PC Motherboard Chipsets and Parts Vendors

Re: PC Motherboard Chipsets and Parts Vendors

"Soundhaspriority" wrote ...
> "Richard Crowley" wrote in...
>> "Soundhaspriority" wrote ...
>>> Several years ago, Intel removed ECC capabilities from their
>>> chipsets,
>>
>> Reference? What remarkable twaddle.
>
> Rich,
> Please accept my deepest apology. While so far in this
> discussion I have provided the following references:
> * w w w .eetimes . com /news/latest/showArticle.jhtml?articleID=199601761

This is a Microsoft press release picked up by an
industry rag on a slow news day last year.

> w w w .tezzaron . com /about/papers/soft_errors_1_1_secure.pdf

This is a press release promoting the company's products.
Here are some quotes you may have missed....
"Alpha particle problems have been largely eliminated"
"DRAM soft error rates have dropped."
"SRAM error rates are growing."
"SRAM has become more susceptible to soft errors than DRAM""
"SER modeling and simulation are highly complex and
frustratingly inaccurate. In addition, there are no standard
testing or reporting mechanisms for cosmic ray induced SERs."
"One reputable source reported anomalously low rates;"

I must admit that you have given me an increased respect
for the marketing people.

> Now let's try to help each other out on this. One important
> clarification is that a motherboard may incorporate a chipset
> that supports ECC, while the BIOS does not. For several years, there
> were no desktop motherboards on the market past the Pentium IV level
> that supported ECC.

It sounds implausible. Without some hard data, it still
sounds like FUD to me.

Furthermore, is that meant to be an argument FOR ECC
or against it? Seems to me that if we have lived without
ECC for so long without society falling apart, maybe it
isn't as critical as you think?

What kind of ECC are you running in the computer you
are reading this on?

> Steve L. tells us that some Intel products once again support ECC.

Then perhaps you should take that up with Steve L. who
appears to be your expert.

> Since you are an Intel insider, I respectfully request some document
> links so we can establish what Intel desktop motherboard products do
> support ECC.

All Intel motherboards are documented online at:
w w w .intel . com Feel free to reference whatever you wish.
No insider access necessary.

Re: PC Motherboard Chipsets and Parts Vendors


"Richard Crowley" <rcrowley@xp7rt . net > wrote in message
news:o_qdndzm_sIyRrfVnZ2dnUVZ_gudnZ2d@posted.pcez...
> "Soundhaspriority" wrote ...
[snip]
>
>> Since you are an Intel insider, I respectfully request some document
>> links so we can establish what Intel desktop motherboard products do
>> support ECC.
>
> All Intel motherboards are documented online at:
> w w w .intel . com Feel free to reference whatever you wish.
> No insider access necessary.
I found the page,
* w w w .intel . com /support/motherboards/desktop/sb/cs-009023.htm

and posted it where Mike Rivers will see it.

Bob Morein
(310) 237-6511

Re: PC Motherboard Chipsets and Parts Vendors


"Richard Crowley" <rcrowley@xp7rt . net > wrote in message
news:68uj5oF2u8m7mU1@mid.individual . net ...
> "Soundhaspriority" wrote ...
>> You should take that advice yourself. Microsoft has no vested reason
>> to promote the use of ECC,
>
> You must have missed the nine other vested reasons I cited.
>
>> other than to improve the customer experience.
>
> Or FUD.
> * en.wikipedia.org/wiki/Fear%2C_uncertainty_and_doubt
>
>> But the hardware industry doesn't want to charge consumers for ECC
>> protection, because they know it's cheaper to dupe them out of the added
>> reliability.
>
> They also know that the customers won't pay for solid-gold
> keyboards either because it provides no better reliabilty.
>
[snip]

Here are some numbers for you. This paper was revised 2004, the most up to
date source I could find on the web.
w w w .tezzaron . com /about/papers/soft_errors_1_1_secure.pdf

The Micron estimate in the table is 2-4 errors per year per 256 mb. In a
computer with 4 gigabytes of ram, this translates to 32-64 errors in memory,
per year, or approximately:

ONE PER WEEK.

Please consult your keychain before you reply.

