Exploring AMD's Error Correction RAS Engineering

[@Level1Techs]

This video has audio issues during the interview sections due to filming complications. Steps were taken to try and fix as much as we could. It's not perfect though. ALSO!! 1:07 has the wrong broll! This is what it should be: github.com/mchehab/rasdaemon Sorry in advance! ~Editor Autumn

[@hydroponicgard]

Thank you for clearing this up, it just shows how much people at L1T care. :)

[@VideoReal-m3p]

A few Remarks / questions, some might be comically Funny, but never the least. Why in 2024 are people sitting down, Holding Mics ? 2 Why are we not in DDR10 phase? 3 If Error correction is correcting, Why does it fail often? I'm not Elon, but I have questions myself.

[@IRefuseToUseThisStupidFeature]

It's ok, I don't know what a broll is and I won't look it up.

[@tommihommi1]

so does rasdaemon already do PCIe errors or is it some future plan? what exactly does it *do* except filter stuff out of dmsg

[@Vincentl2189]

Intel changed my mind about "its never the cpu"

[@gerogyzurkov2259]

X86*

[@lolerie]

There is no such rule. Skylake for example literally had a bug in loop detector and they had to fix it with a ucode

[@Marc.Google]

13-14 gen 🔥

[@mejesster]

This level of RAS features were more or less available on SPARC and POWER systems 15+ years ago. They had some incredible tech that fell by the wayside.

[@umeng2002]

Do you ever think that Wendell has only one camera and just films these segments 3 or 4 times?

[@Level1Techs]

Wendell has two cameras and usually films in one go. He does sometimes do multiple takes though. That's when Editor Grant and I get to choose which one we like better. Oftentimes, I end up mixing the multiple takes together to make 1 really good take lol ~Editor Autumn

[@Machistmo]

Shhh don’t tell

[@johnmijo]

@@Level1Techs Autumn, it's all about Trolling your boss :p

[@Marc.Google]

@@Level1Techs you do a great job Autumn and Grant. Wendell is a top tier host! 👌🏼

[@Delease]

The irony of a video about error correction having several editing errors should be appreciated. That being said, it was still very informative.

[@napoficial7123]

Thank you Wendell

[@pibyte]

Back to you Steve

[@MickeyMishra]

This really is LONG overdue. Finding the issues in the hardware is still a mess in large clusters. It really is about time.

[@abavariannormiepleb9470]

At this point Intel should apply Error Correction to their entire company.

[@dgo4490]

I love correcting errors!

[@pibyte]

Me too. I am going to marry my true ECC RDIMM one day.

[@Marc.Google]

Thank you to all the great beards out there! 🙏🏼 also, fantastic intro.

[@Sommyie]

16:18 I like that the protective wrapping is still on the laptop charger

[@treyquattro]

I was expecting some funky new Row Address Strobe tech! Good job I tuned in to find out.

[@FrenziedManbeast]

Ilove that this started out as essentially a pitch for Prestige Worldwide!

[@Marc.Google]

Boats and hoes!

[@parrottm76262]

Probably the single best video I have seen lately for fellow gray beards, like me. :)

[@birlalovesyou]

"Reliability"... "Availability".... "Serviceability" .... loved the way you said those IT jargons 😂😂😂

[@dylf14]

Thanks for covering this stuff Wendel!

[@italianbasegard]

This video is an example of “we’ll talk about that more in a bit” recursion

[@timkarsten8610]

21:09, i wonder why Wendell is surprised about that. Is it statistically relevant that the systems around that were commissioned at the same time, and similar hardware?

[@MikeBob2023]

Thank you, Wendellman! 🙏🏼

[@JohnWilliams-gy5yc]

The tangent roll is on point. Kudos to the crews.

[@zyxyuv1650]

Wendell, I'm disappointed, you did not test this enough, you need to get a hold of some depleted uranium or plutonium so we can find out how AMD's new RAS engineering holds up under isotopic decay with stray alpha particles.

[@EdToml]

Interesting. I'm a retired 'gray beard' and this video has pointed me to a place that may lead to solving AER errors on my box. Thanks.

[@insu_na]

my smol asrock server with a 5700x has had 2 MCE errors since I installed the rasdaemon 6 months ago or so, and both occurred within the CPU cache. kinda wild. no ecc errors reported from my ECC memory, but I'm not sure if that's because of flaky ECC support on Ryzen or if the memory just never threw errors..

[@NoizyCr1cket]

Rasdaemon or Rasdaman? I saw both in the video.

