I also think having it is lots better than not having it, but especially in the case of permanent errors somewhere on the die the correction just manages to make a defective die usable. It can not protect against transitional faults anymore in that region.
On a fully working die it should protect against transient faults (those will happen more often on DDR5) but there is no way to know if it is correcting something at any point in time or not. Meaning there could be chips so defective that only the on-die-ECC makes them viable to use. Those would have to be identified on errors occuring during runtime or with memory stress testing, just as normal RAM would have to be. I have not seen error rates for DDR5 yet, but I would suspect those are not massively better than non-ECC DDR4 RAM. With DDR5 we see smaller process nodes and higher clocks being used than for DDR4. There is actually even DDR4 RAM using on-die-ECC to increase yields and get more usable chips out of a production process (https://microarch.org/micro53/papers/738300a282.pdf)
In the meantime, until ECC “yes or no” is confirmed officially by Framework, I asked support to remove the RAM from my current batch 1 order. Once I know what is used, I’ll change the order again or get my Ram somewhere else.
Does anybody here know, if ECC functionality needs additional traces on the motherboard between CPU und FP7r2 socket in order to function? If that is the case and framework did not plan these on their PCB design, we won’t even need to be concerned with digging into BIOS or microcontroller issues that may prevent ECC functionality.
But if ECC is working on Framework 16, I would also very much love to see it in the Framework 13. AMD Ryzen 7480U also does support ECC according to AMD specs.
I have been contacting some other manufacturers using 7840HS or 7840U in their Notebooks about ECC support. I will let you know, if anyone of them confirms support.
Edit:
As soon as Framework 13 and 16 with AMD are released, maybe someone could check the traces on the motherboard.
Update 1:
One of the manufacturers I contacted is SolidRun. They sell industrail PCs and their newest model the Bedrock R7000 uses a 7840HS and lists ECC RAM as compatible in the specs. I contacted them and they confirmed, that the PC can run unbuffered DDR5 ECC RAM. Interestingly, the PC does come with the regular, non-pro Version of the CPU. So ECC RAM seems not to be a Pro-Version feature.
If have also contacted AMD and some other manufacturers. Apart from SolidRun only XMG did answer as of now. Their laptop models do not support the ECC Ram functionality, but they will discuss it for future models with AMD processors.
Traces are absolutely needed between FP7r2 and the SO-DIMM slots. The parity data has to be transmitted on those traces to the memory controller in the CPU and is checked there. So no traces means 100% no ECC support as you say. Traces connected could still mean anything from fully working to not working at all.
Not: we’re working with AMD to improve ECC currently not working.
Just flat out: it’s 100% not gonna be available. link
Yes, it could be a situation like with the AM5 platform where broken ECC gets fixed over time. But I don’t see any effort from Framework to add ECC support. And I am not gonna buy a Framework laptop just to test. I was wise to not preorder.
All four of these processors have exactly the same note on ECC RAM:
Yes (FP7r2 only; Requires platform support)
where FP7r2 is socketed RAM, vs LPDDR5 RAM which is soldered.
That covers the low end of AMD processors available on Framework 13 and 16 laptops.
Just for clarity, even though that’s the official documentation, don’t take that as gospel. In this very thread nrp mentions that Framework has seen those webpages, but haven’t gotten confirmation from AMD and haven’t seen ECC work in testing. It sounds like the product pages are either wrong, or somehow irrelevant. Maybe the one AMD person who knows how to make it work decided to leave engineering and be a construction worker instead.
Update 2:
From the companies I asked about ECC Support for their AMD Ryzen 7000 products I got another answer. I asked ASRock about the 4X4 BOX-7840U which is a NUC-like Mini-PC. This is their answer:
" … Regarding your question, 4X4-7840U could support non-ECC up to DDR5 5600MHz.
We also had checked with AMD, they claimed that ECC is only supported by PRO CPU skus officially, and the website information will be revised later. … "
Don’t know what is going on at AMD, but it seems like the right hand is not knowing what the left is doing.
Summary till now:
7840HS: Does support ECC Ram (confirmed by Solidrun)
7840U: according to manucfacturers does not officially support ECC, according to AMD does support ECC - so propably unofficial support
7840U Pro and 7840HS Pro: propably does support ECC Ram
I will keep you updated about further answers from companies.
I think another summary could be:
It’s a mess!
And as with any mess, there is little guarantee of anything.
My guess is that AMD attempted to bring ECC memory to the whole line, hence their website information claiming that it’s supported, and AMD reps repeating that. But AMD ran into issues in getting it to work consistently, they thought they could iron out the issues, but it turned out harder than they hoped. ECC can work, but when issues pop up it may take significant effort with AMD working with the board manufacturer, so they are starting to tell manufacturers that it’s “actually” only supported on higher priced Pro CPUs. Again, just a guess. Looking for a reason for the conflicting information, including from AMD themselves.
Just wanted to note, it is clarified in the sentence right after the line highlighted here that this is AMD’s position, not Framework’s. Framework tried asking the AMD engineers they worked with to create this board, and the engineers told them that “because it’s not a ‘Pro’ processor that it won’t have ECC support”.
This does not mean that Framework was told that they could have made it work and decided it wasn’t worth the trouble. They were told by AMD engineers that they could not enable this support.
Until full ECC support is available, I, for one, will hold off buying this laptop and keep using older Xeon models from other integrators that offer it.
Is the Pro series not available for the time being or will it not be available in the future?
I don’t really understand the needs of consumers in the computer market, but I think there are advantages to laptops that support ECC memory, and almost all laptops equipped with Ryen Pro APUs on the market use soldered RAM (I’m talking about Lenovo and HP).
So from my perspective, I would recommend that the next generation of products consider ECC support when developing.