Official AMD Barcelona Thread

[terrace215]

http://www.supermicro.com/products/m...00P/X7DB8+.cfm

It is Blackford I assume. The boost in scores is most linked to the new compiler I believe. Computex troubles, maybe it's AMD being funny again like in the past. Or not.

Thanks for looking up the mobo. Yeah, compiler, and tweaking the degree of hardware pre-fetching apparently boosted int_rate nicely. So Seaburg should provide some more...


Re: Computex & "AMD fun", I don't think so... folks like to draw an analogy to the K7 thing, but this time AMD is already on record with 2.6GHz "simulations", Henri's early boasts (although he changed to pessimism in early June), PR boasts (yes some of those claims have already been busted), so what is the point in making people think you have a lot of problems, if you don't ?

It can't be good for business.

I wonder at what point in time over the next 2 months (assuming the 'launch' by mid-Aug is correct) the actual clock speeds for launch will become definitively known?

The Q2 CC & following Analyst Day in late July would provide opportunities to disclose them.

[terrace215]

Computex sandbagging is more like it.

Nah, there's no motivation for them to do that...

Besides, it was not just Computex demo parts... it was OEMs, mobo makers, and they pushed out the Phenom timing in their latest roadmap.

AMD is in such trouble they are hardly going to sandbag the whole world and try to make them believe they are in even worse trouble... there's no point to it.

If they had fast parts, they'd be seeding them everywhere right now, making sure incredible scores were leaking all over the net.

[informal]

terrace215, look at Johan's quote i posted from Ace's.

[EternityZX9]

2.4GHz? That's 200MHz shy of their "simulated benchmark" PR.

Not gonna cut it... they'll lose to Clovertown on everything but fp_rate.

Oh, you've seen the final benchmarks or the final product? Please show!

[Shintai]

Computex sandbagging is more like it.

If they sandbagged at computex they should fire everyone at marketing and management of the company. Because that would be the biggest PR blunder ever.

Look at Intel, they demo first Conroe, now Penryn and they are set on PR. It all takes care of itself for free. Massive PR for hardly any money.

Just think on what good any sandbagging would do. And then tell me.

Before K8 AMD was in full PR mode for over a year before.

[informal]

"Sandbagging" in terms of not showing production ready(and bug free) MPUs!
That's why it is such a PITA to "write" a bios for that thing,since it needed some patchwork to run.
I say again,patience,AMD has a very strong product,and in a month(if all goes as planned) we will have server chips that will be reviewed.And there always are those power hungry Asus QuadFx mobos ppl could use to test Barcelona.

[red]

informal, all that Johan has hinted at is that results could be 1-∞% better, thanks again, hype and no numbers. You are very hopeful. Tell/show me/us something we don't know BTW, those poor Quad FX owners will have to hope that Asus gets a new BIOS for it.

Assuming exact scaling with frequency, they'd get 94 with a 2.4GHz Barc...

Did you miss the part where I said a different configuration could significantly sway the score? We have no idea if AMD crippled the results or optimized them.
http://www.spec.org/cpu2006/results/...528-01177.html
http://www.spec.org/cpu2006/results/...214-00467.html
For a supposed CPU test, spec can as be seen here, dependent on many other factors.

[terrace215]

terrace215, look at Johan's quote i posted from Ace's.

All that quote strictly says is that he KNEW the Computex samples had bugs, and underperformed in some benchmark for its speed. Probably the cinebench thing.

So... okay, great. There's still the problem of Barc clockspeeds.

[nemrod]

informal, all that Johan has hinted at is that results could be 1-∞% better, thanks again, hype and no numbers. You are very hopeful.

It could look like just hype... But it'sn't Informal but Johan from anandtech. And the point is:"And no, I am not speculating or guessing :-)" so I has probably in his hand something less buggy, and probably better than 1%. We have to wait to know the truth.

[Brother Esau]

Wow they actually will materialize in Q3 hey? Hence another reason why the Inquire and Fud should be removed from the forum!

[mstp2009]

K10 has 2x the SSE units K8 has
dunno bout their speed...

Do you ever actually do your research before you post?


I hate to say it, but Shintai is right (even though he is an A$$). K10 has the same number of SSE units as K8. They are now 128-bit wide versus 64-bit. They did not double in number.


Upgrades for K10 can be found here (credit to gOJDO) - go down to the bottom for SSE improvements.