Regards,
Bob Morein
(310) 237-6511

Re: PC Motherboard Chipsets and Parts Vendors

"Soundhaspriority" <nowhere@nowhere . com > wrote in message
news:otudnSpZGo7Vu7fVnZ2dnUVZ_g2dnZ2d@giganews . com
> "Richard Crowley" <rcrowley@xp7rt . net > wrote in message
> news:68uj5oF2u8m7mU1@mid.individual . net ...
>> "Soundhaspriority" wrote ...
>>> You should take that advice yourself. Microsoft has
>>> no vested reason to promote the use of ECC,
>>
>> You must have missed the nine other vested reasons I
>> cited.
>>> other than to improve the customer experience.
>>
>> Or FUD.
>> * en.wikipedia.org/wiki/Fear%2C_uncertainty_and_doubt
>>
>>> But the hardware industry doesn't want to charge
>>> consumers for ECC protection, because they know it's
>>> cheaper to dupe them out of the added reliability.
>>
>> They also know that the customers won't pay for
>> solid-gold keyboards either because it provides no
>> better reliabilty.
> [snip]
>
> Here are some numbers for you. This paper was revised
> 2004, the most up to date source I could find on the web.
> w w w .tezzaron . com /about/papers/soft_errors_1_1_secure.pdf

This paper says that the most common irreducable source of soft errors is
cosmic rays:

"SER* modeling and simulation are highly complex and frustratingly
inaccurate [2][14][16][32]. In addition, there are no standard testing or
reporting mechanisms for cosmic ray induced SERs [3][5][19][22] and most
manufacturers don’t reveal their test results [5][23], so comparisons and
conclusions are difficult to come by. The measurements and estimates in the
following table were drawn from a variety of documents."

> The Micron estimate in the table is 2-4 errors per year
> per 256 mb. In a computer with 4 gigabytes of ram, this
> translates to 32-64 errors in memory, per year, or
> approximately:

> ONE PER WEEK.


The error here is that this old paper is based on the areal density of bits
on Micron DDR RAM chips in 2004, which was 4 years ago.

Larger objects are more frequently hit by cosmic rays, smaller objects are
hit less frequently. As RAM chips become more dense, the cells become
proportionately smaller.

RAM chips have become about 4 times as dense since 2004, so the results in
the paper are pessimistic by a factor of about 4.

IOW, the error rate in a computer with 4 GB of 2008 DDR RAM is more like

ONE PER MONTH.

Re: PC Motherboard Chipsets and Parts Vendors


"Arny Krueger" <arnyk@hotpop . com > wrote in message
news:OOydnVMF7eeWrLfVnZ2dnUVZ_qTinZ2d@comcast . com ...
> "Soundhaspriority" <nowhere@nowhere . com > wrote in message
> news:otudnSpZGo7Vu7fVnZ2dnUVZ_g2dnZ2d@giganews . com
>> "Richard Crowley" <rcrowley@xp7rt . net > wrote in message
>> news:68uj5oF2u8m7mU1@mid.individual . net ...
>>> "Soundhaspriority" wrote ...
>>>> You should take that advice yourself. Microsoft has
>>>> no vested reason to promote the use of ECC,
>>>
>>> You must have missed the nine other vested reasons I
>>> cited.
>>>> other than to improve the customer experience.
>>>
>>> Or FUD.
>>> * en.wikipedia.org/wiki/Fear%2C_uncertainty_and_doubt
>>>
>>>> But the hardware industry doesn't want to charge
>>>> consumers for ECC protection, because they know it's
>>>> cheaper to dupe them out of the added reliability.
>>>
>>> They also know that the customers won't pay for
>>> solid-gold keyboards either because it provides no
>>> better reliabilty.
>> [snip]
>>
>> Here are some numbers for you. This paper was revised
>> 2004, the most up to date source I could find on the web.
>> w w w .tezzaron . com /about/papers/soft_errors_1_1_secure.pdf
>
> This paper says that the most common irreducable source of soft errors is
> cosmic rays:
>
> "SER* modeling and simulation are highly complex and frustratingly
> inaccurate [2][14][16][32]. In addition, there are no standard testing or
> reporting mechanisms for cosmic ray induced SERs [3][5][19][22] and most
> manufacturers don't reveal their test results [5][23], so comparisons and
> conclusions are difficult to come by. The measurements and estimates in
> the following table were drawn from a variety of documents."
>
>> The Micron estimate in the table is 2-4 errors per year
>> per 256 mb. In a computer with 4 gigabytes of ram, this
>> translates to 32-64 errors in memory, per year, or
>> approximately:
>
>> ONE PER WEEK.
>
>
> The error here is that this old paper is based on the areal density of
> bits on Micron DDR RAM chips in 2004, which was 4 years ago.
>
> Larger objects are more frequently hit by cosmic rays, smaller objects are
> hit less frequently. As RAM chips become more dense, the cells become
> proportionately smaller.
>
> RAM chips have become about 4 times as dense since 2004, so the results in
> the paper are pessimistic by a factor of about 4.
>
> IOW, the error rate in a computer with 4 GB of 2008 DDR RAM is more like
>
> ONE PER MONTH.
That might be true. It is certainly within reasonable confidence limits. Or,
other factors could intervene. There are other sources of error besides
radiation. Board noise has beome so complex, it is modeled statistically.
There has also been concern that DDR2 buses have reduced noise immunity
versus DDR1. ECC correction is done by the CPU. Thus, single bit bus
errors are caught as well.

