Searching for a new motherboard

FWIW, IIRC, DDR5 has ECC built-in by design... so any DDR5 sticks OP gets will have ECC built in. Trouble is, DDR5 is expensive af. $100 USD will only net you an 8 GB stick of DDR5. Blame the AI-driven rammageddon.
😩
 
FWIW, IIRC, DDR5 has ECC built-in by design... so any DDR5 sticks OP gets will have ECC built in. Trouble is, DDR5 is expensive af. $100 USD will only net you an 8 GB stick of DDR5. Blame the AI-driven rammageddon.
😩
That built-in ECC is worthless. It just hides errors.

There are real ECC DDR5 modules. Both registered and unbuffered.
 
You used to be able to tell what you had by counting the number of chips on the dimm module, but it's got a bit more complicated recently. There are now (in ECC DDR5):- ECC UDIMMs, EC4/x72 RDIMMS and EC8/x80 RDIMMS ('U' for unregistered/unbuffered and 'R' for registered), where the highest server grade is EC8/x80 RDIMMs. The EC4/x72 is a lower-cost version of EC8/x80 (72 bits versus 80 bits), but having lower fault-tolerance.
Eg. see pages 4-5 of https://www.scribd.com/document/807434495/RAM and here https://lenovopress.lenovo.com/lp1618-introduction-to-ddr5-memory
 
I don't recommend the Gigabyte Aorus Master X670E. The board has issues with the igc Ethernet driver, which can stop working once your USB keyboard or mouse begins rapidly connecting and disconnecting (that issues is caused by the usb extension port cable, mobo has some issues to handle that, tried many vendors). When that happens, you can lose not only Ethernet but also your input devices.USB stability on this board is a pain, and these problems are tied to the onboard Intel iwl Wi‑Fi/Bluetooth hardware. Disabling Wi‑Fi and Bluetooth is basically mandatory if you want any stability, to avoid unexpected hw failures.

There are also issues with the M.2 PCIe slots. Installing a KC3000 (same wirh WD Black) often requires 2–3 reboots after a cold start to avoid controller hangs. Additional sysctl tuning or bandwidth limiting helps reduce burst R/W behavior, but once the NVMe hangs, a hard reset is your only friend — just like with all the other issues mentioned above.This is not limited to FreeBSD; I’ve seen the same behavior on Linux and Windows.

Intel confirmed that this motherboard has several components that were improperly mounted.

My longest uptime was 3 months and it hangs... I couldn't get any dump data back for debugging.

My old Asus Sabertooth Z77 is a trillion times more stable than this board -> 7y uptime. UPS failed and had to replace it.
 
USB stability on this board is a pain
I learned that hard on X470 with Rift CV1 PCVR and 4 sensors (Ryzen CPU-side USB went unstable with HMD and 1 sensor; my board luckily had separate ASMedia ports to offload around), and hearing others report issues on 500-series (and now a 600) has me skeptical with USB on Ryzen.

I had an ASUS Prime X470-PRO board and disabling IOMMU was a fix for random PCI errors on Linux.

I use the Asus Prime branded AM4 board
Iirc with that X470-PRO board it had memory daisy-chained in an interesting way (apparently it's easier to reach higher speeds/overclock if they're wired one way; mine was the other way :p but I couldn't run 4x 3666Hz rated sticks because of that and/or 2700X/Zen2)
 
I use the Asus Prime branded AM4 board. Works fine, does ECC. I think the Prime AM5 boards are equally good, but of course the memory costs much more.

thanks for the pointer, I looked up this MB and the datasheet is impressive. it even has an internal serial connector for kernel debug :)

but can you tell us the exact ECC memory you paired it with?
I looked at a lot of refurbished DDR4 ECC server dimms (both registered ECC and unbuffered ECC kind) and none appear on the motherboard's compatibility list, so I am unsure if they will work at all.
 
thanks for the pointer, I looked up this MB and the datasheet is impressive. it even has an internal serial connector for kernel debug :)

but can you tell us the exact ECC memory you paired it with?
I looked at a lot of refurbished DDR4 ECC server dimms (both registered ECC and unbuffered ECC kind) and none appear on the motherboard's compatibility list, so I am unsure if they will work at all.