The 4 dual precision SSE OPs per cycle is referring to packing 2 x 64-bit instructions into each 128-bit execution unit in a clock cycle. Not doubling the number of execution engines.

[GoThr3k]

and since when 128 aint the double of 64?

little comparison, you said there were like for example 128 factories
but the truth is there are 64 factories which are just twice as big as before

(note,i dont say K10 has 64 SSE units btw,i can forsee that someone (inparticular) would quote this phrase and says that K10 doesnt have 64 SSE units,well duh )

[DoubleZero]

I don't know you but if you are desperate to link to an AMD apologist with funny blogs like "Core 2 Duo -- The Embarassing Secrets", "Intel's Chipsets -- The Roots Of Monopoly", "Intel -- The Monopoly Under Siege", "Anandtech Melts Down", "Q4 06 Intel and AMD: Who Won?", "AMD Q4 Earnings -- Good Or Bad?", "The Dishonesty of Overclocking", "Intel's Bluff (Q3 06 Results)", "Tigerton or Kittenton? Memory Amnesia.", well yeah, that speaks for itself

Ok, mea culpa on that... the titles look suspicious.
But i didn't link a blog i linked a article, and i stand for what i've said, after a week that went from K10 in Q2 at 2.4ghz unable to go further, to K10 in Q1 at 2.6Ghz to 2.4Ghz at lauch but just because AMD don't need more, that was a refreshing article.

[Shintai]

Ok, mea culpa on that... the titles look suspicious.
But i didn't link a blog i linked a article, and i stand for what i've said, after a week that went from K10 in Q2 at 2.4ghz unable to go further, to K10 in Q1 at 2.6Ghz to 2.4Ghz at lauch but just because AMD don't need more, that was a refreshing article.

LOL, love your logic. AMDs inability to raise clocks is turned into another hype from a fanboy.. Just like brisbanes performs better too right, thats why they are just to hit 2.7Ghz now while 90nm can do 3Ghz.

And if you cant see the site in questions uber AMD bias you are worse than all those together you call fanboys of another company.
An article or blog doesn´t matter if the writer is the same and using the same mentality all the way. But I guess its ok just to makeup all kind of things for the love of AMD. Lets just not hope it dissapoints as much as R600.

I cant wait till we see the actual product in action, besides at Computex.

Seems like

[Shintai]

LOL, you win the fanboy award!

Because your brisbane and a few others doesn´t mean everyone clocks to the sky. Just like all C2Ds dont do 3.6Ghz.

And I´m sure AMD would have replaced their prescott style 3Ghz CPUs with 65nm if they could. But again, you seem out of educational reach. i wonder what your excuse will be when a 2.4Ghz K10 aint enough, unless specfp_rate is all you do.

[informal]

Shintai,AMD needs other (90nm) fab up and running...I guess you can understand that,can you?
It means they use all they have and balance the product portfolio:90nm F3 for highest speed bins in A64 X2 and Opteron line,90nm F2 for mid range A64 X2 and single core line and 65nm for low-mid range(Bisbanes).
They do need the capacity for the K10 QC parts since the die size is quite large.
K10 dual cores will be much easier to produce and bin higher(even if it costs them some % in higher TDP specs)

[nn_step]

LOL, you win the fanboy award!

Because your brisbane and a few others doesn´t mean everyone clocks to the sky. Just like all C2Ds dont do 3.6Ghz.

And I´m sure AMD would have replaced their prescott style 3Ghz CPUs with 65nm if they could. But again, you seem out of educational reach. i wonder what your excuse will be when a 2.4Ghz K10 aint enough, unless specfp_rate is all you do.

I believe you mean to say "Just like all C2Ds dont do 3Ghz"
Because every chip has a different wall

[red]

after a week that went from K10 in Q2 at 2.4ghz unable to go further, to K10 in Q1 at 2.6Ghz to 2.4Ghz at lauch but just because AMD don't need more, that was a refreshing article.

??? 1. Where does this arbitrary 2.4Ghz K10 number keep coming from? 2. Also, how do you go about talking about Q2 first, then going about to Q1? Word to the wise, Sci is delusional and can't keep facts and fiction apart and never cites his numbers. So I guess the point of the article was that Sci thinks that AMD doesn't "need" 2.6GHz. Or maybe AMD can't clock them high enough.

