Sandy Bridge-E should hit C2 stepping after launch

[onethreehill]

We knew that Intel's Sandy Bridge-E chips were big and complicated to make, but what we didn't expect was that Intel was going to continuously keep running into problems. The current C1 stepping is apparently having problems with VT-d, not a major issue for most consumers' maybe, but it's a huge problem when it comes to Xeon chips and as such Intel is working on the C2 stepping to fix this "errata".

The only good news here for Intel, if you look at that way, is that its current Extreme Edition CPU's don't support VT-d, neither does the Sandy Bridge K-SKU CPU's, so as far as the consumer platform is concerned it looks like there won't be any further delays, but the initial batch should be C1 stepping rather than C2. The C1 stepping should be in production by now, or by the latest by next week, although we don't know when Intel will kick off production of the C2 stepping, but it's very possible it won't be until next year as qualification samples of the C2 stepping isn't expected until the very end of the year......

http://vr-zone.com/articles/sandy-br...ch/13677.html#

Update: Intel Core i7-3960X and i7-3930K CPUs Transitioning to C2 stepping in January

As previously reported, Intel's first wave of Sandy Bridge-E processors have VT-d (Intel Virtualization Technology for Directed I/O) disabled due to an errata in the C1 stepping. That issue couldn't be resolved in time for the launch but it's getting fixed with the C2 stepping which is set to start rolling out to customers on January 20th, 2012 (samples have already been delivered).

The CPUs moving to the C2 stepping are the hexa-core Core i7-3960X (3.3 GHz) and Core i7-3930K (3.2 GHz). Beside the fixed VT-d, the C2 chips will feature new S-spec and MM numbers so a BIOS update for current motherboards will likely be required......

http://www.techpowerup.com/156625/In...n-January.html

[lutjens]

This is disturbing...I was looking forward to Romley-EP this quarter...no way they'll release it without a working VT-d implementation. What's also disturbing is the report on the price increase on X79...no reason for it other than to bend us (and motherboard makers) over. I really wish AMD would get their crap together and pose some sort of competition. This is the type of crap we can continue to look forward to from Intel if AMD can't....

[alfaunits]

Most consumers couldn'r care less - any of x86/x64 CPUs produced today are more than enough for them, so whether the new VT-d stuff is cheaper or more expensive is of little to no interest.
The high end stuff is only of interest to some of us at XS

[M.Beier]

Hmm, no VT support for workstation solution???? That sounds wrong... It has been supported for years, and very essential when emulating OS'.

[Andi64]

Hmm, no VT support for workstation solution???? That sounds wrong... It has been supported for years, and very essential when emulating OS'.

You are talking about VT-x, which is supported, and it's essential for virtualization. VT-d is not supported, and it's Intel's name for IOMMU virtualization (in short, having DMA on a Virtual Machine.)

Is not as important as VT-x, but those are Workstation CPUs, so it should be available at launch...

Even the mobile parts have it.

It would be nice to know the details of the errata. Maybe it isn't completely broken, but underperforming or something...

[xdan]

This news is written very biased by techpowerup.
VT-d is more for servers. VT-x is for desktop. 2500K, 2600K, 980X have VT-x but don't have VT-d
http://ark.intel.com/products/47932/...GTs-Intel-QPI)
Also VT-d is an extension of VT-x so the virtualization it remains.
I don't heard anybody playing games with servers...
VT-d it's more about using multiple OS's and running programs and aplications from secondary OS, with more speed.
An to quote more explaning :

This part of the "news" is your own addition and has a nice scent of bias. You can say that it's your own analysis of the news bit on VR, but there are a few flaws with it.

The most important thing to note is that VT-d is not present on desktop platforms. Not only the CPU has to support VT-d, but also key motherboard components (NB/SB). On desktop consummer products these "features" are disabled. They are also disabled on desktop consummer CPUs.

