EDAC MC0: Giving out device to module amd64_edac

See my blog post about it (it was top of HN): https://sunshowers.io/posts/am5-ryzen-7000-ecc-ram/

I would assume your particular motherboard to operate with proper SECDED+-level ECC if you have capable, compatible DIMM, enable ECC mode in the firmware, and boot an OS kernel that can make sense of it all.

I am writing this message on such an ASUS MB with a Ryzen CPU and working ECC memory. You must check that you actually have a recent enough OS to know your Threadripper CPU and that you have installed any software package required for this (e.g. on Linux "edac-utils" or something with a similar name).

Some businesses (and governments) try and unify their purchasing, but this seems to make things worse, with the purchasing department both not understanding technology and being outwitted by vendors.

I've been building my own gaming and productivity rigs for 20 years and I don't think memory has ever been a problem. Maybe survivorship bias, but surely even budget parts aren't THIS bad.

A computer with 64 GB of memory is 4 times more likely to encounter memory errors than one with 16 GB of memory.

When DIMMs are new, at the usual amounts of memory for desktops, you will see at most a few errors per year, sometimes only an error after a few years. With old DIMMs, some of them will start to have frequent errors (such modules presumably had a borderline bad fabrication quality and now have become worn out, e.g. due to increased leakage leading to storing a lower amount of charge on the memory cell capacitors).

For such bad DIMMs, the frequency of errors will increase, and it may become of several errors per day, or even per hour.

For me, a very important advantage of ECC has been the ability to detect such bad memory modules (in computers that have been used for 5 years or more) and replace them before corrupting any precious data.

I also had a case with a HP laptop with ECC, where memory errors had become frequent after being stored for a long time (more than a year) in a rather humid place, which might have caused some oxidation of the SODIMM socket contacts, because removing the SODIMMs, scrubbing the sockets and reinserting the SODIMMs made disappear the errors.

    94 2025-08-26 01:49:40 +0200 error: Corrected error, no action required., CPU 2, bank Unified Memory Controller (bank=18), mcg mcgstatus=0, mci CECC, memory_channel=1,csrow=0, mcgcap=0x0000011c, status=0x9c2040000000011b, addr=0x36e701dc0, misc=0xd01a000101000000, walltime=0x68aea758, cpuid=0x00a50f00, bank=0x00000012
    95 2025-09-01 09:41:50 +0200 error: Corrected error, no action required., CPU 2, bank Unified Memory Controller (bank=18), mcg mcgstatus=0, mci CECC, memory_channel=1,csrow=0, mcgcap=0x0000011c, status=0x9c2040000000011b, addr=0x36e701dc0, misc=0xd01a000101000000, walltime=0x68b80667, cpuid=0x00a50f00, bank=0x00000012

It's always the same address, and always a Corrected Error (obviously, otherwise my kernel would panic). However, operating my system's memory at this clock and latency boosts x265 encoding performance (just one of the benchmarks I picked when trying to figure out how to handle this particular tradeoff) by about 12%. That is an improvement I am willing to stomach the extra risk of effectively overclocking the memory module beyond its comformt zone for, given that I can fully mitigate it by virtue of properly working ECC.

This was running at like, 1866 or something. It's a pretty barebones 8th gen i3 with a beefier chipset, but ECC still came in clutch. I won't buy hardware for server purposes without it.

Edit: it's probably because I switched it to "energy efficiency mode" instead of "performance mode" because it would occasionally lock up in performance mode. Presumably with the same root cause.

Last winter I was helping someone put together a new gaming machine... it was so frustrating running into the fake ecc marketing for DDR5 that you mention. The motherboard situation for whether they support it or not, or whether a bios update added support or then removed it or added it back or not, was also really sad. And even worse IMO is that you can't actually max out 4 slots on the top tier mobos unless you're willing to accept a huge drop in RAM speed. Leads to ugly 48 GB sized sticks and limiting to two of them... In the end we didn't go with ECC for that someone, but I was pretty disappointed about it. I'm hoping the next gen will be better, for my own setup running ZFS and such I'm not going to give up ECC.

Some vendors use hamming codes with “holes” in them, and you need the CPU to also run ECC (or at least error detection) between ram and the cache hierarchy.

Those things are optional in the spec, because we can’t have nice things.

That is mostly to assist manufacturers in selling marginal chips with a few bad bits scattered around. It's really a step backwards in reliability.

However in the past there have existed very few CPU models and MBs that supported either kind of DIMMs, while today this has become completely impossible, as the mechanical and electrical differences between them have increased.

In any case, today, like also 20 years ago, when searching for ECC DIMMs you must always search only the correct type, e.g. unbuffered ECC DIMMs for desktop CPUs.

