Hacker News new | ask | show | jobs
Asus Prime X370-PRO specs suggest AMD Ryzen processors may support ECC memory (webcache.googleusercontent.com)
128 points by en 3408 days ago
17 comments

It is good that AMD now look like they will finally be pushing Intel out of complacency. ECC should be the norm, not the exception and hopefully this will help drive the price of ECC memory down so that happens.
Finally ? AMD used to always support ECC, I believe their latest consumer chips didn't but that was mostly because it was never used because most motherboards makers don't want to add its support to their non-server variants.

Because why would they ? The very people for who that matters are those who have the needs and funds to buy the more expensive variant. Same reasoning as to why Intel isn't adding it to their desktop cpu.

Enterprises are also professional cost cutters, so if AMD can provide enterprise-class features for less, then they will find a market.
That's demonstrably not true, AMD has supported ECC more or less across the board since the Athlon64.

Enterprises may be professional cost cutters, but before that they're professional ass coverers, otherwise Oracle would have died 20 years ago.

No one got ever fired for buying Intel

Even if Ryzen will be twice as fast enterprise will still go with Intel. Unless some company with reasonable size (like DigitalOcean or Linode) will adopt Ryzen in their servers and brag about it.

We are absolutely looking at buying large quantities of (Ry)zen, if the numbers work out. We're most concerned about multi-core / NUMA penalty compared to Intel.

I assume we also get some sort of volume discount with Intel that may not be reflected in consumer pricing. But maybe we would get the same with AMD.

Also, CPUs are a tiny tiny portion of our costs of goods sold. So in the end, it doesn't matter too much how expensive Intel is relative to AMD. That said, we would prefer to see competition for the long run.

> No one got ever fired for buying Intel

Even after the whole kerfuffle about Intel paying off OEMs? [0] They got some pretty huge fines for that.

But in all seriousness, I think that if Ryzen can hit the perf/watt numbers of Intel, companies will seriously look at purchasing AMD systems again.

Back in the NetBurst days, lots of people were buying Opteron servers because they kicked the pants off Intel. If AMD can prove their competitive with Intel again in the data center market, people will buy their chips.

Competition is always good, even for large enterprise, and if the TCO of an AMD Ryzen system becomes on par with Intel, enterprise will happily pit them against each other in a price war. Enterprise will absolutely buy from both.

[0] http://money.cnn.com/2010/07/23/technology/dell_intel/index....

> No one got ever fired for buying Intel

Exactamently. Even if Ryzen has a way better performance/price ratio than Intel, it's a risk:

* risk of losing favour with Intel or vendors (aka getting "renegotiated contract")

* support risk

* less coherent hardware risk

* software compatibility risk

Ryzen is lot of risk, and the reward is not huge because at the end of the day CPU is a pretty minor investment. Unless Intel completely fucks up a generation as they did with P4 few company will be willing to take that risk. Hell even then most companies remained on the Intel ship.

I suspect that Intel would offer hardware nearly at cost to convince them to do otherwise.
Microsoft used (and still does) AMD CPUs for their first generation of Azure cloud instances. This did not do much in promoting AMD.
AMD does not support ECC on consumer platforms after AM3+. Socket FM1 through FM2+ are all incapable of supporting ECC.

For socket AM1 there are reports that it may be ECC capable, however no actual products exist that implement this support.

The socket can't support it, the processor most certainly can since the memory controller is built-in.
Hype headline, change it. AMD has almost always had ECC support in their desktop processors as long as I've used them. The motherboard makers are to blame for not implementing it.
Not much of a surprise coming from AMD, but it will most likely end up like their previous generations of CPU: most motherboards won't support ECC memory, forcing you to buy specialized hardware anyway. And if you're buying specialized hardware, you might as well go all the way.
>forcing you to buy specialized hardware anyway.

The link is literally a MBO claiming support for ECC and from what I can tell it will be ~170$ - which is not really unreasonable.

This smells:

> - Supports HDMI 1.4b

Erm, Bristol Ridge already has 2.0 http://www.anandtech.com/show/10705/amd-7th-gen-bristol-ridg...

> Supports DisplayPort with max. resolution 4096 x 2160 @ 60 Hz

That's DisplayPort 1.2. No way. It's going to support DP 1.4, Polaris already does.

Am I the only one finding "3-D printing friendly motherboard with patent-pending mounting design" ironic? (along with the classic "Proprietary" adjective marketing people like to put everywhere they can). This comment is for one of their boards on the original page.

Keep in mind these are motherboard specs. I would suggest not reading too much into them.

I am quite excited about Zen, but I hope they will make consumer motherboards that support more than 64GB RAM.

