PC Motherboard Chipsets and Parts Vendors

Re: PC Motherboard Chipsets and Parts Vendors

In article <691c6aF2v01nrU1@mid.individual . net >,
"Richard Crowley" <rcrowley@xp7rt . net > wrote:

> "To alleviate this problem, Intel has proposed a cosmic ray
> detector which could be integrated into future high-density
> microprocessors, allowing the processor to repeat the last
> command following a cosmic ray event."
> * en.wikipedia.org/wiki/Cosmic_ray

That sounds like a hell of a trick. How do you ensure that the cosmic
ray that passes in a random direction through your RAM also passes
through your detector (unless your detector encompasses the entire
computer)?

Re: PC Motherboard Chipsets and Parts Vendors


"Ralph Barone" <invalid@not_real.ca> wrote in message
news:invalid-68FD4F.21014414052008@shawnews...
> In article <691c6aF2v01nrU1@mid.individual . net >,
> "Richard Crowley" <rcrowley@xp7rt . net > wrote:
>
>> "To alleviate this problem, Intel has proposed a cosmic ray
>> detector which could be integrated into future high-density
>> microprocessors, allowing the processor to repeat the last
>> command following a cosmic ray event."
>> * en.wikipedia.org/wiki/Cosmic_ray
>
> That sounds like a hell of a trick. How do you ensure that the cosmic
> ray that passes in a random direction through your RAM also passes
> through your detector (unless your detector encompasses the entire
> computer)?

I was wondering about that. The detector may actually be part of a chip
substrate. If the substrate is hit anywhere, I'm thinking, it could generate
a measurable event. Solid state radiation detectors do exist. One detector
per chip.

Bob Morein
(310) 237-6511

Re: PC Motherboard Chipsets and Parts Vendors

Soundhaspriority wrote:
> "Ralph Barone" <invalid@not real.ca> wrote in message
> news:invalid-68FD4F.21014414052008@shawnews...
>> In article <691c6aF2v01nrU1@mid.individual . net >,
>> "Richard Crowley" <rcrowley@xp7rt . net > wrote:
>>
>>> "To alleviate this problem, Intel has proposed a cosmic ray
>>> detector which could be integrated into future high-density
>>> microprocessors, allowing the processor to repeat the last
>>> command following a cosmic ray event."
>>> * en.wikipedia.org/wiki/Cosmic ray
>> That sounds like a hell of a trick. How do you ensure that the cosmic
>> ray that passes in a random direction through your RAM also passes
>> through your detector (unless your detector encompasses the entire
>> computer)?
>
> I was wondering about that. The detector may actually be part of a chip
> substrate. If the substrate is hit anywhere, I'm thinking, it could generate
> a measurable event. Solid state radiation detectors do exist. One detector
> per chip.
>
> Bob Morein
> (310) 237-6511

Isn't there a Star Trek boxed set waiting for you at amazon . com ?

Re: PC Motherboard Chipsets and Parts Vendors


"Romeo Rondeau" <eveyone@ease . com > wrote in message
news:OMRWj.4492$nW2.3249@nlpi064.nbdc.sbc . com ...
> Soundhaspriority wrote:
>> "Ralph Barone" <invalid@not_real.ca> wrote in message
>> news:invalid-68FD4F.21014414052008@shawnews...
>>> In article <691c6aF2v01nrU1@mid.individual . net >,
>>> "Richard Crowley" <rcrowley@xp7rt . net > wrote:
>>>
>>>> "To alleviate this problem, Intel has proposed a cosmic ray
>>>> detector which could be integrated into future high-density
>>>> microprocessors, allowing the processor to repeat the last
>>>> command following a cosmic ray event."
>>>> * en.wikipedia.org/wiki/Cosmic_ray
>>> That sounds like a hell of a trick. How do you ensure that the cosmic
>>> ray that passes in a random direction through your RAM also passes
>>> through your detector (unless your detector encompasses the entire
>>> computer)?
>>
>> I was wondering about that. The detector may actually be part of a chip
>> substrate. If the substrate is hit anywhere, I'm thinking, it could
>> generate a measurable event. Solid state radiation detectors do exist.
>> One detector per chip.
>>
>> Bob Morein
>> (310) 237-6511
>
> Isn't there a Star Trek boxed set waiting for you at amazon . com ?

