My opinion is this, logically thinking about it:
If you need ECC…then isn’t it very likely that you don’t want to bank it on 1st revision products? (Be it Apple, HP, Lenovo…etc)
OFF Topic: Phoronix has made some benchmarks on ECC vs. non ECC Ram for those who are interested.
[AMD Ryzen 9 7900X Performance With ECC DDR5 Memory - Phoronix](https://AMD Ryzen 9 7900X Performance With ECC DDR5 Memory)
Thought, as this was a thread about ECC it would be interesting for those thinking about it.
And again, Framework released a list of compatible RAM for the 13" Laptop AMD boards:
I suppose it will apply to the 16" model.
It is stated that: DDR5 modules have on-die ECC support and ECC modules will allow the system to function, however, ECC parity functionality is not supported by the non-Pro series APUs that Framework offers.
I blame Intel for starting the trend of using ECC as a enterprise feature to milk more money. Most cpu dies from Intel and amd have the hardware to fully support ECC, they just have it disabled.
It is absolutely disgraceful that we are still using non ECC RAM. That should be a thing of the past.
AMD updated their spec sheets on the 7040 non-PRO U processors and marked ECC as “No”. SolidRun still advertises ECC for U processors.
The page for the 6800U for example did not (yet) change but maybe no one pestered AMD with this and I do not know of actual working ECC with that processor.
I have good hopes for my Kingston 5600MT ECC at least working fine without parity since the 4800MT version seems to be. So here is to having ECC support sometime down the line (be it in a FW mainboard or whatever DDR5 SO-DIMM hardware):
Wow they actually did! Presumably AMD decided that’s gonna be a nice artificial market segmentation for their Pro CPU line. Given AMD’s track record on the desktop I really thought they would not do the same shenanigans als Intel.
Imagine if you relied on that spec sheet that’s been online for months. Who will still trust AMD’s product pages?
To quote Louis Rossmann:
Informative and unfortunate.
PS: My thanks goes to the Framework team. All things considered you were really transparent about current state of ECC support in your products as more details became available.
You will also find failures on other websites.
I’ve seen mainboards with misnamed connectors, Intel CPUs with wrong features (even “basic stuff” like HyperThreading yes/no), misleading numbering of hard drive slots in desktop cases, accessories which are not delivered with every unit, concerts beginning at the wrong time,…
In big companies not every required information finds it ways to every department. Even more if the final product is still in progress.
To be fair, all non-PRO desktop-APUs (at least) since Zen did not support ECC. The PRO SKUs did, but were essentially unobtanium since they were only sold to OEMs. So if the mobile non-PRO APUs do not support ECC that is at least consistent. Even so it is a bit disappointing since the hardware does not seem to differ on the PRO/non-PRO APUs.
I wonder how big the cost difference is between those SKUs. I would have paid extra for ECC support, especially when it is official.
I do however know that for the vast majority ECC isn’t even a consideration so having either everybody paying more or having to maintain a separate SKU is probably not sustainable.
If there are several hundred $ between the framework SKUs maybe having the high end one using a PRO CPU would be an alternative. That way there would be the same number of SKUs to maintain, higher price for PRO only hits the enthusiasts shelling out for the high end anyway and there are some other benefits like RAM encryption (SME) as well.
At least some of the non-PRO Zen SKUs supported ECC. My non-PRO Threadripper 2950X (a Zen 1+ CPU) and its motherboard (a MSI consumer board) both support ECC. On/off switch in the BIOS IIRC. I have 4x32GB of ECC in it right now for working with large databases, VMs, and containers.
It really is frustrating that we are not all using ECC by now. Absolutely no reason for it beyond gatekeeping enterprise hardware features.
What about cost? I think the average user doing gaming, office work, or creative work doesn’t really care about ECC or benefit from it. Having more segmentation allows people to choose between features and price. From the perspective of AMD or Intel, if you care about ECC, you’re most likely a professional that can afford to fork out a lot of money for it.
For a lot of consumer chips, cost versus performance is a big factor in purchasing decisions. Additional features that add cost but are unutilized can really skew that equation.
It would cost them nothing to leave support enabled in their CPUs. Having ECC support doesn’t necessarily force you to buy ECC ram that may be more expensive.
1 word: money
If you got enough money, AMD and Intel will be happy to leave that toggle on.
It’s no different with Nvidia with their Quadro line in the past.
That’s right back to my point of gatekeeping (expensive) enterprise hardware. Glad we can agree. But it’s a terrible reason for everyone but Intel.
This post also explains a lot of why ECC is important:
https://www.realworldtech.com/forum/?threadid=198497&curpostid=198647
Yes, I use it too on my 3700X and set it up working on Threadripper 1920X and Ryzen 5800X. But you need to differentiate between APU (monolithic with iGPU) and CPU (chiplets and until Ryzen 7000 no GPU). APUs generally didn’t support ECC outside of PRO-SKUs.
Didn’t think about Threadripper having PRO-SKUs while there the differentiation is not ECC but other features.
They may not realize they benefit from it or the benefit might be marginal but more stability or data integrity is always a good thing. They may not care enough to pay more for it but they would benefit, however slight or marginal.
I’ll agree with you on that but it is as you mentioned, it maybe be slight or marginal. For casual users, I think data security on SSDs are more of a concern or at the forefront.
And how can you ensure that, if every single bit going to the SSD must go through the maybe failure inducing non-ECC RAM? How to even detect such errors?
I have collected lots of data over the years and there are at least dozens of files I have seen with some kind of corruption (most easily found in compressed video/images). I have not found such a file from the timespan where my NAS has used ECC yet.
That might have been software bugs, disk errors, file system shenanigans or any number of causes but I take the added safety of ECC for the 10% to 20% additional cost on RAM any day over not even being able to discover errors at all.
My NAS is currently a cheap Intel Celeron on a workstation chipset mATX board, that whole system without disks cost me roundabout 250€ years ago. There is no way I’m going back to non-ECC on anything that runs longer than a few hours at a time.
I guess there is some (a lot?) more cost in testing the whole platform (all possible combinations CPU, MB, RAM) to guarantee correct functioning of it. Plus added support costs due to more support requests of people not knowing how it works.
If you had read what I linked in my post, you would have seen that AMD likely didn’t do any of that with their Ryzen CPUs, and just left support there. Hence the “unofficially supported”.