ES Xeon Discussion

[RolloZ170]

The sellers mentions compatible with GIGABYTE. But are for Q0KG (or other E0 SPR-SP) , are besides Asus and GIGABYTE other dual socket or C741 boards from brands like Supermicro, Asrock known to be working or not working?
The ASRock SP2C741D16X-2T would be quite amazing 7 pcie slots on dual socket wow.

currently all C741 boards use a 806F8 MCU container supporting the following steppings:
806F4, "E0/S1"
806F5, "E2/B1"
806F6, "E3/S2"
806F7, "E4/S2"
806F8, "E5/S3/B3"

 

[Kizune]

not fully correct. nothing insinde looks at brand strings, there are working parts with "0000%@"
but they are SPR-WS ES steppings E0 and up.

I’m not saying it does. It’s just my way to figure out the earlier generations of engineering samples. Ones who identify themselves as a proper model usually pretty stable and have all the model specific features enabled. And this generation it is just a coincidence that WS motherboards accept earlier versions but do not accept latest ones. Fun fact - even Intel XTU software is tricked by that engineering sample so it starts and works with D0 SPR-SP as if it was Xeon W without X - meaning multiplier is locked. But still - that is probably the first time I see XTU not complaining about unsupported CPU and exiting right away.

 

[RolloZ170]

And this generation it is just a coincidence that WS motherboards accept earlier versions but do not accept latest ones.

after stepping D0, intel reworked the SPR silicon. maybe it is very easy, E0 and up SPR-SP doesn't like DMI4,
D0 ES2 support DMI4, but later intel decided to give C741 only DMI3.

Intel Emerald Rapids Backtracks on Chiplets – Design, Performance & Cost

Competitive Analysis Considerations, Sapphire Rapids Stealthy Redesign

www.semianalysis.com

 

[DHamov]

you don't understand.
where have i sayd that ?

Sorry my typing mistake, all these abbreviations are still new and not in my fingers.
It good to know that that W790 does not support SPR-SP cpu's.
But for example 8461V is 'scalable' but its only single socket, and also not WS
So if i understand correctly in asus w790 the d0 stepping of these SPR-SP 44xx, 64xx and 84xx cpu's are strangely enough supported but later steppings even official versions are not supported in Asus w790.

currently all C741 boards use a 806F8 MCU container supporting the following steppings:
806F4, "E0/S1"
806F5, "E2/B1"
806F6, "E3/S2"
806F7, "E4/S2"
806F8, "E5/S3/B3"

Does this mean that for these steppings one can buy any C741 board, and as long as one does not update the bios, these steppings are supported? Sorry i dont understand what a 806F8 MCU container, is and google was not very helpful.

 

[RolloZ170]

It good to know that that W790 does not support SPR-SP cpu's.
But for example 8461V is 'scalable' but its only single socket, and also not WS

HWinfo
scalable: no (don't ask)
wayness: 1S
and of course don't work on ASUS W790

So if i understand correctly in asus w790 the d0 stepping of these SPR-SP 44xx, 64xx and 84xx cpu's are strangely enough supported but later steppings even official versions are not supported in Asus w790.

yes, but i doubt it is the stepping only. W790 likes to run DMI4 with the CPU, D0 ES can, but stepping E0 and later not.
originaly intel reported for both C741&W790 DMI4, but changed C741/SPR-SP to DMI3 late.

Does this mean that for these steppings one can buy any C741 board, and as long as one does not update the bios, these steppings are supported?

i dont understand what a 806F8 MCU container, is and google was not very helpful.

i'll try to explain:
you see a microcode in the BIOS, 806F8, but hey ? stepping E3 CPU-ID is 806F6, why does it work ?

here is the microcode 806F8 platform Type 87
note "F8 06 08 00" and "05 00 00 00" is the intel x86 little endian format of 0x000806F8 and 0x00000005
HxD shows extended header of mcu.
the 05 00 00 00 is the count of entries.
then 5 supported cpuid's of this microcode.
F8 06 08 00
F7 06 08 00
F6 06 08 00 stepping E3
F5 06 08 00
F4 06 08 00

UEFITool can decode this too.

and MCE.py ( MCExtractor, most possible latest version )

 

[tesla100]

Thanks, o I wish i asked this one day before... i just ordered QYFS, but now i see that Q0KG E0 is just a bit more expensive. Maybe i can still change it. But strange that E0 should work, on the ASUS dual socket board, while a while ago you wrote : 'stepping E0 was tested, doesn't POST on W790 Ace...' for the single socket board. And for D0, things are the opposite (works at single socket w790, not on dual socket). Quite a puzzle all of this, at least to me.