[@Level1Techs]

rasdaemon sorry about that editing oopsie..or engagement challenge? lol Link to what it should be github.com/mchehab/rasdaemon

[@marcogenovesi8570]

rasdamemamean

[@AnirudhTammireddy]

I only heard ramen

[@KRAVER_]

wow this is really exciting cant wait to see where this goes in a few years. RAS FTW !!

[@xani666]

...so instead of nesting mapping file somewhere accessible in firmware I need to look for more vendor specific garbage *again* ? I swear the industry does not learn shit, it's same garbage SNMP had

[@yoppindia]

they will stop support after sometime.

[@marcogenovesi8570]

>SNMP mentioned >(PTSD vietnam flashback moment)

[@xani666]

@@marcogenovesi8570 such a garbage protocol. It made some sense when devices had single digit MB of RAM but whole OID stuff quickly became utter waste of time for everyone involved

[@yoppindia]

if a particular core is faulty can they be turned off? is there similar tool for windows and what about intel? what about bus errors?

[@abavariannormiepleb9470]

PCIe Bus Errors - The Bane of my existence.

[@timkarsten8610]

Mine is Excel.

[@gijoeljamesmaldonado]

Don't know what I'm looking at, but I like it some how.

[@pl0dman925]

Amit, where does LN2 fit in and where is Steve from GN.

[@hallsofvalhalla1749]

I agree. Reproduction is definitely the most important thing.

[@andrewvirtue5048]

Ah yes. Cunningham's law. People are more willing to correct errors than to just flat out give a correct answer.

[@noth606]

Ey man! Rastaman 'ere! Err Rasdaman, Rasdafarai!

[@Ozz465]

Not something i can get into ($) but always good to see Wendell having a good time .

[@Maxjoker98]

I would pronounce Rasdaman as "Rasta man" instead of "RAS daemon", but that it mostly because it's more fun.

[@hareesh2581]

@level1Techs do we have vibs for esxi?

[@axn40]

I hope you bought some AMD shares before publiqhing this analyss 😅

[@tolpacourt]

Is it true that if a motherboard officially supports an Epyc CPU it will support ECC RAM?

[@J_..._]

Leaded solder is still used in mission critical environments, afaik.

[@JoaoPedro-ki7ct]

The previous title was fun

[@viruslab1]

can I use this software on my Intel pc?

[@tulsatrash]

Very exciting.

[@RFC3514]

I'm pretty sure it's meant to be pronounced "ras-da-man" (like "rastaman", hence the Rastafari / Ethiopian flag), not "ras-damon".

[@v4lgrind]

Donät know if I would say building dimm maps is fun... It's satisfying to complete one for sure, but the syntax is obtuse and confusing to say the least.

[@Machistmo]

Remote Access Server went through my head

[@JKGarageBMW]

Now do ADDDC!

[@icq6686]

haaahahahahah mr. W. i like the new style! lolz

[@yellowtoblerone]

bring me every camera what you mean every EVERY CAMERA ANGLE

[@cckeysify]

how about a rasdaman for your diet

[@axn40]

That's definitely not some normie interview 😂

[@marcogenovesi8570]

Wendell brings all the nerds to the yard

[@Marcel-dg3qd]

“Retraining memory” ? 🤔

[@mytech6779]

Why is it so hard to get real rdimm ECC on desktop‽ I mean without absurdly overbuilt xeon/threadripper platforms that cost 5×→10× more than a basic desktop. I only need a 4→8 core CPU and a few PCIe lanes, but good full fat error checking would be extremely valuable for the sort of data I handle. That is on top of the availability/cost of ECC-rdimm vs ECC-udimm. I could cobble up some franken-box out of Haswell-era used server parts, but then its sucking up tons of electricity 24/7 when only actively working 5% of the time, and really not well suited for doubling as a desktop so a second machine is needed… (Not to mention space for the extra box, and future disposal.)

Trillions of transistors yet no readouts for mounting pressure, plug insertion/contacts, etc ... . Motherboards still using CR2032 with it conveniently placed under the GPU or chipset heatsink. Low volume designs are stale and high volume designs are all about cost savings. Like it would be easier to have more ram with more channels if the processor was hexagonal!

[@wskinnyodden]

Hmm sue them for financial damages due to negative invalid publicity as they have basically stated you are bad memory! (Bad DIMM) hehehehe

[@cvasirocket1401]

So , its a more advanced Code Analyst?

[@ThaexakaMavro]

500 card in stock InTeL iS sElLiNg Out 🙄 Msrp is also fake

[@IRefuseToUseThisStupidFeature]

11:30 Does anyone else find the slow mouse speed disturbing? Like the delay is dragging on your soul.

[@IRefuseToUseThisStupidFeature]

More and less? What a deal! Sign me up