The Intel detector is probably a PIN diode. See
* w w w .carroll-ramsey . com /detect.htm

Bob Morein
(310) 237-6511

Re: PC Motherboard Chipsets and Parts Vendors

In article <invalid-68FD4F.21014414052008@shawnews>,
Ralph Barone <invalid@not_real.ca> wrote:
>In article <691c6aF2v01nrU1@mid.individual . net >,
> "Richard Crowley" <rcrowley@xp7rt . net > wrote:
>
>> "To alleviate this problem, Intel has proposed a cosmic ray
>> detector which could be integrated into future high-density
>> microprocessors, allowing the processor to repeat the last
>> command following a cosmic ray event."
>> * en.wikipedia.org/wiki/Cosmic_ray
>
>That sounds like a hell of a trick. How do you ensure that the cosmic
>ray that passes in a random direction through your RAM also passes
>through your detector (unless your detector encompasses the entire
>computer)?

Remember. the chip is flat. Putting a flat detector behind the chip will
detect anything going through the chip other than particles that are
parallel to the plane of the chip, or very close to it. You don't need to
detect where it on the chip it struck, just that it went by.
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

Re: PC Motherboard Chipsets and Parts Vendors

"Richard Crowley" <rcrowley@xp7rt . net > wrote in message
news:691c6aF2v01nrU1@mid.individual . net

> An excellent point. If SER was any kind of significant
> issue for the average computer user, you can be sure that the
> people who market them would jump at the chance to prove
> to customers that the extra $$ was worth it. Instead,
> even people who make ECC admit that it isn't really
> necessary in your average PC.

Here's a real-world study:

* w w w .ece.rochester.edu/~xinli/usenix07/

212 servers with about 4 GB RAM each were monitored for about 90 days.

There were 2 errors.

This is a small enough number of errors that it seems to be statistically
questionable. IOW the same experiment might have found 1 error, or 3 errors
or more or less, when repeated. However 10 errors or more seems to be highly
unlikely

We might then think that 50 servers running for a year would give similar
results - 10 errors or less.

A 4 GB PC might thus run for 5 years (1 to 2 times its useful life) and not
develop any errors the whole time it was turned on.

Re: PC Motherboard Chipsets and Parts Vendors


"Arny Krueger" <arnyk@hotpop . com > wrote in message
news:ef-dnWpMfbKdoLHVnZ2dnUVZ_gydnZ2d@comcast . com ...
> "Richard Crowley" <rcrowley@xp7rt . net > wrote in message
> news:691c6aF2v01nrU1@mid.individual . net
>
>> An excellent point. If SER was any kind of significant
>> issue for the average computer user, you can be sure that the
>> people who market them would jump at the chance to prove
>> to customers that the extra $$ was worth it. Instead,
>> even people who make ECC admit that it isn't really
>> necessary in your average PC.
>
> Here's a real-world study:
>
> * w w w .ece.rochester.edu/~xinli/usenix07/
>
> 212 servers with about 4 GB RAM each were monitored for about 90 days.
>
> There were 2 errors.
>
> This is a small enough number of errors that it seems to be statistically
> questionable. IOW the same experiment might have found 1 error, or 3
> errors or more or less, when repeated. However 10 errors or more seems to
> be highly unlikely
>
> We might then think that 50 servers running for a year would give similar
> results - 10 errors or less.
>
> A 4 GB PC might thus run for 5 years (1 to 2 times its useful life) and
> not develop any errors the whole time it was turned on.
That is reasonable. However, there is another issue, which has to do with
noise modeling. Because of the extremely large number of current patterns in
a motherboard, it's impossible to simulate them all. The same is true with
RAM chips, which is why memory tests work so poorly -- they fail to account
for pattern sensitivity.

The EMI landscape changes constantly. In complexity, it's much like the CPU
test problem -- it's impossible to completely test a CPU for faults. The
board cannot be run through all the patterns, because there are too many.

Server motherboards used buffered ram for two reasons: it allows more ram on
a bus, but it also increases noise margins. Desktop machines don't use
buffered ram. When Microsoft made their "complaint" about hardware, they
fingered ECC, but the ram makers fingered the motherboard makers in turn.
They suggested a "reread" facility, reading the ram again, which, they
stated, would knock out some of the errors caused by the EMI landscape. But
ECC checks the ram bus as well, so unless the noise induced error goes
beyond one bit, it will reduce noise induced error as well.

The paper you cite is extremely interesting. However, I would not directly
transfer the results, except loosely, as you have done, to desktop boards.
Server boards are designed for reliability at the expense of performance, a
choice I happen to prefer for all my computers, because the performance hit
is not large.