Brisbanes clock better and cooler (F3 revision not included)

ok... I am not sure why the cooler part was thrown in as Shintai was talking about process performance I believe.

informal, how long is their 90nm going to run? It seems like AMD will continue to pump out their fastest K8s on it even near the end of 2007 http://images.tomshardware.com/2006/...m_dualcore.jpg
And where is the evidence that AMD will make native dual K10 rather than disabled quads?
===
http://www.vr-zone.com/?i=5078
[AMD Phenom Launch Schedule & Roadmap]

* AMD targets Barcelona to launch in Summer 2007 and expects ramp in frequency and volume subsequently. We can expect low clocked Barcelona in August timeframe
* Phenom series to be launched in Nov-Dec with DVT samples in Sep/Oct timeframe
* Phenom FX AM2+ part called FX-80 in Q4 timeframe and TDP for the FX series are yet to be determined
* 45nm Deneb, Propus, Regor and Sargas STARS cores in H2 2008
* Single-core Lima has been removed from the roadmap so single-core Orleans will be replaced by dual-core Rana in Q1 next year. The mainstream segment will shift from single core to dual core by early next year
* AMD expects to provide next schedule update in July

Nothing too new, but in a pretty chart.

[informal]

If I recall correctly ,in one interview D.Mayer said 90nm production will last until the end of 2007.
And chart you linked shows a 6000+ at TDP of 89W in Q3 which is remarkable for the die size and clocks for that CPU on 90nm process.

[Dresdenboy]

little comparison, you said there were like for example 128 factories
but the truth is there are 64 factories which are just twice as big as before

(note,i dont say K10 has 64 SSE units btw,i can forsee that someone (inparticular) would quote this phrase and says that K10 doesnt have 64 SSE units,well duh )

I think, your example is a bit off. How would you explain the parallel addition of 2 64 bit doubles in those widened factories?

IMO a better analogy is the lock-step operation of two 64 bit mem controllers to build a 128 bit mem channel. BTW, "lock step operation" is even mentioned in an AMD patent regarding the new FPU.

They really physically added 64 bit capable copies of the original subunits of the FPU and even a somewhat narrower register file (which doesn't need to support 80 bit FP values). Thus it's very likely that SSE 128 works by additionally using those extensions to do the calculations of the upper half of the values in the SSE registers.

So actually all the terms "widened execution", "doubled units", "extended FPU" fit

Here is a link to Hans de Vries' page with a K10 die plot, which names the different units. Those of the original FPU are the same as on K8, which he described in detail in his often cited Opteron article.
http://chip-architect.com/news/2007_...us_Images.html

[madcho]

FASN8 wanted here

[mstp2009]

I believe you mean to say "Just like all C2Ds dont do 3Ghz"
Because every chip has a different wall

More NN BS.

Show me a C2D that won't do 3GHz and I guarantee you it is because of a board limitation. Not the chip.

[informal]

AP
Ahead of the Bell: AMD
Thursday June 21, 9:16 am ET
Advanced Micro Devices Shares Jump in Premarket Session After Analyst Upgrades to 'Buy'

NEW YORK (AP) -- Shares of chip maker Advanced Micro Devices Inc. jumped in premarket trading on Thursday after an analyst upgraded the stock, and suggested the company has regained market share from rival Intel Corp.

ADVERTISEMENT
Stifel Nicolaus analyst Cody Acree upgraded the stock to a short-term "Buy" from "Neutral," and set a $17 price target on it.

"We believe the first-quarter's underperformance was not representative of AMD's true end demand and that second-quarter revenue will likely recover a portion of first-quarter's shortfall," Acree wrote in a client note.

Acree also thinks the Sunnyvale, Calif., company has regained one or two percentage points of market share, and believes second-quarter sales will exceed the consensus estimate for 1.5 percent sequential growth.

In April, the company forecast second-quarter sales would be flat or slightly higher. Analysts expect, on average, a second-quarter loss of 85 cents per share on $1.25 billion in revenue,

AMD is scheduled to host an analyst day on Thursday, July 26.

AMD shares traded at $13.88 in the premarket session, up 24 cents from Wednesday's closing price of $13.64. Shares of Santa Clara, Calif.-based Intel traded at $23.91, down 3 cents from Wednesday's closing price of $23.94.

http://biz.yahoo.com/ap/070621/ahead..._amd.html?.v=1

[red]

Um, isn't this the Barcelona thread?

[informal]

Yes it is,but i didn't want to spam the News Section with this one since we are talking about AMD's fab resources and manufacturing,and this is closely related with that.Seems that 90nm fab produces quite competitive products to be fazed out just yet l