Bob Morein
(310) 237-6511

Re: PC Motherboard Chipsets and Parts Vendors

Arny Krueger wrote:

> IOW, the error rate in a computer with 4 GB of 2008 DDR RAM is more like
> ONE PER MONTH.

What's the symptom or consequence of these soft errors? How would I know
if I had one? My computers don't crash, so what else happens?

--
If you e-mail me and it bounces, use your secret decoder ring and reach
me here:
double-m-eleven-double-zero at yahoo -- I'm really Mike Rivers
(mriv...@d-and-d . com )

Re: PC Motherboard Chipsets and Parts Vendors

On May 14, 8:51 am, Mike Rivers <mriv...@d-and-d . com > wrote:
> Arny Krueger wrote:
> > IOW, the error rate in a computer with 4 GB of 2008 DDR RAM is more like
> > ONE PER MONTH.
>
> What's the symptom or consequence of these soft errors? How would I know
> if I had one? My computers don't crash, so what else happens?
>
> --
> If you e-mail me and it bounces, use your secret decoder ring and reach
> me here:
> double-m-eleven-double-zero at yahoo -- I'm really Mike Rivers
> (mriv...@d-and-d . com )

I have had my computer for about 4 months and I think it crashed once,
and not because of RAM.
I have 4 GB of RAM, and while what you might say might be true (The
Micron estimate in the table is 2-4 errors per year per 256 mb.) , I'm
pretty much sure it's not entirely.

See, if you have 2 errors per YEAR on 256mb, you don't have one error
per month on 4gb since I doubt your 4gb will be all used at the said
time anyway.
What I'm trying to say is: If your computer is running 365/365 days,
I'd take ECC, else just get more Ram (not more than 4gb on a 32bit
system - ie WinXP or Vista 32)

Re: PC Motherboard Chipsets and Parts Vendors

"Mike Rivers" wrote in ...
> Arny Krueger wrote:
>
>> IOW, the error rate in a computer with 4 GB of 2008 DDR RAM is more
>> like
>> ONE PER MONTH.
>
> What's the symptom or consequence of these soft errors? How would I
> know if I had one? My computers don't crash, so what else happens?

It gives Microsoft something else to blame (besides their
buggy code). In 10s of thousands of computer-hours,
I can't remember *ever* having a problem that couldn't
be attributed to something more likely than SER.

If SER really is the problem that Bob thinks it is, I'm more
likely to go out and buy stock in Micron, Terrazon, et.al.
than any ECC memory from them. I would expect much
better return on my investment.

Re: PC Motherboard Chipsets and Parts Vendors

On May 14, 9:02 am, "Richard Crowley" <rcrow...@xp7rt . net > wrote:
> "Mike Rivers" wrote in ...
>
> > Arny Krueger wrote:
>
> >> IOW, the error rate in a computer with 4 GB of 2008 DDR RAM is more
> >> like
> >> ONE PER MONTH.
>
> > What's the symptom or consequence of these soft errors? How would I
> > know if I had one? My computers don't crash, so what else happens?
>
> It gives Microsoft something else to blame (besides their
> buggy code). In 10s of thousands of computer-hours,
> I can't remember *ever* having a problem that couldn't
> be attributed to something more likely than SER.
>
> If SER really is the problem that Bob thinks it is, I'm more
> likely to go out and buy stock in Micron, Terrazon, et.al.
> than any ECC memory from them. I would expect much
> better return on my investment.