Enterprise customers are not "most buyers" and they do know exactly what features they need and what they don't need. Also they would not normally buy a consumer grade desktop computer. They will buy enterprise grade workstations and servers, which is the Xeon brand, with Xeon compatible motherboards that will support VT-d if they indeed need this feature.

Games, like "first person shooters", are not an really what servers are for. Your trying to plant an idea in the minds of gullable TPU readers (I really hope they buy Bulldozers, I don't want them on the Intel camp), that SB-E has some major flaw that will affect their gaming performance. In reality this doesn't concern anyone since VT-d is disabled at the hardware level on consumer grade desktops.

So why is a bug in a disabled feature important? It's not.

By the time we get Xeons in the channel, they will all be C2. There will be no need for any recalls and "free replacements".

[PatRaceTin]

feel like playing 1366 when it launch in 2008

[Andi64]

VT-d it's more about using multiple OS's and running programs and aplications from secondary OS, with more speed.

No. It's not that. What is failing on these CPUs is VT-d, you can still run guest OS's, a bunch of them, and run aplications from them with almost no performance hit. I have at least 10 Virtual Machines at home on three hosts (Phenom II X6, Athlon II X3, A6-3500) and they don't support AMD-Vi (Intel VT-d).

VT-d enables DMA on a Virtual Machine. It has nothing to do with CPU and Memory virtualization, but with devices virtualization. You can't install specific drivers without it, that's why a VM has a crappy VGA, and legacy "hardware".

[qcmadness]

This is a blow for Intel to defend against Bulldozer.

[FischOderAal]

This is disturbing...I was looking forward to Romley-EP this quarter...no way they'll release it without a working VT-d implementation. What's also disturbing is the report on the price increase on X79...no reason for it other than to bend us (and motherboard makers) over. I really wish AMD would get their crap together and pose some sort of competition. This is the type of crap we can continue to look forward to from Intel if AMD can't....

AMD is getting "their crap together". Their goal is to finally make some profit, not compete with Intel performance-wise. If you compare the amount of R&D Intel and AMD is spending then AMD is doing fairly well...

[M.Beier]

Thanks for clearing that up for me guys, and you are spot on right.

[lutjens]

I don't heard anybody playing games with servers...

Bite your tongue... Some of us crazies actually do game on a server based system (as well as work on it when we're forced to)...

Whether or not VT-d is needed by most people, Intel will not ship a 2S or 4S processor without it (ie: release a flawed processor). Especially considering the price of 2P and 4P processors...people would be very annoyed paying premium prices for a flawed processor. The issues with VT-d may fly in the 1S market (where the feature isn't present to begin with), but not the others.

[Lanek]

Right with you, and this is where is the problem for Xeon processors ( if confirmed ) and why they will be delayed. Now for SB-EP so I7 3960-3930, the issue should get less impact for enthusiast, gamers, benchers.

Anyway i find a little bit strange this infos is falling just after IDF show. But it is maybe explain why Intel have remove a part of his conference and speak mostly about IvyBridge, 22nm etc.

[lutjens]

I just wonder how many C1 stepping EP Xeons with this issue were stockpiled for launch and now all have to be reprocessed (rebranded) and castrated to convert them to i7 or Xeon 1P chips. At least the 1P market will have decent supply...

Maybe Intel will forget to neuter their 2P ability and just go ahead and unlock them...

[[XC] Synthetickiller]

Vt-d is also really nice for hardware passthrough to VMs. (I think that's been said, but I'm just dumbing it down.)

If you need a cheap Vt-d solution, either Xeon E3 or Opteron (G34). LGA2011 would not be a good solution compared to those offerings (price wise mostly).

X79 just doesn't seem to be a good solution for anyone at all, besides benchers, at least till ivy comes out....

[paulbagz]

This is a blow for Intel to defend against Bulldozer.

Gotta see BD do any damage first

-PB

[Mumak]

VT-d is a kind of extension to improve performace on virtualized systems. It's not mandatory for whole VT, but for TXT it is.
VT-x is a pure CPU thing, VT-d virtualizes I/O operations (DMA, Interrupt and IO device remapping), so it's more related to the IIO part of CPU.
The bug found in C1 is related to a semaphore which doesn't work properly and because there's no workaround for this erratum, it must be fixed with a new stepping.

