To make a NAS / server I am looking for a Raspberry Pi with ECC support.

I gather that the Pi 5 and the CM 5 use RAM which has internal ECC.
That's good. But do they inform the OS of when there are corrections and uncorrectable errors?

Have you guys any idea as to whether the OS running on the Pi5 or CM5
will be able to tell whether RAM has had errors or not?

Feedback welcome!


--
James Harris



--- PyGate Linux v1.5.1
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

James Harris wrote:

That's good. But do they inform the OS of when there are corrections
and uncorrectable errors?

Not that I 'rely' on the tools I use...

gglAIov:

The Raspberry Pi Compute Module 5 (CM5) supports on-die ECC (Error
Correction Code), which silently corrects single-bit errors but does
not report them to the operating system. This is different from
traditional ECC, which would also report errors. The Pi5 does not
support traditional ECC with OS-level error reporting, so while
errors are corrected, the system won't be alerted to potential
issues with the RAM

--
Mike Easter

--- PyGate Linux v1.5.1
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

On 01/12/2025 22:43, Mike Easter wrote:

James Harris wrote:

That's good. But do they inform the OS of when there are corrections
and uncorrectable errors?

Not that I 'rely' on the tools I use...

gglAIov:

The Raspberry Pi Compute Module 5 (CM5) supports on-die ECC (Error
Correction Code), which silently corrects single-bit errors but does
not report them to the operating system. This is different from
traditional ECC, which would also report errors. The Pi5 does not
support traditional ECC with OS-level error reporting, so while
errors are corrected, the system won't be alerted to potential
issues with the RAM

I take it you mean that that's from an AI engine and they are not
reliable. If so, agreed in spades. FWIW the following is where Grok told
me that on CM5 ECC was integrated with the OS. It's quite categoric. But
is it true?

--- Start ---

"Unlike the standard Raspberry Pi 5, the Compute Module 5 exposes the
ECC side-band signals and the BCM2712 RAS controller, so the Linux EDAC
driver can report and correct errors." - Raspberry Pi Forums Thread
(November 2024): James Adams (Head of Compute Module Engineering).

...

Jeff Geerling's Testing (Blog/Video, December 2024): He verified this
hands-on by injecting errors and logging corrections with the
brcmstb_edac driver on a CM5 prototype.

...

Linux Kernel Documentation: The merged EDAC driver (Linux 6.12+)
explicitly supports BCM2712 ECC on CM5 (but not standard Pi 5), with
details on how it reads RAS registers for error counts.... (search for "brcmstb_edac" commits).

--- End ---

As you can see, it does say clearly that Linux can handle ECC reports on
the CM5 but thus far I've not found any sources to back it up.


--
James Harris



--- PyGate Linux v1.5.1
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

Anssi Saari <anssi.saari@usenet.mail.kapsi.fi> wrote:

James Harris <james.harris.1@gmail.com> writes:

As you can see, it does say clearly that Linux can handle ECC reports
on the CM5 but thus far I've not found any sources to back it up.

Why not ask your chosen AI tool to simply provide links to sources and
verify yourself? With a quick search I didn't find support for any of
this ECC support stuff either, except for the standard on-die ECC.

It's funny how the AI provides direct text quotes from specific posts and Youtube videos, which would be trivial to search for. But when you do so, nothing is found. It's almost like they don't exist.

Theo

--- PyGate Linux v1.5.1
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

James Harris <james.harris.1@gmail.com> writes:

As you can see, it does say clearly that Linux can handle ECC reports
on the CM5 but thus far I've not found any sources to back it up.

Why not ask your chosen AI tool to simply provide links to sources and
verify yourself? With a quick search I didn't find support for any of
this ECC support stuff either, except for the standard on-die ECC.

--- PyGate Linux v1.5.1
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

On 03/12/2025 14:18, druck wrote:

On 01/12/2025 21:16, James Harris wrote:

To make a NAS / server I am looking for a Raspberry Pi with ECC support.

I gather that the Pi 5 and the CM 5 use RAM which has internal ECC.
That's good. But do they inform the OS of when there are corrections
and uncorrectable errors?

Have you guys any idea as to whether the OS running on the Pi5 or CM5
will be able to tell whether RAM has had errors or not?

As far as I know it's only for error correction within the RAM
controller, and not reported to the OS. There is a possibility the SOC
might know about it via some proprietary Broadcom mechanism.

That is my impression too, based on admittedly sketchy data.
Lets face it at that level, ecc will simply make slightly dodgy ram rock
solid and merely postpone the day when there is a kernel panic due to corrupted RAM.

I cant see than knowing about it helps much.

I have found a webpage where some idiots attempting to fry a Pi 5 by
running AI on it without a cooler got a fatal ECC error, but my money is
on they just hallucinated the entire thing.

Well indeed.

The internet may be made of cats, but these days its full of
(commercially sponsored and politically motivated) lies.



--
The higher up the mountainside
The greener grows the grass.
The higher up the monkey climbs
The more he shows his arse.

Traditional


--- PyGate Linux v1.5.1
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

James Harris <james.harris.1@gmail.com> writes:

IME asking Grok to provide links is an exercise in futility. The URLs
look ideal - they have all the components in them that you would think
would answer your query - but when you click on them you get a 404 or
the wrong article.

So maybe you should reconsider your choice then? I've never had a
hallucinated link from chatgpt but I'm not a big user of AI.

--- PyGate Linux v1.5.1
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

On 03/12/2025 09:48, Anssi Saari wrote:

James Harris <james.harris.1@gmail.com> writes:

As you can see, it does say clearly that Linux can handle ECC reports
on the CM5 but thus far I've not found any sources to back it up.

Why not ask your chosen AI tool to simply provide links to sources and
verify yourself? With a quick search I didn't find support for any of
this ECC support stuff either, except for the standard on-die ECC.

IME asking Grok to provide links is an exercise in futility. The URLs
look ideal - they have all the components in them that you would think
would answer your query - but when you click on them you get a 404 or
the wrong article.

Grok says it's called AI hallucination and is common to other AI
engines, too, and that its makers are working on fixing it in future.
For now, however, its links are rubbish.

So are some of its quotations.

Both problems seem to be something to do with recognising patterns and generating corresponding patterns


--
James Harris



--- PyGate Linux v1.5.1
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

On 03/12/2025 14:37, The Natural Philosopher wrote:

On 03/12/2025 14:18, druck wrote:

On 01/12/2025 21:16, James Harris wrote:

To make a NAS / server I am looking for a Raspberry Pi with ECC support. >>>
I gather that the Pi 5 and the CM 5 use RAM which has internal ECC.
That's good. But do they inform the OS of when there are corrections
and uncorrectable errors?

Have you guys any idea as to whether the OS running on the Pi5 or CM5
will be able to tell whether RAM has had errors or not?

As far as I know it's only for error correction within the RAM
controller, and not reported to the OS. There is a possibility the SOC
might know about it via some proprietary Broadcom mechanism.

That is my impression too, based on admittedly sketchy data.
Lets face it at that level, ecc will simply make slightly dodgy ram rock solid and merely postpone the day when there is a kernel panic due to corrupted RAM.

I cant see than knowing about it helps much.

If it tells you there is a problem before it becomes too severe to
correct, then it may be useful so that you can replace the system.

I have found a webpage where some idiots attempting to fry a Pi 5 by
running AI on it without a cooler got a fatal ECC error, but my money
is on they just hallucinated the entire thing.

Well indeed.

The internet may be made of cats, but these days its full of
(commercially sponsored and politically motivated) lies.

--- PyGate Linux v1.5.1
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)