I disagree with the title, I don't think it says all Ryzen will support ECC.

>>4 x DIMM, Max. 64GB, DDR4 2666/2400/2133 MHz ECC and non-ECC, Non-ECC, Un-buffered Memory

That looks like a motherboard spec - because it is. And it supports Ryzen processors of various types. I read that as: One CPU supports "ECC and non-ECC", another CPU specifies "Non-ECC". At least that's how I read it. Otherwise there is no reason to put non-ECC in there twice.

The second one doesn't say Ryzen. I'm not up to date on AMD, but are you sure it's the same thing?
But it's a AM4 socket and is that only compatible with Ryzen atm?
AMD Ryzen™ Processors

4 x DIMM, Max. 64GB, DDR4 2666/2400/2133 MHz ECC and non-ECC, Non-ECC, Un-buffered Memory

AMD 7th Generation A-series/Athlon™ Processors

4 x DIMM, Max. 64GB, DDR4 2400/2133 MHz Non-ECC, Un-buffered Memory

That seems pretty clear to me that there's AM4 Ryzen and also there will be an AM4 A-series and an AM4 Athlon line. AMD has said repeatedly that AM4 is going to be the one socket for all their new chips from the server through to mobile. Look for the Opteron (or its replacement if the new server chips get a name change), the A series, the Athlon series, and Ryzen all to use the same socket but offer different memory and PCI configurations based on both the chipset and the CPU/APU.

That's awesome, I think this is the perfect CPU for my next ZFS NAS. If you try to find ECC support in Intel's lineup, it's hard to find much. Either the low end Pentium, or the high end Xeon ($$$).
Nonsense. Xeon e3 all have ECC AFAIK; the whole line. My old low end Xeon e3-1225 has it, and the latest e3-1225 v5 has it too. It's a $200 chip and is eminently suited for ZFS NAS (that's what mine is doing) - as well as desktop.
If all you want from it is a NAS, then said low-end Pentiums will do fine. The C2750, for instance, is plenty fast enough to drive a large number of HDDs.

Although right now there's a bricking bug to worry about, so maybe not that particular one right away.

And I believe they've removed it from the latest-gen Pentiums
This is not true. VT-d and ECC were thought to be removed, but these features were simply unlisted, not removed. Intel Ark now shows them as available, e.g:

http://ark.intel.com/products/97143/Intel-Pentium-Processor-...

Good to know, thanks for the info.
This maybe true for later models but not initial offers.

If you look a the Linux 4.10 change log [1] you will see that while zen supports ECC, it ignores the error correction message. So you can use ECC ram, but you get none of the benefit of ECC ram

[1] https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux....

That looks like it just adds a class of errors, "deferred", without explanation, though this commit [1] implies that these are for ECC errors whose handling was not urgent, and does not specify further how those would occur (unless the MC knows something about the page being used for cache/re-readable data or something, so it can just drop the page on the floor and reread it).

[1] - https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux....

Are warning messages like these common in the Linux kernel? Cursory overview makes it seem like the logic should never get there. Funny.

> WARN(1, "Something is rotten in the state of Denmark.\n");

> 4 x DIMM, Max. 64GB, DDR4 2666/2400/2133 MHz ECC and non-ECC, Non-ECC, Un-buffered Memory

To be fair, that "ECC and non-ECC, Non-ECC" reads like they were still editing the page and hadn't made up their minds yet if the motherboard supported ECC. In other words, they were deciding between "ECC and non-ECC" and "Non-EC".

Can someone tell me how many errors over the past 10 years or so would have been fixed by ECC memory. I've no idea at all how much better it would be and what the current error rate it prevents is? It sounds more nice to have than vital? Is there a performance cost to it at all?
Would prevent Rowhammer attacks.
DDR4 is innately much more resilient to rowhammer than previous standards.
More resilient maybe, but not Rowhammer-proof: https://arstechnica.com/security/2016/03/once-thought-safe-d...
Not completely, since those can produce multi-bit flips.
With SECDED ECC, in the event that two bits are flipped, the system can detect this and halt.
Regardless of this being something new one can hope that it marks a point in time where access to consumer-geared motherboards and CPUs that support ECC-memory becomes more available. Simply because people actually want it.
Ryzen also moved lots of the chipset logic to the chip. This has reflected on cheaper prices for motherboards, but could also easy the broad support of features like ECC.
So this would effectively compete against the newer Skylake/Kaby Lake Xeons, especially with them being only 4 core vs 8 for Zen.
In the same way every previous generation of AMD CPU "effectively competed" against equivalent-generation Xeons.
I think one issue is the amount, I was under the impression am4 was only going to support 64gb.
64GB is also the limit of Intels desktop i7 processors. I don't think Ryzen is intended as server processors, so 64GB should cover the target market very well.
Nope. X99 i7s support 128GB and quad channel memory. Most of X99 boards come with 8 slots as well. However the setup is way over-priced, so I'm going to preorder Ryzen on day 1 if available. It just seems like phenomenal value.
Sandy-E to Haswell-E i7's do not officially support 128gb, though there have been reports of it working.

