Aceshardware

Carfax · Joined: 07 Aug 2007 Posts: 40

If anyone has been keeping up with the nightmare that has become the Barcelona launch, this article from DailyTech may help shed some light on some of the controversy surrounding the launch.

At first, Dave Graham over at AMD Zone stated that the B1 chips had issues concerning the North Bridge, and that the BA stepping would be used in the retail versions.

DailyTech confirms this, but then it becomes even more confusing.

In his article, Kubicki alludes that the chips used in Monday's launch were B1 stepping, but then also states in the same article that they used B2 stepping in the preview aswell.

Apparently, the 2.5ghz parts use the B2 stepping, and are supposedly errata free.

Yet, both the Anandtech and Tech-Report preview/review seemed to have some inconsistencies in the K10's performance.

So now, I'm left wondering whether Anandtech and Tech-Report really did have B2 stepping chips, or if the B2 stepping chips also have some weird issues that would also affect performance one way or another.

someoneelse · Joined: 13 Sep 2007 Posts: 17

Lol, a nightmare? I think that's overstating it. Kristopher Kubicki isn't old enough to remember Opteron launch it seems. The B stepping Opterons had reduced performance due to SSE issues.

someoneelse · Joined: 13 Sep 2007 Posts: 17

Let me also add, if it's at 1.5 volts, it ain't B2.

Paul DeMone · Joined: 29 Aug 2007 Posts: 530 Location: Great white north

Carfax wrote:

Apparently, the 2.5ghz parts use the B2 stepping, and are supposedly errata free.

Sure as long as you don't care if integer division doesn't always produce
the correct result.

Check out errata #244 on page 12 and 19 here:

http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/41322.pdf

To update the old joke from about a decade ago - "AMD inside? Don't divide!"

AtWork · Joined: 31 Jul 2007 Posts: 139

someoneelse wrote:

Let me also add, if it's at 1.5 volts, it ain't B2.

I don't know what stepping mine are, but they have to be some of the earliest shipped and they're showing 1.205v. They are not HE series.

Temps are 29C on CPU 1 and 30C on CPU2 (at idle).

All values from Tyan System Monitor V2.2.4 with patch 2 installed.

Heatsink/fans are standard 2U. http://www.serversdirect.com/product.asp?pf_id=FN2015

I had these heatsink/fans available. They are waaay more than is needed for Barcelona.

z42 · Joined: 13 Sep 2007 Posts: 2

Paul DeMone wrote:

Sure as long as you don't care if integer division doesn't always produce
the correct result.

Or if you want accurate results in FP dont buy a Q6xxx-check errata AK20 (work around add 3 Nops )for the Intel Quad Core series (there are other AK errata all leading to FP corruption)
:roll:

The Int div bug in Barca only happens in fairly rare circumstances.

uf · Joined: 12 Sep 2007 Posts: 6 Location: Moscow, Russia

Paul DeMone wrote "AMD inside? Don't divide!".</i>

In reality, a decade ago there was serious division error in Petium. And true joke was "Intel inside? Don't divide"

Richard Van Den Boom · Joined: 13 Aug 2007 Posts: 84

uf wrote:

In reality, a decade ago there was serious division error in Petium. And true joke was "Intel inside? Don't divide"

That's why Paul was "updating" the joke....... You're were explaining it for the younger ones? ;-)

To go back on topic, Johan said he had been testing B1 stepping for his article in another thread.
I find it a bit light from AMD to send non-final stepping samples to online reviewers at launch.

Dresdenboy · Joined: 07 Sep 2007 Posts: 12 Location: Berlin, Germany

Paul DeMone wrote:

Sure as long as you don't care if integer division doesn't always produce the correct result.

Check out errata #244 on page 12 and 19 here:

http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/41322.pdf

To update the old joke from about a decade ago - "AMD inside? Don't divide!"

You forgot to mention the REP MOVS erratum. ;)

Anyway.. both instructions are vector decoded and seem to be fixable by applying a microcode patch (see the mentioning of a BIOS update). For FDIV (even on K10) this wouldn't have been possible.

I wonder that apparently no fix happened to these published errata going from rev BA to B2.

Dresdenboy · Joined: 07 Sep 2007 Posts: 12 Location: Berlin, Germany

Richard Van Den Boom wrote:

I find it a bit light from AMD to send non-final stepping samples to online reviewers at launch.

There is one possible excuse: This was not the real launch.

Richard Van Den Boom · Joined: 13 Aug 2007 Posts: 84

Dresdenboy wrote:

There is one possible excuse: This was not the real launch.

Do you mean the real launch will be when the processors will actually be available to buy?

Markace · Joined: 31 Aug 2007 Posts: 11

No, this is a more esoteric version of a paper launch; it is known as a For Advertisement Deception launch or a “FAD launch” for short.

Richard Van Den Boom wrote:

Dresdenboy wrote:

There is one possible excuse: This was not the real launch.

Do you mean the real launch will be when the processors will actually be available to buy?

tsayin · Joined: 02 Aug 2007 Posts: 26

I've never heard of a "BA" revision. Revisions typically have a letter followed by a number, with the letter changing and the number reseting to zero if the transistor layers change, but the number changing by itself if only metal layers change.

And wasn't Richard at the VP level, well above any architectural decisions (though clearly resposible for busness decisions)?

I don't know, but Dailytech never seems to actually know what it is talking about.

Groo · Joined: 22 Jul 2007 Posts: 178

tsayin wrote:

I've never heard of a "BA" revision. Revisions typically have a letter followed by a number, with the letter changing and the number reseting to zero if the transistor layers change, but the number changing by itself if only metal layers change.

And wasn't Richard at the VP level, well above any architectural decisions (though clearly resposible for busness decisions)?

I don't know, but Dailytech never seems to actually know what it is talking about.

While I do agree with the sentiment, BA is a productized B1. I am not sure what they changed other than the name, and it is the first time I have heard that particular nomenclature used, but it is real.

-Charlie

Petr Koc · Joined: 25 Aug 2007 Posts: 13 Location: Prague

This is all very strange. Anandtech tested revision DR-B2 - http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3092&p=2

As you can see, it has CPUID of 100F22h (stepping 2).

On the other hand, AMD Compare website (http://products.amd.com/en-us/OpteronCPUFilter.aspx) lists new Opterons as having revision DR-BA. DR-BA has CPUID of 100F2Ah (stepping A) what would suggest it is newer than B2 but the errata list on the other hand has more errrors in BA what basically means B2 is newer than BA.

Why would AMD send B2 to Anandtech and release BA to public? Maybe B2 is very fresh and they do not have enough processors manufactured? And maybe B2 is significantly faster and they wanted to influence results so they prefered to send B2 and not BA for review?