Hi, can I ask where you ordered it and for how much? Thanks

 

[DHamov]

Hi, can I ask where you ordered it and for how much? Thanks

I have send you a PM with the links.

 

[DHamov]

i'll try to explain:

I tried to understand, and i will try again. But until now it was not clear. Maybe i need some more sleep, and probably also some extra education. But one thing became clear by the language settings of your screenshots.

currently all C741 boards use a 806F8 MCU container supporting the following steppings:

Also includes the single socket C741 right?

 

[RolloZ170]

currently all C741 boards use a 806F8 MCU container supporting the following steppings:

Also includes the single socket C741 right?

sure. single socket boards usualy use same BIOS but hidden second socket settings.
you just ask "will E-Steppings work ?"

 

[RolloZ170]

But one thing became clear by the language settings of your screenshots.

yes, german blood, like D. Trump LOL.
have corrected that, now in english.

 

[RolloZ170]

I saw mentions of higher TDP's for some 4677 processor like 385W.

only custom OEM (confidential SKUs for selct customers )

 

[DHamov]

i'll try to explain:

It is really interesting, but sorry i am afraid that is all too difficult for me to understand.
Can you also modify Bioses if you know these microcodes?

 

[DHamov]

But last week friday night there were a bunch of coincidence offers and discounts that seemed very good which made me decide in a hurry a bit to go for QYFS + Asus W790. But honestly, i am still dreaming of a dual system with E stepping. But no one on the web was writing about experiences with that. And i also do not really have the time to take the risk.

One other consideration is that some people say that ES steppings might have a short life span and bad reliability. To those with long years experience with them. How true is has that been in the past?
Were there particular ES series that started to become unreliable, unstable in time, as i they were aging? My experience with normal hardware is that usually when i buy it new i test it hard with some torture test, and if it survives, it usually survives the rest of my usage. But maybe that does not apply to ES cpus?

 

[RolloZ170]

One other consideration is that some people say that ES steppings might have a short life span and bad reliability. To those with long years experience with them. How true is has that been in the past?

If you're afraid of that, you have to buy retail processors, but then you only have 3 years of peace.
even a brand new processor can die in the first 3 month.
ES we can buy are made for engineers to build motherboards and write drivers, with a unstable ES this can be go to very hard work thought.
but indeed some series of ES got issues after some months, remember the BUG in the intel C2000 series ? was not ES....

 

[RolloZ170]

One other consideration is that some people say that ES steppings might have a short life span and bad reliability. To those with long years experience with them. How true is has that been in the past?

forgot one thing: some sellers of ES provide warranty.
if you looking at long lifespan do not buy (worn out) used prod.units(retail) cheap at ebay.
and instead of QYFS for $400 you better go with retail Platinum 8470, its about $12000 usd.
of course there is a risk using ES, but some have no other chance

 

[Civiloid]

One other consideration is that some people say that ES steppings might have a short life span and bad reliability. To those with long years experience with them. How true is has that been in the past?
Were there particular ES series that started to become unreliable, unstable in time, as i they were aging? My experience with normal hardware is that usually when i buy it new i test it hard with some torture test, and if it survives, it usually survives the rest of my usage. But maybe that does not apply to ES cpus?

Main problem with ES is that I think changelog (list of bugs) fixed in each stepping was distributed under NDA to customers who were designing their systems. Each stepping means that something was fixed in between, but what exactly and how important that is - you probably won't know (unless some one accidentally publishes changelog for the stepping or will tell you in private what to expect).