Bob Morein
(310) 237-6511

Re: PC Motherboard Chipsets and Parts Vendors

"Soundhaspriority" <nowhere@nowhere . com > wrote in message
news:1fSdnbpD8-C23rHVnZ2dnUVZ_sninZ2d@giganews . com
> "Arny Krueger" <arnyk@hotpop . com > wrote in message
> news:ef-dnWpMfbKdoLHVnZ2dnUVZ_gydnZ2d@comcast . com ...
>> "Richard Crowley" <rcrowley@xp7rt . net > wrote in message
>> news:691c6aF2v01nrU1@mid.individual . net
>>
>>> An excellent point. If SER was any kind of significant
>>> issue for the average computer user, you can be sure
>>> that the people who market them would jump at the
>>> chance to prove to customers that the extra $$ was worth it. Instead,
>>> even people who make ECC admit that it isn't really
>>> necessary in your average PC.
>>
>> Here's a real-world study:
>>
>> * w w w .ece.rochester.edu/~xinli/usenix07/
>>
>> 212 servers with about 4 GB RAM each were monitored for
>> about 90 days. There were 2 errors.
>>
>> This is a small enough number of errors that it seems to
>> be statistically questionable. IOW the same experiment
>> might have found 1 error, or 3 errors or more or less,
>> when repeated. However 10 errors or more seems to be
>> highly unlikely We might then think that 50 servers running for a year
>> would give similar results - 10 errors or less.
>>
>> A 4 GB PC might thus run for 5 years (1 to 2 times its
>> useful life) and not develop any errors the whole time
>> it was turned on.

> That is reasonable. However, there is another issue,
> which has to do with noise modeling. Because of the
> extremely large number of current patterns in a
> motherboard, it's impossible to simulate them all. The
> same is true with RAM chips, which is why memory tests
> work so poorly -- they fail to account for pattern
> sensitivity.

That issue was sddressed by the real-world test.

Look, there's no end of ghosts that you can conjur and chase, particuarly if
you are prone to obsessing.

If you need golden audio components and PCs to be comfortable when you work,
please do so - at your own expense. Just don't get disappointed if a lot of
people with real-world experience tend to look down on you when you try to
share your opinions.

Re: PC Motherboard Chipsets and Parts Vendors


"Arny Krueger" <arnyk@hotpop . com > wrote in message
news:7c2dnb6cmYqm2bHVnZ2dnUVZ_gSdnZ2d@comcast . com ...
> "Soundhaspriority" <nowhere@nowhere . com > wrote in message
> news:1fSdnbpD8-C23rHVnZ2dnUVZ_sninZ2d@giganews . com
>> "Arny Krueger" <arnyk@hotpop . com > wrote in message
>> news:ef-dnWpMfbKdoLHVnZ2dnUVZ_gydnZ2d@comcast . com ...
>>> "Richard Crowley" <rcrowley@xp7rt . net > wrote in message
>>> news:691c6aF2v01nrU1@mid.individual . net
>>>
>>>> An excellent point. If SER was any kind of significant
>>>> issue for the average computer user, you can be sure
>>>> that the people who market them would jump at the
>>>> chance to prove to customers that the extra $$ was worth it. Instead,
>>>> even people who make ECC admit that it isn't really
>>>> necessary in your average PC.
>>>
>>> Here's a real-world study:
>>>
>>> * w w w .ece.rochester.edu/~xinli/usenix07/
>>>
>>> 212 servers with about 4 GB RAM each were monitored for
>>> about 90 days. There were 2 errors.
>>>
>>> This is a small enough number of errors that it seems to
>>> be statistically questionable. IOW the same experiment
>>> might have found 1 error, or 3 errors or more or less,
>>> when repeated. However 10 errors or more seems to be
>>> highly unlikely We might then think that 50 servers running for a year
>>> would give similar results - 10 errors or less.
>>>
>>> A 4 GB PC might thus run for 5 years (1 to 2 times its
>>> useful life) and not develop any errors the whole time
>>> it was turned on.
>
>> That is reasonable. However, there is another issue,
>> which has to do with noise modeling. Because of the
>> extremely large number of current patterns in a
>> motherboard, it's impossible to simulate them all. The
>> same is true with RAM chips, which is why memory tests
>> work so poorly -- they fail to account for pattern
>> sensitivity.
>
> That issue was sddressed by the real-world test.
>
[snip]

No, it wasn't, because it was done on server class motherboards, which are
engineered differently from desktop motherboards.

