Raven Ridge Desktop APUs Come Out Tomorrow, The Likely Linux Requirements

[starshipeleven]

What. The actual. Hell.
My AsRock mobo does indeed support ECC... I don't recall if it had support when it was launched, but still, it is surprising and nice to have a workstation feature like that on an inexpensive board like the AB350M Pro4.

Asrock (and Asrock Rack) are pretty unconventional. For example Asrock has announced a mATX board for Threadripper, which is probably going to be the only one around.

Other manufacturers might have some boards with ECC support and some others not. One I'm keeping tabs on is a Gigabyte mini-itx board with Displayport. I really hope the APUs support ECC so I can make my wet dream build.

https://www.gigabyte.com/Motherboard/GA-AB350N-Gaming-WIFI-rev-10

A few years down the line this might provide a cheap upgrade path for people that want a 64GB RAM machine for a hobby or work, seeing as ECC DIMMs usually sell for "pennies" on ebay.

Not the case, sadly. The ECC ram selling for cheap is the "registered" or "fully buffered" ECC ram. This type of ram only goes on mid and high-end servers, and that's why it is on sale for cheap (nobody buys it at all).

Ryzen, socket 155x Intels, and even Threadripper are using unbuffered ram (ECC or not), which is still expensive even if you buy used.

[Brophen]

Asrock (and Asrock Rack) are pretty unconventional. For example Asrock has announced a mATX board for Threadripper, which is probably going to be the only one around.

Love that they made that mATX board. If I had the money I would see if I couldn't do a dual GPU build with it - that would be fun

[DrYak]

I think AMD has enabled ECC support in their consumer SKUs prior to Ryzen as well, but I don't remember ECC having such a widespread adoption in consumer motherboards. Of course I might be mistaken, but I seem to remember that you had to buy select AM3+/FM2+ motherboards to have ECC support, and those were typically not positioned in the low-end or mid-range.

ECC has been always enabled in the AMD CPUs almost since forever (mostly because they were the first to move the memory controller on the CPU die, and thus it didn't make sens to create two separate variants between the server and consumer markets. Only the APUs which are never going to be seen in servers and workstations lack ECC (These weren't designed with ECC support to begin with).
The Zen-based APUs are new in that point of view, because they are the first APUs to also target workstations - hence availability in the AMx series of sockets and not only in laptop sockets).

Thought ECC wasn't widely advertised. Neither on the CPU side, nor on the motherboard side.

I've been using ECC in my home server for quite some time (AM2+ socket).
Even back then there was quite a choice of motherboard, including manufacturer who typically target consumers and including small cheap motherboard, not only huge multi-socket expensive monsters for servers. (mine is a mid-range ASUS mini-ATX).

This is completely different from the Intel world where ECC is considered a "premium" feature and marketed at a high price for high-range hardware (mostly targeting the rich "corporate" segment).

Most people don't realize this, but all computers produce multiple memory errors every single day. It might manifest itself a momentary glitch on screen, or a segmentation fault in an app, or even a corrupted file on storage. But whether you realize it or not it happens multiple times every single day....

Well, if you have *multiple memory errors every single day*, you should have done more memtest86+ back when you built it.
(Might be imperfect memory module, might be flaky power supply, might be insufficient cooling).

That, or don't build your data center in Pripyat :-D

Memory error can occur, but they shouldn't occur that often under normal circumstances.

[chithanh]

Only the APUs which are never going to be seen in servers and workstations lack ECC (These weren't designed with ECC support to begin with).
The Zen-based APUs are new in that point of view, because they are the first APUs to also target workstations - hence availability in the AMx series of sockets and not only in laptop sockets).

I think that is wrong. The only APUs to not support ECC were Zacate and Llano.
From Kabini/Trinity on, they had always ECC support, however the AM1/FM2/2+platform lacked necessary pins for ECC. But BGA based mobos would still offer ECC support.

Well, if you have *multiple memory errors every single day*, you should have done more memtest86+ back when you built it.
(Might be imperfect memory module, might be flaky power supply, might be insufficient cooling).

That, or don't build your data center in Pripyat :-D

Memory error can occur, but they shouldn't occur that often under normal circumstances.

From the often-quoted Google study about DRAM errors, they found error rates of 8% per DIMM per year.