It can be reliability problem, it can be short life span, or it can be reduced performance or it can be something in a case which you'll never use in your system and you won't even notice a difference between D0 and E5 for your particular workload.

Then question is if you want to risk that or not.

I myself think that for homelab/test setup that is acceptable risk (as there is no other way to get enough cores, etc for that amount of money, and in worst case you'll have a downtime of that setup, so what if it doesn't cost you reputation and revenue?). If I'd be a startup that need HW for something - probably I'll consider the risks of getting ES system (especially on early stages). If I'd be a serious company - I'd probably wouldn't go for buying ES now.

You can tell that probably later steppings should have more bugs fixed and if you'll have something it probably will be a performance issues under some conditions and not stability problems, just based on a common sense, as it was said above - ES2 and later were used by devs to develop motherboards, drivers, etc. so they should be stable enough for that under most common conditions otherwise their job would've been a nightmare (but that is common sense, not experience).

The choice is always yours. But you just have to be aware that there are risks involved here.

 

[scouzi]

Main problem with ES is that I think changelog (list of bugs) fixed in each stepping was distributed under NDA to customers who were designing their systems. Each stepping means that something was fixed in between, but what exactly and how important that is - you probably won't know (unless some one accidentally publishes changelog for the stepping or will tell you in private what to expect).

It can be reliability problem, it can be short life span, or it can be reduced performance or it can be something in a case which you'll never use in your system and you won't even notice a difference between D0 and E5 for your particular workload.

Then question is if you want to risk that or not.

I myself think that for homelab/test setup that is acceptable risk (as there is no other way to get enough cores, etc for that amount of money, and in worst case you'll have a downtime of that setup, so what if it doesn't cost you reputation and revenue?). If I'd be a startup that need HW for something - probably I'll consider the risks of getting ES system (especially on early stages). If I'd be a serious company - I'd probably wouldn't go for buying ES now.

You can tell that probably later steppings should have more bugs fixed and if you'll have something it probably will be a performance issues under some conditions and not stability problems, just based on a common sense, as it was said above - ES2 and later were used by devs to develop motherboards, drivers, etc. so they should be stable enough for that under most common conditions otherwise their job would've been a nightmare (but that is common sense, not experience).

The choice is always yours. But you just have to be aware that there are risks involved here.

I agree. This should not be used in production. You could even have wrong results! (this has happened in the past with Intel). But for a personal HP desktop - it's worth the risk. The fact that Gigabyte and others (and they most likely know the bugs) maintain the CPU IDs in their BIOS probably means it's stable enough to be kept there. For a small shop that is trying to get ahead, perhaps hedge your bets and compete with deep deep pockets with a mix of production grade CPUs and ES for dev/test. I would also mix different models of ES.

 

[RolloZ170]

The fact that Gigabyte and others (and they most likely know the bugs) maintain the CPU IDs in their BIOS probably means it's stable enough to be kept there

It's strange that this stepping D0 cpuid 806F3 is marked as production.
cpuid 806F3 was first PRE C stepping
after that stepping, Intel restructured the silicon.

 

[Civiloid]

It's strange that this stepping D0 cpuid 806F3 is marked as production. after that stepping, Intel restructured the silicon.

I might be wrong, but there were talks that Intel shipped some CPUs to early adopters and then found bug or series of bugs that resulted in redesign, but that didn't affect those adopter's workloads.

Probably that means that those adopters still use D0 CPUs in production because bugs that were found are just irrelevant to them and Intel is not forced to support it until customer moves away to something newer.

But I might be just wrong, vaguely remember reading that in news year-ish ago.

 

[RolloZ170]

I might be wrong, but there were talks that Intel shipped some CPUs to early adopters and then found bug or series of bugs that resulted in redesign, but that didn't affect those adopter's workloads.

intels redesign improved the die count of a wafer. some bugs kept in, but most can resolved by microcode, unfortunaly stepping D0 is not supported by intel since 19.04.2022

Probably that means that those adopters still use D0 CPUs in production because bugs that were found are just irrelevant to them and Intel is not forced to support it until customer moves away to something newer.

i doubt that. is more likely: intel thought ok ! we are ready, this is PRD stepping, we launch soon.