You're getting sloppy.

Bob Morein
(310) 237-6511

Re: PC Motherboard Chipsets and Parts Vendors

"Soundhaspriority" <nowhere@nowhere . com > wrote in message
news:wN6dnUvvmY3E27HVnZ2dnUVZ_obinZ2d@giganews . com
> "Arny Krueger" <arnyk@hotpop . com > wrote in message
> news:7c2dnb6cmYqm2bHVnZ2dnUVZ_gSdnZ2d@comcast . com ...
>> "Soundhaspriority" <nowhere@nowhere . com > wrote in message
>> news:1fSdnbpD8-C23rHVnZ2dnUVZ_sninZ2d@giganews . com
>>> "Arny Krueger" <arnyk@hotpop . com > wrote in message
>>> news:ef-dnWpMfbKdoLHVnZ2dnUVZ_gydnZ2d@comcast . com ...
>>>> "Richard Crowley" <rcrowley@xp7rt . net > wrote in message
>>>> news:691c6aF2v01nrU1@mid.individual . net
>>>>
>>>>> An excellent point. If SER was any kind of
>>>>> significant issue for the average computer user, you can be sure
>>>>> that the people who market them would jump at the
>>>>> chance to prove to customers that the extra $$ was
>>>>> worth it. Instead, even people who make ECC admit
>>>>> that it isn't really necessary in your average PC.
>>>>
>>>> Here's a real-world study:
>>>>
>>>> * w w w .ece.rochester.edu/~xinli/usenix07/
>>>>
>>>> 212 servers with about 4 GB RAM each were monitored for
>>>> about 90 days. There were 2 errors.
>>>>
>>>> This is a small enough number of errors that it seems
>>>> to be statistically questionable. IOW the same experiment
>>>> might have found 1 error, or 3 errors or more or less,
>>>> when repeated. However 10 errors or more seems to be
>>>> highly unlikely We might then think that 50 servers
>>>> running for a year would give similar results - 10
>>>> errors or less. A 4 GB PC might thus run for 5 years (1 to 2 times its
>>>> useful life) and not develop any errors the whole time
>>>> it was turned on.
>>
>>> That is reasonable. However, there is another issue,
>>> which has to do with noise modeling. Because of the
>>> extremely large number of current patterns in a
>>> motherboard, it's impossible to simulate them all. The
>>> same is true with RAM chips, which is why memory tests
>>> work so poorly -- they fail to account for pattern
>>> sensitivity.
>>
>> That issue was sddressed by the real-world test.

> No, it wasn't, because it was done on server class
> motherboards, which are engineered differently from
> desktop motherboards.

Of course they are engineered differently - they are different. That they
are different does not prove the existence of a difference that is relevant
to RAM soft errors.

Fact of the matter is that high volume parts (e.g. desktop and laptop system
boards) are more gtenerally more carefully engineered because their volume
is much greater, and the financial exposure to the manufacturer is greater.

Re: PC Motherboard Chipsets and Parts Vendors


"Arny Krueger" <arnyk@hotpop . com > wrote in message
news:kLKdnUzIcZy8w7HVnZ2dnUVZ_hCdnZ2d@comcast . com ...
> "Soundhaspriority" <nowhere@nowhere . com > wrote in message
> news:wN6dnUvvmY3E27HVnZ2dnUVZ_obinZ2d@giganews . com
[snip]
>
> Fact of the matter is that high volume parts (e.g. desktop and laptop
> system boards) are more gtenerally more carefully engineered because their
> volume is much greater, and the financial exposure to the manufacturer is
> greater.
>
The above is ludicrous.

Bob Morein
(310) 237-6511

Re: PC Motherboard Chipsets and Parts Vendors

"Soundhaspriority" <nowhere@nowhere . com > wrote:
>
> "Arny Krueger" <arnyk@hotpop . com > wrote in message
> news:kLKdnUzIcZy8w7HVnZ2dnUVZ hCdnZ2d@comcast . com ...
>> "Soundhaspriority" <nowhere@nowhere . com > wrote in message
>> news:wN6dnUvvmY3E27HVnZ2dnUVZ obinZ2d@giganews . com
> [snip]
>>
>> Fact of the matter is that high volume parts (e.g. desktop and laptop
>> system boards) are more gtenerally more carefully engineered because their
>> volume is much greater, and the financial exposure to the manufacturer is
>> greater.
>>
> The above is ludicrous.
>
> Bob Morein
> (310) 237-6511

What a wonderful world it would be if it was true, though...