In general, registered ECC DIMMs are easier to find, because wherever "server memory" is advertised, that is what is meant. For desktop ECC memory, you must be careful to see both "ECC" and "unbuffered" mentioned in the module description.

In my experience, it's generally unwise to push the platform you're on to the outermost of its spec'd limits. At work, we bought several 5950X-based Zen3 workstations with 128GB of 3200MT/s ECC UDIMM, and two of these boxes will only ever POST when you manually downclock memory to 3000MT/s. Past a certain point, it's silicon lottery deciding if you can make reality live up to the datasheets' promises.

[0] https://en.m.wikipedia.org/wiki/Row_hammer

I also have a fractal define case with anti noise padding material and dust filters, but my temperatures are great and the computer is almost inaudible even when compiling code for hours with -j $(nproc). And my fans and cooler are much cheaper than his.

That should of course be sound padding...

Because CPUs can get much hotter in specific spots at specific pins no? Just because you're reading 100, doesn't mean there aren't spots that are way hotter.

My understanding is that modern Intel CPUs have a temp sensor per core + one at package level, but which one is being reported?

Anyway, OP's cooler should be able to cool down 250W CPUs below 100C. He must have done something wrong for this to not happen. That's my point -- the motherboard likely overclocked the CPU and he failed to properly cool it down or set a power limit (PL1/PL2). He could have easily avoided all this trouble.

https://www.techpowerup.com/review/intel-core-ultra-9-285k/2... lists maximum temperature as 88.2C with the previous gen NH-D15 cooler.

For a long time, my Achille's heel was my Bride's vacuum. Her Dyson pulled enough amps that the UPS would start singing and trigger the auto shutdown sequence for the half rack. Took way too long to figure out as I was usually not around when she did it.

You said the right words but with the wrong meaning! On Gigabyte mobo you want to increase the "CPU Vcore Loadline Calibration" and the "PWM Phase Control" settings, [see screenshot here](https://forum.level1techs.com/t/ddr4-ram-load-line-calibrati...).

When I first got my Ryzen 3900X cpu and X570 mobo in 2019, I had many issues for a long time (freezes at idle, not waking from sleep, bios loops, etc). Eventually I found that bumping up those settings to ~High (maybe even Extreme) was what was required, and things worked for 2 years or so until I got a 5950X on clearance last year.

I slotted that in to the same mobo and it worked fine, but when I was looking at HWMon etc, I noticed some strange things with the power/voltage. After some mucking about and theorising with ChatGPT (it's way quicker than googling for uncommon problems), it became apparent that the ~High LLC/power settings I was still using were no good. ChatGPT explained that my 3900X was probably a bit "crude" in relative quality, and so it needed the "stronger" power settings to keep itself in order. Then when I've swapped to 5950X, it happens to be more "refined" and thus doesn't need to be "manhandled" — and in fact, didn't like being manhandled at all!

But if your UPS (or just the electrical outlet you're plugged into) can't cope - dunno if I'd describe that as cratering your CPU.

It looks completely bonkers to me. I overclocked my system to ~95% of what it is able to do with almost default voltages, using bumps of 1-3% over stock, which (AFAIK) is within acceptable tolerances, but it requires hours and hours of tinkering and stability testing.

Most users just set automatic overclocking, have their motherboards push voltages to insane levels, and then act surprised when their CPUs start bugging out within a couple of years.

Shocking!

Yeah. If Asahi worked on newer Macs and Apple Silicon Macs supported eGPU (yes I know, big ifs), the choice would be simple. I had NixOS on my Mac Studio M1 Ultra for a while and it was pretty glorious.

I had the same issue with my MSI board, next one won't be a MSI.

My modern CPU problems are DDR5 and the pre-boot timing thing never completing. So a build of a 9700x that I did that WAS supposed to be located remotely from me has to sit in my office and have its hand held thru every reboot cuz you never know quite know when its doing to decide it needs to retime and randomly never come back. Requires pulling the plug from the back and waiting a few minutes then powering back, then waiting 30 minutes for 64gb of ddr5 to do its timing thing.

This isn't really true. On the same process node the difference is negligible. It's just that Intel's process in particular has efficiency problems and Apple buys out the early capacity for TSMC's new process nodes. Then when you compare e.g. the first chips to use 3nm to existing chips which are still using 4 or 5nm, the newer process has somewhat better efficiency. But even then the difference isn't very large.

