Zambezi ES performance weirdness

[TheBlueChanell]

The start of back to school season is just about a month away. Llano needed to be out by then, that's a killer time for Notebooks. Release the parts now, get a ton of good press and then they'll be available by the time people are ready to shop for back to school.

Your tech people will suggest to their layman friends that they pick-up one of these notebooks when asked for recommendations. I know I will be recommending Llano based laptops to pretty much anyone looking for something in the ~$800 range. For the everyman, Llano is perfect and a great stepping stone towards something awesome.

[RussC]

The start of back to school season is just about a month away. Llano needed to be out by then, that's a killer time for Notebooks. Release the parts now, get a ton of good press and then they'll be available by the time people are ready to shop for back to school.

Your tech people will suggest to their layman friends that they pick-up one of these notebooks when asked for recommendations. I know I will be recommending Llano based laptops to pretty much anyone looking for something in the ~$800 range. For the everyman, Llano is perfect and a great stepping stone towards something awesome.

Its too late, they already missed the Back to school season for systems.

RussC

[Manicdan]

^based on graduation happening just last week?!?, lol

[duploxxx]

Its too late, they already missed the Back to school season for systems.

RussC

you must have a very strange season system then
why do you think so many OEM are already showing a nice amount of LIano laptops? they already ordered and got delivered large amount of APU. Larger then expected by AMD after also the success of brazos shipments.

[XRL8]

@RussC

Dude, i get realism.I get pessimism.
But summer hasnt even started yet and youre telling amd missed back to school season already.Stop smoking that .

[JF-AMD]

JF when will first commercial servers based around Orochi ship to customers? Will it be at the server BD launch or you're holding cards (CPU's) close to your chest before then?

I have a little server to upgrade and wondered if it's worth holding a month or two for it

First commercial servers based on interlagos will be sold, under NDA, prior to launch, as we have said in the past. I will most likely be in Asia for the launch, but don't know for sure yet.

@JF, without saying what AMD did to ES in this special "design" we wont know nothing and youre not willing to say so thats impass.

It would be good to know if there are newer samples, but i doubt JF is going to tell us.

When you make samples, they are based on the same design that you plan to take to market. You can run the fab hot, get really high speeds and really low yields. You can run the fab cold and get lower speeds and higher yields.

When you are developing silicon having a larger number of chips in your hands for validation is far more important than having the highest clock speeds. Your yield and results will tell you what you can expect from a standard distribution.

Now please give it a rest. There is no conspiracy here, people are trying to create controversey when there should be none.

[XRL8]

...
Now please give it a rest. There is no conspiracy here, people are trying to create controversey when there should be none.

Thanks for clearing that up.Thats what i have been saying all along.ES have lower clocks and thats that part about "designed for lower perfomance".
It does not however explain the weirdness of B0 ,and as you said earlier, AMD does not intentionally cripple ES.So they must be crippled on their own, ie BUG.
Its no conspiracy theory.Pure logic, based on every other ES, and your comments.I just dont get why people "got on to me" about my comments.
As to the core of the problem, informal on his blog has some rather interesting take on it.However im afraid well never know.
And yes, im giving it a rest ;-)

[zalbard]

Now please give it a rest. There is no conspiracy here, people are trying to create controversey when there should be none.

I totally agree.
It's very obvious, logical and well-explained. There is no need to discuss this further.

[JF-AMD]

Thanks for clearing that up.Thats what i have been saying all along.ES have lower clocks and thats that part about "designed for lower perfomance".
It does not however explain the weirdness of B0 ,and as you said earlier, AMD does not intentionally cripple ES.So they must be crippled on their own, ie BUG.
Its no conspiracy theory.Pure logic, based on every other ES, and your comments.I just dont get why people "got on to me" about my comments.
As to the core of the problem, informal on his blog has some rather interesting take on it.However im afraid well never know.
And yes, im giving it a rest ;-)

There is not a bug. I am sorry that you can't comprehend this fact. There is nothing more I can say because you just aren't getting it.

[almighty15]

When you are developing silicon having a larger number of chips in your hands for validation is far more important than having the highest clock speeds..

Exactly, Most important thing is to iron out ALL the kinks and the more chips you have the higher the chance that you'll catch everything.

[XRL8]

There is not a bug. I am sorry that you can't comprehend this fact. There is nothing more I can say because you just aren't getting it.

Well , you could.Just explain why they perform worse than the athlon II.Not give out ALL possibile reasons for it.

Of course if im led to believe what youre saying, it means BD will have lower performance than deneb clock for clock.Ok.

Guys.
Do you really think statement "tuned for lower performance due to lower clocks" explains the behaviour of this samples ? And not only low clocks ?
If the chip doesnt have a bug and isnt crippled (ie microcode) on purpose.It just means That IPC wise it should perform like the final product.

So yes, im probably to stupid to comprehend something explained this vaguely.
"Chip performs as expected due to lower clocks, it has no bug, nor was intentionally crippled,it is slower than Athlon II CLOCK FOR CLOCK but it isnt"

@Allmighty, no bugs, no kinks to iron out, mainboards are final, only problem could be in CPU.

[BeepBeep2]

If it is slower in new steppings than Athlon II CPC, then 3.8 Ghz, and 4.0 launch would not be enough, don't you think?
Stop making assumptions on leaked benchmarks by literally three or four people tested on boards that do not have final bios support.

Rememember someone saying UD5 vs UD7 had a big performance difference? Remember that the rest of the results came on 890 Series boards? I suspect huge Turbo/C&Q issues due to poor bios code. That has nothing to do with the CPU.