--
Mats Peterson
* w w w .geocities . com /matsp888/

Re: PC Motherboard Chipsets and Parts Vendors

"Soundhaspriority" <nowhere@nowhere . com > wrote in message
news:2O-dnZLa_fUQ97HVnZ2dnUVZ_srinZ2d@giganews . com
> "Arny Krueger" <arnyk@hotpop . com > wrote in message
> news:kLKdnUzIcZy8w7HVnZ2dnUVZ_hCdnZ2d@comcast . com ...
>> "Soundhaspriority" <nowhere@nowhere . com > wrote in message
>> news:wN6dnUvvmY3E27HVnZ2dnUVZ_obinZ2d@giganews . com
> [snip]

>> Fact of the matter is that high volume parts (e.g.
>> desktop and laptop system boards) are generally
>> more carefully engineered because their volume is much
>> greater, and the financial exposure to the manufacturer
>> is greater.

> The above is ludicrous.

Prove it.

Re: PC Motherboard Chipsets and Parts Vendors


"Mike Rivers" <mrivers@d-and-d . com > wrote in message
news:IqJWj.17747$%X1.9083@trnddc08...
> Soundhaspriority wrote:
>
>> Mike, this reasoning is not valid. We're talking soft errors, which occur
>> when nothing is wrong with the memory module.
>
> If there's nothing wrong with the memory module, then why is there an
> error? If there's an error, there's something wrong with the memory
> module, if only intermittently.
>
> > The causes are cosmic rays and
>> poorly modeled noise.
>
> I don't know what "poorly modeled noise" is, but you're not going to
> convince me that cosmic rays cause computer errors no matter how many web
> sites you find that say they do. Maybe in the lab, but in my house? Inside
> a solid metal case? Should I be wearing a metal helmet?
>
Mike, here's another one for you:
* w w w .us.design-reuse . com /articles/6930/solutions-for-soft-errors-in-system-on-chip-designs.html

"Compared to embedded systems, desktop processors now utilize large,
high-density memories, which significantly increases the vulnerability of
systems to soft error failure. "

Bob Morein
(310) 237-6511

Re: PC Motherboard Chipsets and Parts Vendors

"Soundhaspriority" wrote ...
> "Mike Rivers" wrote ...
>> Soundhaspriority wrote:
>>
>>> Mike, this reasoning is not valid. We're talking soft errors, which
>>> occur when nothing is wrong with the memory module.
>>
>> If there's nothing wrong with the memory module, then why is there an
>> error? If there's an error, there's something wrong with the memory
>> module, if only intermittently.
>>
>> > The causes are cosmic rays and
>>> poorly modeled noise.
>>
>> I don't know what "poorly modeled noise" is, but you're not going to
>> convince me that cosmic rays cause computer errors no matter how many web
>> sites you find that say they do. Maybe in the lab, but in my house?
>> Inside a solid metal case? Should I be wearing a metal helmet?
>>
> Mike, here's another one for you:
> * w w w .us.design-reuse . com /articles/6930/solutions-for-soft-errors-in-system-on-chip-designs.html
>
> "Compared to embedded systems, desktop processors now utilize large,
> high-density memories, which significantly increases the vulnerability of
> systems to soft error failure. "

"Current research suggests that the average rate of failure for complex
chips may be in excess of four errors per year. "

Dunno about you, but I hit the wrong button and delete a whole
file much more often than that.

"In tightly-coupled memories (TCM), creating a software exception
that executes an error-handling code to refresh the TCM is one
approach. However, the error-handling code itself may be subject
to soft errors. In systems where the expectation of SER is relatively
low, it may be acceptable to handle error correction by initiating a
complete system reset."

And costs nothing.

"Although it is difficult to quantify the true threat presented by soft
errors, the industry is agreed that the trend for SER is clearly set
to rise with increasingly smaller process geometries."

Indeed, in 3-5 generations of computers down the road (if I live
that long) I expect that ECC will be commonplace.

"Any solution therefore needs not only to be pragmatic but also
acceptable to manufacturers, in terms of the cost impact on both
area and performance. "

But the cost/benefit tradeoff even then will still be a sticky problem
to deal with.

It would appear that Bob's own citation disputes him. :-)

I still can't figure out why Bob's got his panties in such a
knot over SER & ECC. Did he just buy a bunch of Micron
stock? And he still hasn't explained "poorly modeled noise".