www.intel.com/design/pentiumxe/specupdt/306832.htm

Perhaps you could drop Wesley Fink at Anandtech a line, and
see if he knows whats up.

Paul

Paul,

Thanks for the truly excellent information. I have sent Wesley an
email as you suggested. I hold Anandtech in very high regard,
so I am optimistic that we shall get the true story very soon. Will
post any reply here, if appropriate. AnandTech may decide to
investigate formally and publish an article on the subject. Wesley's
first reply to me might just be his own initial speculation and
pointless to post here. We don't need to fuel more rumors on the
subject -- we need the facts !


John Lewis

 

[John Lewis]

www.monarchcomputer.com is now taking pre-orders on
a first come, first-serve basis. First shipments due June 16.

All 4 variants of the X2 processor too !!

Unit prices as follows:-

X2 4800+ $999
X2 4600+ $822
X2 4400+ $619
X2 4200+ $579

I would expect Newegg to be about 5-10% lower
(except for the 4800+) once they open up for orders.
Newegg does not take pre-orders.

Looks as if AMD is beating Intel to the RETAIL market
in dual-core. AMD appears to be executing their desktop
(and server) processor strategies with perfection these days,
while Intel is stumbling around in darkness of their own
making.

( Intel dual-core will need a new motherboard, DDR2
a beefier power-supply plus a bunch of extra cooling..... )

Think of all those Socket 939 motherboards waiting to
be upgraded (especially the SLI ones..) -- provided the MB
manufacturers pull their fingers out in offering BIOS upgrades
to support the X2 !!!

For the X2 to work reliably, the Socket 939 MB should
have A64 4000+ or FX-55 power-regulator capability.

I expect initial shipments to be small and heavily over-
subscribed.

John Lewis

 

[NightSky 421]

www.monarchcomputer.com is now taking pre-orders on
a first come, first-serve basis. First shipments due June 16.

All 4 variants of the X2 processor too !!

Unit prices as follows:-

X2 4800+ $999
X2 4600+ $822
X2 4400+ $619
X2 4200+ $579

I won't be buying a new computer until next year, but hopefully these sorts
of processors will have mainstream pricing by then. I'm pretty confident
they will.

 

[John Lewis]

Seems as if big-brothers Intel and M$$ have our true interests at
heart :--

http://www.digitmag.co.uk/news/index.cfm?NewsID=4915

In the case of Intel "pride (arrogance) may come before the fall"
The Pentium-D is already architecturally in deep trouble; why add
this extra burden ?? No doubt depending on Dell's volume and its
technically-naive customer-base to help dig them out for at
least the next year.

John Lewis

 

[Rick]

Seems as if big-brothers Intel and M$$ have our true interests at
heart :--

http://www.digitmag.co.uk/news/index.cfm?NewsID=4915

In the case of Intel "pride (arrogance) may come before the fall"
The Pentium-D is already architecturally in deep trouble; why add
this extra burden ?? No doubt depending on Dell's volume and its
technically-naive customer-base to help dig them out for at
least the next year.

John Lewis

How do the concerns expressed in this article equate to
anyone being a "DRM paranoid"? But in any case thanks
for the link.

 

[Kroagnon]

Seems as if big-brothers Intel and M$$ have our true interests at
heart :--

http://www.digitmag.co.uk/news/index.cfm?NewsID=4915

In the case of Intel "pride (arrogance) may come before the fall"
The Pentium-D is already architecturally in deep trouble; why add
this extra burden ?? No doubt depending on Dell's volume and its
technically-naive customer-base to help dig them out for at
least the next year.

John Lewis

Hyperthreading is the only thing making me still buy Intel P4s for server
use; from now on it's Athlon 64 x2 for me (when price comes down).

 

[John Lewis]

Hyperthreading is the only thing making me still buy Intel P4s for server
use; from now on it's Athlon 64 x2 for me (when price comes down).

Similarly. For lots of excellent reasons other than DRM.

Any so-called "enthusiast" who buys one of the current-generation
Intel dual-core processors (or Intel-based dual-core desktop systems)
should have his/her head examined for embedded sawdust.

John Lewis

( I have only ever built/used systems with Intel processors
up to this point..... )

 

[tlai909]

Besides DRM, what's wrong with Prescott and -D 820?

Not that I care because the only Intel I've paid for is PIII and
Pentium-M...

T.

 

[John Lewis]

Besides DRM, what's wrong with Prescott and -D 820?

Not that I care because the only Intel I've paid for is PIII and
Pentium-M...

How would you like an inefficient room-warmer ? For a start.

See Anandtech and Xbit Labs articles on dual-core
processors. And if you want to really get into
it technically.. read the recent Xbit Labs articles on Intel
Netburst. You are lucky to have PIII and Pentium-M.
Both excellent architectures.