[informal]

Zambezi is designed to offer 30-50% performance increase within the same power envelope as previous design.This means that even with higher core count,it must perform somewhat better than Deneb at the same clock and it must clock higher(in order to achieve 30% better performance in ST workloads). Stop worrying about some ES results as they are irrelevant.What is relevant is end product.
Also,take a look at this article(disclaimer:BSN):

In a talk with company representatives, we learned that Patriot plans to launch DDR3-1866 kits as well. Patriot decided that they will wait for final silicon from AMD in order to qualify their parts and achieve the best possible performance. This was like listening to the same story we heard from a lot of other AMD "partners" - AMD was so tight on keeping the performance of the parts leaking that it actually delayed the performance optimizations of the upcoming products in the ecosystem (especially on the motherboard side). This is a far cry from Intel, which continuously sends performance optimized silicon or microcode updates so that partners can have a tweaked platform at launch.

Then again, AMD deliberately wanted to keep everything under wraps in order not to get beaten to the punch by a more resourceful competitor and it seems that the company really have a winner on their hands in more markets than one (Fusion Llano offers excellent GPU performance, FX to offer 1GHz+ Turbo mode).

[SEA]

XRL8

What is your point, you posting and posting all this stuff?
What you need from samples that were never meant to be any fast?
Of course they must have erratas that fixed by microcode. This lets to use them as normal chips running with no errors though slow.

Again, what you are trying to tell?
You think production chips will be just little bit faster? Everybody had that thought (but dropped it because it won't make any sense)...
Or, you you think you caught JF lying and you have to bring this to everyone?
Or, what? Just aggressively guessing?

Wait just one month.

[LowRun]

First

And yes, im giving it a rest ;-)

And then...

Well , you could.Just explain why they perform worse than the athlon II.Not give out ALL possibile reasons for it.

Of course if im led to believe what youre saying, it means BD will have lower performance than deneb clock for clock.Ok.

Guys.
Do you really think statement "tuned for lower performance due to lower clocks" explains the behaviour of this samples ? And not only low clocks ?
If the chip doesnt have a bug and isnt crippled (ie microcode) on purpose.It just means That IPC wise it should perform like the final product.

So yes, im probably to stupid to comprehend something explained this vaguely.
"Chip performs as expected due to lower clocks, it has no bug, nor was intentionally crippled,it is slower than Athlon II CLOCK FOR CLOCK but it isnt"

@Allmighty, no bugs, no kinks to iron out, mainboards are final, only problem could be in CPU.

[crazydiamond]

5 words come to mind when reading all these posts from xrl8

UN BEE LEAVE A BULL

[rintamarotta]

So amd is running fabs cold for es's, to get max yields and low clocks along with some bug in ES?
Expect retail chips to be made running fabs hot and bugs gone.

[undone]

So amd is running fabs cold for es's, to get max yields and low clocks along with some bug in ES?
Expect retail chips to be made running fabs hot and bugs gone.

I heard that you have seen/had some healthy ES, anymore info please?

[FlanK3r]

http://translate.googleusercontent.c...t1VwR0ptGtBPQw

wtf?

[RussC]

you must have a very strange season system then
why do you think so many OEM are already showing a nice amount of LIano laptops? they already ordered and got delivered large amount of APU. Larger then expected by AMD after also the success of brazos shipments.

@RussC

Dude, i get realism.I get pessimism.
But summer hasnt even started yet and youre telling amd missed back to school season already.Stop smoking that .

Mmm, the back to school season is in 6-7 weeks. Well see. With AMD its always better to be pessimistic than optimistic.

RussC

[XRL8]

@BeepBeep at least one of the tests was on a 990FX final bios mainboard.And no, i dont think AMD would launch a chip in that state.
@SEA im not trying to tell anything, as this thread title says, im trying to understand why B0 performs as it is.Nothing more.And no, JF hasnt made it any clearer,on the contrary.
@Lowrun i did leave it alone, just to be graciously attacked by JF of not being able to comprehend simple things.So i answered exactly what i mean.
@crazydiamond ,very informative and on topic.

Rintamarotta seems to think thare was a bug in this ES ,and he probably had one.
I too expect final chips to be bug free and much higher clocked!

I wont post in this thread as it seems no one is still on topic and it lost its purpose in this regard.So be happy people!
As an addendum i would like to point out that people read in TOO MUCH in what ME and JF was saying.
I didnt think it has anything to do with the chips were going to get in retail.
And JF didnt say anything about actual chips performance or explained why it performed like it was.He only listed the things that could go (theoretically) wrong.And informed that most ES`s are lower clocked due to yields.

[LowRun]

http://translate.googleusercontent.c...t1VwR0ptGtBPQw

wtf?

This is just unreadable

[Manicdan]

the only thing i THINK i understood was that AMD is better at scheduling than windows, and that drivers can improve BD performance

[nex_73]

at least one of the tests was on a 990FX final bios mainboard.

The bios needs to be tweaked for zambezi, so does the microcode. So i would not call it final.
Think they are working on AGESA rev0.0.7.6 right now.
The boards are released as "BD ready"... they will work with a BD right on, but for full support/functionality a bios update is needed...

[BeepBeep2]

@BeepBeep at least one of the tests was on a 990FX final bios mainboard.And no, i dont think AMD would launch a chip in that state.

If you dont want to post in this thread, fine, but I will tell you one thing...

Latest AGESA code for Bulldozer does not equal bios's on released boards tuned for Thuban.

the only thing i THINK i understood was that AMD is better at scheduling than windows, and that drivers can improve BD performance

Considering windows vista/7 are specifically shipped with AMD drivers from K7 to K10.5, that makes sense