[kromosto]

VT-d is absolutely not an important feature for me but i dont like getting broken stuff even if the broken part is something i dont use. So will wait for c2 stepping with my good old 920 i can even wait for c3

[Mumak]

Every CPU (including production units that you have inside) has tens of various bugs. Some of them are considered serious enough that they need a fix for production level and some not. Some errata can be fixed using a microcode update, some using a BIOS or driver workaround and some require a new stepping. For example there are lots of errata in production units that have been considered not to occur in real world (only under specific conditions / synthetic testing), there are also lots of specification violations considered not serious enough (we can live with that)... there are also public (found in specification updates) and secret errata..

[R101]

This function is only supported on server motherboards, or do I get this wrong?

[lutjens]

Every CPU (including production units that you have inside) has tens of various bugs. Some of them are considered serious enough that they need a fix for production level and some not. Some errata can be fixed using a microcode update, some using a BIOS or driver workaround and some require a new stepping. For example there are lots of errata in production units that have been considered not to occur in real world (only under specific conditions / synthetic testing), there are also lots of specification violations considered not serious enough (we can live with that)... there are also public (found in specification updates) and secret errata..

Thanks for the information and insight...very good to know. I imagine that each bug when found, its impact is very closely examined before being allowed to continue to exist in production chips. More than likely quite a few high ups would be needed to sign off on the depriorization of a bug, especially in a Xeon. Would this be accurate?

[Olivon]

VT-d is a kind of extension to improve performace on virtualized systems. It's not mandatory for whole VT, but for TXT it is.
VT-x is a pure CPU thing, VT-d virtualizes I/O operations (DMA, Interrupt and IO device remapping), so it's more related to the IIO part of CPU.
The bug found in C1 is related to a semaphore which doesn't work properly and because there's no workaround for this erratum, it must be fixed with a new stepping.

Thanks Mumak

[Mumak]

This function is only supported on server motherboards, or do I get this wrong?

VT-d is supported on any platform.

Thanks for the information and insight...very good to know. I imagine that each bug when found, its impact is very closely examined before being allowed to continue to exist in production chips. More than likely quite a few high ups would be needed to sign off on the depriorization of a bug, especially in a Xeon. Would this be accurate?

Yes, each bug is closely investigated. When a workaround is found to be sufficient, there's no need to perform a silicon change and most bugs are fixed using software workarounds (BIOS or driver) or CPU microcode/uncore updates. There are certain criteria that a chip must met in order to pass a milestone (ES0, ES1, ES2, Pre-ES2, QS, Production). During the ES-phase bugs are usually called Sightings and those which are accepted for production are then 'promoted' to Erratas

[Andi64]

This function is only supported on server motherboards, or do I get this wrong?

It's supported on AMD 890FX (kinda, it depends on the board manufacturer) and on 970, 990X, 990FX. I mean "supported" as in "it should work and the option is available", but I doubt AMD, or any virtualization company, will give you support and troubleshooting with those Chipsets.

[onethreehill]

Intel Core i7-3960X and i7-3930K CPUs Transitioning to C2 stepping in January

As previously reported, Intel's first wave of Sandy Bridge-E processors have VT-d (Intel Virtualization Technology for Directed I/O) disabled due to an errata in the C1 stepping. That issue couldn't be resolved in time for the launch but it's getting fixed with the C2 stepping which is set to start rolling out to customers on January 20th, 2012 (samples have already been delivered).

The CPUs moving to the C2 stepping are the hexa-core Core i7-3960X (3.3 GHz) and Core i7-3930K (3.2 GHz). Beside the fixed VT-d, the C2 chips will feature new S-spec and MM numbers so a BIOS update for current motherboards will likely be required......

http://www.techpowerup.com/156625/In...n-January.html