Advice request for building a Xeon E3-1275 ECC-based system from ASUSP8B WS motherboard

Castor Nageur wrote:
On 26 août, 20:26, Castor Nageur wrote:

If you ask on the 2cpu.com forums, they may be able to tell you of
other sources of the Tyan motherboards.

I posted my request but it seems that nobody could help me on the 2cpu
forums.
Anyway, I succeeded in contacting the French company LDLC.
Here is the product's reference:

http://www.ldlc.com/fiche/PB00115676.html

In facts, they never store this motherboard because the demand is very
low. They just order when you order and you have to wait for a minimum
of 2 weeks before can get it.
The European demand seems to be very low for that kind of product. The
hotliner told me that they only had one ordered in 6 months.

Moreover, he explains me that it is difficult for them to order Tyan
products (I suppose this is because of the low demand again !).


If you have a lot of money to waste, you could investigate one of these.
I notice it uses C206, has ECC, uses Xeon. You'll pay a significant
premium for the privilege of owning one.

http://www.anandtech.com/show/4668/h...amped-quarters

Paul

On 31 août, 16:35, Paul wrote:

If you have a lot of money to waste, you could investigate one of these.

Unfortunately, I do not have a lot of money and that's why I want to
build a reliable but cheap configuration using ECC memory. That's also
why I want to build my system myself.

You wrote this when I asked about the ASUS board:

You can check the download page, and see if a new BIOS is available.
If the BIOS has release notes, it may even indicate if changes
have been made in the ECC area.

I downloaded the Tyan S5512 motherboard manual and read the BIOS
section.
Unfortunately and like it is with ASUS, there is no ECC entry in the
Tyan BIOS.

* I know Tyan is a very reliable brand (ASUS is quite reliable too) so
how can you be sure that the Tyan S5512 board really uses your ECC
memory feature without a BIOS notification ?

* Does Tyan provide some diagnostic tools which can simulate/detect
ECC errors ? Could MemTest86+ detect ECC is working ?

Castor Nageur wrote:
On 31 août, 16:35, Paul wrote:

If you have a lot of money to waste, you could investigate one of these.

Unfortunately, I do not have a lot of money and that's why I want to
build a reliable but cheap configuration using ECC memory. That's also
why I want to build my system myself.

You wrote this when I asked about the ASUS board:

You can check the download page, and see if a new BIOS is available.
If the BIOS has release notes, it may even indicate if changes
have been made in the ECC area.

I downloaded the Tyan S5512 motherboard manual and read the BIOS
section.
Unfortunately and like it is with ASUS, there is no ECC entry in the
Tyan BIOS.

* I know Tyan is a very reliable brand (ASUS is quite reliable too) so
how can you be sure that the Tyan S5512 board really uses your ECC
memory feature without a BIOS notification ?

* Does Tyan provide some diagnostic tools which can simulate/detect
ECC errors ? Could MemTest86+ detect ECC is working ?




In a quick search, I can see AMD offers an ECC error injection tool for
AMD processors. I don't see an equivalent function offered by Intel
for their processors. Perhaps I'm using the wrong search terms.

(An example showing AMD cares...)
http://wwwd.amd.com/AMD/SReleaseF.nsf/85a5bd877288a89c87256f330051687b/0a089488bf07da6b87257825005c0eb7/$FILE/ECC_MEM_TOOL_1_0_3.ZIP

You could investigate how Linux handles ECC, and whether Linux has
a utility or monitor that allows error injection. You could do a quick
Linux install, load a package used for EDAC monitoring or testing,
and test that way. Linux gives an alternative to the regular
Windows ecosystem, and their idea of a solution might be better
than the lack of solution I see from Intel. I haven't investigated
all the details, but at least someone in the Linux world thought about
the problem.

Obviously, for error injection to work, there has to be an injection
function in the processor, in some register. At work, we had a department
that did nothing but add mechanical injection points (solder wires onto
hardware and then test whether faults were detected or not), but nobody
wants to do that on hardware they expect to use for a purpose. The people
doing that testing, generally made a mess of the hardware they were given
(as a lot of soldering was required).

I took a look through the E3 processor datasheets (volume 1 and volume 2)
and can't find an error injection function. I searched on the word ECC
and it didn't seem to be discussed. It has to be there, somewhere...
Maybe it just isn't documented.

As for the topic of ECC in general, it's poorly handled. Any time this
kind of discussion comes up, the same problems arise - little visible
feedback, poor tool support, poor documentation, almost like ECC
isn't important. As a user, you have to be pretty motivated, to
spend the time verifying it all works.

Paul

On 2 sep, 19:01, Paul wrote:

Again, thanks very much for your precious advices.

(An example showing AMD cares...)http://wwwd.amd.com/AMD/SReleaseF.ns...7256f330051687...

Yes, I can see that AMD cares about ECC.
Moreover, it seems that nearly all the recent AMD CPU handle ECC
memory.

You could investigate how Linux handles ECC, and whether Linux has
a utility or monitor that allows error injection. You could do a quick
Linux install, load a package used for EDAC monitoring or testing,

I investigated and did not find any EDAC support for the C20x chipset
(Bromolow platform) in their wiki whereas most recent AMD platforms
are supported by EDAC.

I took a look through the E3 processor datasheets (volume 1 and volume 2)
and can't find an error injection function.

So I finally decided not to go for the Tyan S5512 motherboard : it's
very hard & long (& expensive) for me to get it and AMD which is
cheaper supports ECC much better than Intel.
ECC is the most important thing for me : I already had the problem 2
times and I had a lot of corrupted data each time.

I will probably go for a system based on an AMD Phenom II X6 1100T
processor.