* searchwincomputing.techtarget . com /tip/0,289483,sid68 gci1251848,00.html
Nice read.

Re: PC Motherboard Chipsets and Parts Vendors


"Mike Rivers" <mrivers@d-and-d . com > wrote in message
news:BfBWj.311$LL.90@trnddc04...
> Arny Krueger wrote:
>
>> IOW, the error rate in a computer with 4 GB of 2008 DDR RAM is more like
>> ONE PER MONTH.
>
> What's the symptom or consequence of these soft errors? How would I know
> if I had one? My computers don't crash, so what else happens?
>
It all depends upon what bit is hit. Here are some of the possibilities:

1. The affected bit is written back to before it is read. In this case,
since the erroneous data was never read, there is no consequence. Same if
the memory location is not in use.

When erroneous data is written out to files, it causes a persistent error:

2. In the body of a data file: a video or audio artifact. It could be a
miscolored pixel, a click, or inaudible. Same for a "data page" in ram.

3. In the header area of a data file: If the metadata is changed, it could
change the way the file is interpreted by a computer program. If it hits a
metadata tag that isn't being used, such as timecode when you're not using
timecode, there is no consequence.

4. In a code page, ie., executable code: A program crash, abnormal
termination, slightly changed behavior, or nothing at all.

5. In the Windows kernel: This is the most serious case of all.

a. If the memory management tables are corrupted, BSOD.
b. If the ram image of the file system is corrupted, it results in a
read error or a write error. If it's a read error of a user process, it
usually causes an application to misbehave or crash.
c. If it's a write error, it can cause a write to the wrong area of
disk. The worst consequence is erasure or corruption of critical operating
files, requiring a complete reinstallation.

The next logical question is, what are the chances of something significant
happening to you? Most consumers work on simple projects of short scope.
They do their Word, Excel, etc. Anything that happens to that data will be
pretty visible to them. On the other hand, video and audio editors rely on
complete integrity of their projects over long periods of time. This is
particularly so with video -- huge amounts of data, project durations of
weeks or months, complex process flows that are extremely inconvenient to
correct. The leader video editing, Avid, will not qualify a workstation
without ECC. All Mac Pros have ECC as well.

My personal feeling is that if I'm affected by a soft error once a year,
it's too much. But one could argue that unless your computer is connected to
a UPS -- mine are -- you are not at that level of concern. Just remember
that building in reliability will save you something, sometime, even if it's
years down the road.

Bob Morein
(310) 237-6511

Re: PC Motherboard Chipsets and Parts Vendors

"Mike Rivers" <mrivers@d-and-d . com > wrote in message
news:BfBWj.311$LL.90@trnddc04
> Arny Krueger wrote:
>
>> IOW, the error rate in a computer with 4 GB of 2008 DDR
>> RAM is more like ONE PER MONTH.

> What's the symptom or consequence of these soft errors?

The consequences are that one randomly-chosen bit in RAM changes value.

> How would I know if I had one?

Depends where that bit is. If it is in a running program or the operating
system, the program or the operating system might crash.

> My computers don't crash, so what else happens?

Some data some place might get changed. Or, the bit may be in a place that
doesn't matter.

For example, if the changed bit is in memory that is note being used or
going to be discarded or refreshed before the next use, then the change will
be wiped out before it can matter.

My computers don't crash, either.

Well OK, this desktop machine I'm typing on used to get a little goofy
(usually could be straightened out by logging off and logging back on) every
month or two. But it just got a new motherboard/CPU/RAM upgrade and it
hasn't yet been a month since the last boot.

I have 3 computers in routine use, one with 2 GB of DDR2-800 RAM and the
other two with 4 GB. Their CPUs are 3000+ single core, and 4000+ & 6000+
dual core. Every month or two one might get a little goofy and get rebooted
when it has nothing else going on.