Broadwell-E i7's now officially support 128gb.

All the Xeon equivalents (e5-1650/60,80, etc) support 256-768gb depending on the model, if paired with a consumer board that limit is usually 64-128gb.

No excuse now, Intel . Get to it.
AMD has supported ECC for years, IIRC Athlon64 supported ECC across the board, though motherboard didn't necessarily wire it.
I wonder if they have. I don't know if this is water-cooler gossip, but I have heard they are going to have another release in 6-8 months and Kaby lake is simply a placeholder.

I wonder if they are far ahead enough to drop something that beats Ryzen on every metric including price. Either way, that would push it up quick.

Evidently, while Cannonlake (the 10nm next gen chips for low-power SKUs) got punted to 2018 [0], Coffee Lake (the 14nm next gen chips for desktops) are still on for 2017 [1].

[0] - https://www.fool.com/investing/2017/02/18/intel-corporation-...

[1] - https://www.technewsinc.com/intel-8th-gen-coffee-lake-proces...

Didn't Intel release new CPUs last month? they'll be running out of lakes soon to overcharge us for 10% improvements.
Desktop Kaby Lake launched last month, portable Kaby Lake started shipping to OEMs Q2 2016.

Intel used to try for a tick or tock every year - so Sandy Bridge in 2011, IVB in 2012, Haswell in 2013, then Broadwell was incredibly late except for a handful of SKUs in 2014, Skylake in 2015, Kaby Lake (mobile) in 2016...

Really the problem is that Intel ends up having these staggered releases, probably to address limited yield on their new processes and the most impact/value (since they've been so focused on improved work per watt, and mobile cares about that for a lot more than just heat dissipation reasons...)

will there be any "cheapish"(consumer) dual socket motherboards?
AMD just launched a new interconnect, so it could be a while before motherboard makers start cranking them out. http://wccftech.com/amds-infinity-fabric-detailed/
The dual/multi-socket platform(s) are a way off. AMD is focussing on the desktop to launch Zen, in contrast to what they did with Barcelona (iirc).
Who needs a Xeon now? The only chink in the armour here is the 64GB max, which admittedly is bad for the science/Machine learning folks, but the vast majority of these chips is used for databases, web serving, app serving, etc, for which these ECC-enabled Ryzens are a game changer on price. Cluster them up on premises or in cloud, and you're looking at a total winner, especially that the 16GB dimms are still much cheaper than the 32s so you'll be able to create very decent servers on the cheap.
ECC support in CPU is just a half of success. Now somebody have to convince motherboard manufacturers to enable it. It was the same thing with previous AMD CPUs. They all had ECC support but there was virtually no motherboards that would support it.
Yes, I still remember what a PITA it was to identify a motherboard that both supported ECC and allowed enabling it in the BIOS for my last AMD build (circa 2007). I still remember reading about boards that offered "ECC support", but had no way to enable it in the BIOS, change the scrubbing time, etc. And how relieved I was that whatever I wound up with did indeed have full ECC support.

Then again, buying Xeon boards with ECC support for desktop use has been "fun" as well, but for the opposite reason. I actually had to go with a USB sound card when I built my wife's e3 xeon setup.

Asus was the most forth-coming about that. If they mentioned ECC support on a motherboard's specifications page, then it supported ECC and it couldbe enabled, etc. If they didn't, then there was no ECC support. Other manufacturers were far less informative about that. Maybe because it was rare for them to support ECC. E.g. I think that none of Gigabyte AMD motherboards had ECC.
There are plenty of SuperMicro desktop boards that include sound (and support ECC).
Yes, I'm typing this from one (X10SRA) now. But there was some drawback to getting sound in ~2011. And even when I built this one, I wasn't really happy with the selection.

A poor selection and higher prices is a consequence of fragmenting the market and making workstations with ECC a fairly rare thing.

these are the specs on an ASUS motherboard.
The 64GB limit you quote appears on one vendor's motherboard spec sheet. Given that this board also supports older generation CPUS, perhaps this limit is a limit of the motherboard, and not the Ryzen CPU itself?

Do we know anything more about what Ryzen supports in terms of memory? Eg, number of memory channels, speed, etc?

It is a limitation of the chip because it only has two memory channels. http://pcworld.com/article/3149118/computers/what-we-know-ab...