And the processors made on the same node often make for inconvenient comparisons, e.g. the M4 uses TSMC N3E but the only x86 processor currently using that is Epyc. And then you're obviously not comparing like with like, but as a ballpark estimate, the M4 Pro has a TDP of ~3.2W/core whereas Epyc 9845 is ~2.4W/core. The M4 can mitigate this by having somewhat better performance per core but this is nothing like an unambiguous victory for Apple; it's basically a tie.

> I have an M1 Max, a few revisions old, and the only thing I can do to spin up the fans is run local LLMs or play Minecraft with the kids on a giant ultra wide monitor at full frame rate. Giant Rust builds and similar will barely turn on the fan. Normal stuff like browsing and using apps doesn’t even get it warm.

One of the reasons for this is that Apple has always been willing to run components right up to their temperature spec before turning on the fan. And then even though that's technically in spec, it's right on the line, which is bad for longevity.

In consumer devices it usually doesn't matter because most people rarely put any real load on their machines anyway, but it's something to be aware of if you actually intend to, e.g. there used to be a Mac Mini Server product and then people would put significant load on them and then they would eat the internal hard drives because the fan controller was tuned for acoustics over operating temperature.

This anecdote perfectly describes my few generation old Intel laptop too. The fans turn on maybe once a month. I dont think its as power efficient as an M-series Apple CPU, but total system power is definitely under 10W during normal usage (including screen, wifi, etc).

One of the many reasons why snapdragon windows laptops failed was both amd and Intel (lunar lake) was able to reach the claimed efficiency of those chips. I still think modern x86 can match arm ones in efficiency if someone bothered to tune the os and scheduler for most common activities. M series was based on their phone chips which were designed from the ground up to run on a battery all these years. AMD/Intel just don't see an incentive to do that nor do Microsoft.

Intel is just bad at the moment and not even worth touching.

Definetly not that one if you plan to pair with a dedicated GPU! The 5700X has twice the L3 cache. All Ryzen 5000 with a GPU have only 16MB, 5700 has the GPU deactivated.

A common approach is to go into the BIOS/UEFI settings and check that c6 is disabled. To verify and/or temporarily turn c6 off, see https://github.com/r4m0n/ZenStates-Linux

I have always run B series because I've never needed the overclocking or additional peripherals. In my server builds I usually disable peripherals in the UEFI like Bluetooth and audio as well.

On desktop PCs, thermal throttling is often set up as "just a safety feature" to this very day. Which means: the system does NOT expect to stay at the edge of its thermal limit. I would not trust thermal throttling with keeping a system running safely at a continuous 100C on die.

100C is already a "danger zone", with elevated error rates and faster circuit degradation - and there are only this many thermal sensors a die has. Some under-sensored hotspots may be running a few degrees higher than that. Which may not be enough to kill the die outright - but more than enough to put those hotspots into a "fuck around" zone of increased instability and massively accelerated degradation.

If you're relying on thermal throttling to balance your system's performance, as laptops and smartphones often do, then you seriously need to dial in better temperature thresholds. 100C is way too spicy.

If nothing else, it very clearly indicates that you can boost your performance significantly by sorting out your cooling because your cpu will be stuck permanently emergency throttling.

Smartphones have no active cooling and are fully dependent on thermal throttling for survival, but they can start throttling at as low as 50C easily. Laptops with underspecced cooling systems generally try their best to avoid crossing into triple digits - a lot of them max out at 85C to 95C, even under extreme loads.

Intel's basic 285K spec's - https://www.intel.com/content/www/us/en/products/sku/241060/... - say "Max Operating Temperature 105 °C".

So, yes - running the CPU that close to its maximum is really not asking for stability, nor longevity.

No reason to doubt your assertion about gaming laptops - but chip binning is a thing, and the manufacturers of those laptops have every reason to pay Intel a premium for CPU's which test to better values of X, Y, and Z.

Competitors to NVidia really need to figure things out, even for gaming with AI being used more I think a high end APU would be compelling with fast shared memory.

I gather it's very difficult and expensive to make a board that supports more channels of RAM, so that seems worth targeting at the platform level. Eight channel RAM using common RAM DIMMs would transform PCs for many tasks, however for now gamers are a main force and they don't really care about memory speed.

I've never overclocked anything and I've never felt I've missed out in any way. I really can't imagine spending even one minute trying to squeeze 5% or whatnot tweaking voltages and dealing with plumbing and roaring fans. I want to use the machine, not hotrod it.

I would rather Intel et al. leave a few percent "on the table" and sell things that work, for years on end without failure and without a lot of care and feeding. Lately it looks like a crapshoot trying to identify components that don't kill themselves.

So if your CPU is drawing "more than 200W" you're pretty much at the limits of your cooler.

[1] https://www.bequiet.com/en/cpucooler/5110

But I agree this should not be a problem in the first place.