I bought 4x 32 GB DDR4 unreg ECC on the Crucial store.

Generally I have few problems with RAM except Nemix (Newegg house brand I think?), which is junk.
 
On-chip ECC can correct internal errors, theoretically, but cannot detect and correct errors between DIMM and CPU (connectors, PCB patterns, noises from outside, ...). So it alone is different from ECC DIMMs.
 
DIMMs that are labelled Samsung, SK-Hynix and Micron are the three I usually go for, or the major system manufacturers like Dell, Lenovo, HP that re-badge those three makes as their own, and Kingston. I expect Crucial are fine but I haven't used them. I had some Elpida (japanese) ram years ago that was fine too, but I think Elpida went out of business. I wouldn't touch anything else. If the motherboard manual says it will work with PC4-1700P-R or whatever spec they state, it should work fine with dimm's of that spec from any of those three main makes. The 'approved' dimms they list for the motherboard, are the ones they have actually qualified, ie the ones they have tested it with, but you should have no problem with any of the big 3 (samsung, hynix and micron) provided they are of the same spec.

For example, below is a classic samsung DDR4 32GB ECC RDIMM. The main spec you can read on the label is PC4-2400T-R, meaning DDR4, the bus speed, and registered (homework: look up the difference between '2400' and '2400T'). You can tell it's ECC because there are 9 sets of chips instead of 8 that you would find on a non-ECC (ie, desktop) DIMM (at least, that rule of thumb works for DDR4. DDR5 is a bit more complicated), and you know it's registered because of the extra chip in the centre. Note that in this case we have a Samsung DIMM and the DRAM chips you can see are also from Samsung, labelled 'SEC' which is Samsung Electronics Corp'. There is a useful guide here https://rampricesusa.com/guides/server-ram-guide . If you're lucky and it has a FRU number on the label you can search for the FRU number to check the exact specification. The wikipedia page for DDR4 https://en.wikipedia.org/wiki/DDR4_SDRAM describes the meanings of the JEDEC standard terminology (see the table under 'JEDEC standard DDR4 module'). Just make sure you get the same JEDEC spec as the motherboard web page says it supports and you should be fine. Also check if you need any special type like low-profile dimms to fit the case. The other rule is generally to put all the same type on a given motherboard, so if there are 8 slots then fill them all with the same type dimm, don't mix different types. So if you have some slots already installed, just buy some more of the same spec. that you already have. If the mobo data says it supports 3 different specs, for example, then choose one spec and populate it with dimms all of the same spec. That's what I do, anyway :-)

DIMM.jpg

(Answer to homework: https://superuser.com/questions/1291897/whats-the-difference-between-pc2400t-and-pc2400-memory . It's just a more precise timing spec.)
 
I place G.Skill memory in all my client machines for decades.

Not one single failure.
Corsair was endless failures.
Micron regular and ECC has also been flawless.
The little bit of Samsung and Kingston I see have not failed.

Over many years in my service business, RAM is not a significant failing point.

Cheap PSU is the root cause of non-disk failures I’ve seen
 
I place G.Skill memory in all my client machines for decades.

Not one single failure.
Corsair was endless failures.
Micron regular and ECC has also been flawless.
The little bit of Samsung and Kingston I see have not failed.

Over many years in my service business, RAM is not a significant failing point.

Cheap PSU is the root cause of non-disk failures I’ve seen
Yay — that’s one of those background, “forbidden” units that taught me a hard lesson as well.
 
I place G.Skill memory in all my client machines for decades.

Not one single failure.
Corsair was endless failures.
Micron regular and ECC has also been flawless.
The little bit of Samsung and Kingston I see have not failed.

Over many years in my service business, RAM is not a significant failing point.

Cheap PSU is the root cause of non-disk failures I’ve seen
Interesting, I never used G.Skill. I think its been extremely rare to have a dimm fail, I can't remember the last time it happened here. DRAM just goes on forever. Occasionally if I had a board that was being unstable, reseating the dimms and/or wiping the gold contacts with some IPA on a cotton bud to de-grease them usually does the trick; actual failure of memory dimms themselves seem to be very rare. It's pretty incredible really.
 
Back
Top