John Lewis

 

[Conor]

Besides DRM, what's wrong with Prescott and -D 820?

The fact that you'll not be able to get it running at full whack for
long periods because it'll overheat and slow itself down.

 

[Bob]

The fact that you'll not be able to get it running at full whack for
long periods because it'll overheat and slow itself down.

My son runs a 3.2 GHz P4 Prescott with a Zalman copper heat sink

http://www.directron.com/cnps7700cu.html

and the CPU temp stays below 40C. The retail box cooler was a joke.

 

[Conor]

My son runs a 3.2 GHz P4 Prescott with a Zalman copper heat sink

http://www.directron.com/cnps7700cu.html

and the CPU temp stays below 40C. The retail box cooler was a joke.

Very good. Now try running 3DMArk05 CPU test in a constant loop and see
how long it is before the speed drops.

 

[John Lewis]

Very good. Now try running 3DMArk05 CPU test in a constant loop and see
how long it is before the speed drops.

With a Zalman Cu, the 3.2GHz P4 Prescott should be fine.
However try with the same heatsink and running 2-thread CPU stress
on the Pentium-D 820......slow-down pretty-well guaranteed without
additional forced-air cooling of the heatsink. Intel, in their great
big-brother wisdom even deliberately left the power-saving SpeedStep
out of the P-D 820. Must be afraid that people will try to overclock
their least expensive offering instead of buying the higher members
of the P-D family.

Intel's current bumblings and arrogant big-brother activities with
their desktop processors deserve reward. I hope AMD takes a big
chunk of this business. Dell will continue to bale Intel out no doubt,

although Michael must be getting very uncomfortable --- especially
on the rapidly-growing server side of Dell's business, where the
Opteron is now the obvious choice from a performance/power
point-of-view.

John Lewis

 

[Bob]

Very good. Now try running 3DMArk05 CPU test in a constant loop and see
how long it is before the speed drops.

If he pins the CPU at 100% (according to Task Manager) when he rips
DVDs, why would running a CPU stress utility be any different?

 

[Bob]

Intel's current bumblings and arrogant big-brother activities with
their desktop processors deserve reward.

Has anyone consulted the Patriot Act to see if there is a requirement
for this mandated by the federal govt.

Actually the govt doesn't need any Act to do whatever it wants. For
example Congress passed the so-called Brady Law which was responsible
for "Instant Checks" on a prospective gun buyer's background. The law
stated explicitly that the govt (in this case the FBI) could keep the
information about the buyer for 48 hours then it had to destroy it.

Fast forward a year or so and we discover that the FBI is blatantly
breaking the law by keeping data on gun buyers processed by the
Instant Check system. So the NRA takes the FBI to court and the case
works its way up to the US Supreme Court.

The US SC ruled that the FBI could continue to their practive of
keeping information even though it is explicitly outlawed by Congress.
That's because it is the "right" of the FBI to do whatever it damn
pleases. Nevermind it completely ignored all the warnings about 911,
including a delegation of intelligence people working for the Massoud
who came to Washington personally to show the FBI its intelligence
gathering. No one would see them when they arrived.

The point of all this is there is ample precedent for the govt acting
like a fascist dictatorship with the full blessing of the highest
court in the land. And Intel is just doing its part to fight
terrorism. HA!

Anyone remember Richard Jewell? Or Ruby Ridge? How about the Waco
Massacre? Can you say "Seig Heil!". There, I knew you could.

Intel: "Big Brother Inside!"

 

[Kroagnon]

Besides DRM, what's wrong with Prescott and -D 820?

Not that I care because the only Intel I've paid for is PIII and
Pentium-M...

Heat and power consumption on the Prescott.

As for the P-D there are similar heat and power consumption issues and I
believe that it also lacks x64 capability.

 

[Bob]

Since AMD is also an American company and they AREN'T doing this, i'd
say you're way off base here.

Intel is fully capable of shooting off their feet without help from the
government.

There is a lot more to this than a company shooting off its feet.

 

[John Lewis]

Heat and power consumption on the Prescott.

As for the P-D there are similar heat and power consumption issues

and I
believe that it also lacks x64 capability.

Nope, you are thinking of the even-numbered CPUs in the 5xx series
540,550 etc... Some odd-numbered ones are coming with 64-bit
enabled.... and people said that AMD's processor numbering scheme
is screwed up ? Nothing in comparison to Intel's present mess.
What other stupid company would give a Pentium-D (dual-core with no
HT) and the Pentium Extreme (dual-core with HT) exactly the same
number --- 840.

The 6xx and 8xx have EM64T enabled. How well does Intel's
64-bit hardware work ? Nobody has yet done a side-by-side
comparison of Intel's implementation with either the A64
RevE (with SSE3) or the X2.


John Lewis