Re: PC Motherboard Chipsets and Parts Vendors

Arny Krueger wrote:

> I have 3 computers in routine use, one with 2 GB of DDR2-800 RAM and the
> other two with 4 GB. Their CPUs are 3000+ single core, and 4000+ & 6000+
> dual core. Every month or two one might get a little goofy and get rebooted
> when it has nothing else going on.

What's chances that "getting goofy" is a result of a memory error?
Generally we blame Windows, but maybe Windows is really OK and all of
its quirks are a result of memory errors. Maybe Microsoft is right?


--
If you e-mail me and it bounces, use your secret decoder ring and reach
me here:
double-m-eleven-double-zero at yahoo -- I'm really Mike Rivers
(mriv...@d-and-d . com )

Re: PC Motherboard Chipsets and Parts Vendors

Mike Rivers wrote:
> Arny Krueger wrote:
>
>> I have 3 computers in routine use, one with 2 GB of DDR2-800 RAM and
>> the other two with 4 GB. Their CPUs are 3000+ single core, and 4000+ &
>> 6000+ dual core. Every month or two one might get a little goofy and
>> get rebooted when it has nothing else going on.
>
> What's chances that "getting goofy" is a result of a memory error?
> Generally we blame Windows, but maybe Windows is really OK and all of
> its quirks are a result of memory errors. Maybe Microsoft is right?
>

Normally blaming Windows is the right thing to do, but we should not
totally exclude other errors., e.g. in hardware.

--
Chel van Gennip (chel vangennip nl)
Visit Serg van Gennip's site * w w w .serg.vangennip . com

Re: PC Motherboard Chipsets and Parts Vendors

Chel van Gennip wrote:

> Normally blaming Windows is the right thing to do, but we should not
> totally exclude other errors., e.g. in hardware.

I put a certain amount of stock in the principle that software doesn't
crash, and that computer crashes are due to hardware failures. But I
know that programs really DO crash, but I wouldn't expect an operating
system to do so under normal circumstances.


--
If you e-mail me and it bounces, use your secret decoder ring and reach
me here:
double-m-eleven-double-zero at yahoo -- I'm really Mike Rivers
(mriv...@d-and-d . com )

Re: PC Motherboard Chipsets and Parts Vendors


"Mike Rivers" <mrivers@d-and-d . com > wrote in message
news:CnLWj.1515$LL.1134@trnddc04...
> Chel van Gennip wrote:
>
>> Normally blaming Windows is the right thing to do, but we should not
>> totally exclude other errors., e.g. in hardware.
>
> I put a certain amount of stock in the principle that software doesn't
> crash, and that computer crashes are due to hardware failures. But I know
> that programs really DO crash, but I wouldn't expect an operating system
> to do so under normal circumstances.
>
>
> --
> If you e-mail me and it bounces, use your secret decoder ring and reach me
> here:
> double-m-eleven-double-zero at yahoo -- I'm really Mike Rivers
> (mriv...@d-and-d . com )

There are so many shades of gray in this.

Operating systems do crash, unfortunately. But whether they do depends upon
the pattern of use. Afew years ago, the VPN (virtual private network) built
into the Windows kernel was crash vulnerable. Windows offers many kernel
level services, some of them very obscure, or rarely used. The less a
service is used, and the more scenarios it has to cover, the more likely it
is to cause problems. The answer to it is not to use obscure services while
editing.

A bad driver can do it. Unlike the Mac, which is based on Carnegie Mellon's
Mach microkernel, all the Windows drivers and kernel services live in the
same process space. A single bad driver can step on a lot of toes.

But the parts of the OS that are used by everyone all the time have really
been pounded on. If you're running a DAW application, with good drivers, you
won't be using a rarely used kernel service. Memory leaks from application
programs are frequent problems, but you won't be running a bad version of
Firefox while you're editing.

In a typical DAW scenario, the contribution of Windows to reliability issues
is probably zero. But in the past year, Steinberg has issued patches for
crashes.

In my opinion, in which I differ with Rich, the only logical course is to
make the best choice one can in every relevant way. Spending $50 for ECC is
one step. Apply KISS to use of the computer while editing. Keep an eye out
for application patches and crash reports. It all comes together or it don't
but, IMHO, it's the only way to play the game.

Bob Morein
(310) 237-6511