ECC memory is also often the last line of defense against Rowhammer.

[duby229]

ECC has been always enabled in the AMD CPUs almost since forever (mostly because they were the first to move the memory controller on the CPU die, and thus it didn't make sens to create two separate variants between the server and consumer markets. Only the APUs which are never going to be seen in servers and workstations lack ECC (These weren't designed with ECC support to begin with).
The Zen-based APUs are new in that point of view, because they are the first APUs to also target workstations - hence availability in the AMx series of sockets and not only in laptop sockets).

Thought ECC wasn't widely advertised. Neither on the CPU side, nor on the motherboard side.

I've been using ECC in my home server for quite some time (AM2+ socket).
Even back then there was quite a choice of motherboard, including manufacturer who typically target consumers and including small cheap motherboard, not only huge multi-socket expensive monsters for servers. (mine is a mid-range ASUS mini-ATX).

This is completely different from the Intel world where ECC is considered a "premium" feature and marketed at a high price for high-range hardware (mostly targeting the rich "corporate" segment).



Well, if you have *multiple memory errors every single day*, you should have done more memtest86+ back when you built it.
(Might be imperfect memory module, might be flaky power supply, might be insufficient cooling).

That, or don't build your data center in Pripyat :-D

Memory error can occur, but they shouldn't occur that often under normal circumstances.

You can run memtest, it doesn't matter where you are, what equipment you use, where it bought from, or how you use it. It's still gonna happen. It happens on every computer every day.

[grok]

Same, I was also thinking a NAS, I think all of Ryzen is supposed to support ECC but it's left up to the motherboard maker whether to implement ECC or not

If it's for a NAS why not get an A6-9500E.

[grok]

I am still looking forward to some Kabini-successor. My Kabinis are still fine, and probably will be for the next years, but I guess AMD is busy delivering to all the many market segments (their semi custom people, dGPUs, RyZen/TR, server Epyc, mobile, embedded segment) so that the low margin little APUs will probably follow later. But those things make nice litte secondary/tertiary/q. computers, HTPCs, SOHO file servers (if given enough SATA connectors), kiosk stations, machine controllers (if given an RS-232 and the likes) and whatnot. Low power consumption, possibly passively cooled, niceprice, enough bang for most daily workloads and still not headless if you want a display. And also offering a moderate GPU power so you can have desktop effects plus a freedom driver.
Also for some mobile typewriters / text readers those might be interesting (combine with a good display and keyboard).

For some or most of those things, a one-watt CPU would do!
But otherwise the Kabini successor, Atom competitor is expected for 2019 as a 7nm Zen 2.

[DrYak]

You can run memtest, it doesn't matter where you are, what equipment you use, where it bought from, or how you use it. It's still gonna happen. It happens on every computer every day.

I'm not saying that error are impossible. They *DO* happen.

I'm just saying that the rates you're reporting are rather *extremely* high.

That's something to be expected maybe on a crappy cheap machine (the can you can find bargained in some supermarkets. Using the cheapest components possible, of mediocre quality. Stuffed in badly cooled case.)

That's not something to be expect in a correctly built machine.

It's not that error do not occur on high quality machines, it's just that they don't occur *multiple times per day* under normal circumstances.

If you're getting occasional errors and bit flips every once in a while : Yes, ECC is the answer.
If you're getting almost all your daily data completely fucked up by corruption : Then you're having a couple of more fundamental problems.

[grok]

That's something to be expected maybe on a crappy cheap machine (the can you can find bargained in some supermarkets. Using the cheapest components possible, of mediocre quality. Stuffed in badly cooled case.)

That's not something to be expect in a correctly built machine.

I disagree a bit. This stuff uses low power CPUs, sometimes 10 watts (less on laptop!). My favorite example is the skylake/kabylake Celeron as its TDP is just over 50 watts but it consumes about half!
Really low end stuff has an Atom or Jaguar with two cores disabled.
Sometimes low end stuff like motherboards are quite reliable because with high volumes, issues get fixed.
It's known that OEMs use lowest grade motherboards though and you might want to avoid their BIOS.

If anything the objectionable issue might be if the RAM is soldered lpddr3 (can't be replaced)
that's on Atom netbooks, some micro desktops, and the netbooks Apple calls macbook pro.