AMD Zambezi news, info, fans !

[Manicdan]

"10% more" can mean alot of things
if the module as a whole got 10% more then it could be because a single thread got stronger, or the second thread is getting 90% instead of 70% gains thus turning a module into 190% instead of 170% which is ~10%.
or it could just be a few % IPC and a few % clock rates

i honestly hope single threaded perf got 10% better IPC, without any sacrifice in clock rates (and maybe like 5% more OC room too just cause the process is more mature)

[-Boris-]

i honestly hope single threaded perf got 10% better IPC, without any sacrifice in clock rates (and maybe like 5% more OC room too just cause the process is more mature)

I think there is a strong possibility that it's going to be just like that.

[AbortRetryFail?]

IPC gains, as high as 10%, could be seen in simply using an optimized code path in certain software environments.

Just sayin' ...

[rog]

Here's a link with some good info.... if you can get around the bad German>English translation.


http://translate.googleusercontent.c...LwmU_kXexnggow

Donanimhaber is turkish not german.

[Manicdan]

IPC gains, as high as 10%, could be seen in simply using an optimized code path in certain software environments.

Just sayin' ...

its worse than that, it says based on "projections", meaning we have no idea what can really happen.

[BeepBeep2]

The FX-8120 is unlocked, right?

[Canis-X]

Yes, it has been stated that all FX chips are unlocked.

[Daveburt714]

MORAL OF THE STORY: Be careful cranking the current capacity on the IMC with Thuban....

Dude, I turned up the "Current Capacity" NOT the voltage! IMC voltage was @ 1.278 which is well within acceptable range IMO.

It was probably just a fluke, but thanks for trying to make me look stupid....

And yes, I was refering to the "uncore" which is actually called CPU-NB or IMC (integrated memory controller) on AMD systems.

[MaddMutt]

Sir,

I was not trying to make you look stupid. I'm sorry if it sounded or came out that way. If you look at my post and the numbers that I got, to me it seemed like i was getting poorer performance with each voltage increase. With my x4 955 it was the opposite, better score with more voltage. I've just started playing with my x6 and trying to get/acquire as much Knowledge from more experienced members. I apologize for many of my stupid questions, But as I have been told " You can not learn if you don't ask". I may not be putting them in the correct format or complete clarity for you MASTERS As I have said in several of my posts " Thank You for any Knowledge that you are willing to part with...in teaching me

The way I understood the post was that you had allowed to much Voltage to be sent to the IMC. From my limited knowledge, this is the cpu-nb spot in the bios. I read some were in a post that by increasing the Voltage of this can help improve performance by lowering the memory latency.


And again
Thank You for any Knowledge that you are will to part with...in teaching me

[Smartidiot89]

its worse than that, it says based on "projections", meaning we have no idea what can really happen.

I am quite sure AMD knows what IPC-gains awaits them with Piledriver, especially since they have shown Trinity up and running.

So assuming projections = "we have no clue", which I personally do believe they do, the only thing they don't know for sure is probably what clock-speeds it will get as that is the final thing they do with a new architecture. Since its a refresh of Bulldozer I am quite confident it should clock the same or even better.

[dess]

or it could just be a few % IPC and a few % clock rates

How could they know the stock clock rates (within a given TDP class) long before volume production? Also, the clock rate/power consumption ratio is to be raised even for a given die (new steppings/revs, better yields, etc.), and through the lifecycle of the uarch.

IPC gains, as high as 10%, could be seen in simply using an optimized code path in certain software environments.

That could be said about BD, as well.

Perhaps they're referring to the Converged BMI Instructions. But, would they've written "x86 performance", then?

[-Boris-]

I am quite sure AMD knows what IPC-gains awaits them with Piledriver, especially since they have shown Trinity up and running.

So assuming projections = "we have no clue", which I personally do believe they do, the only thing they don't know for sure is probably what clock-speeds it will get as that is the final thing they do with a new architecture. Since its a refresh of Bulldozer I am quite confident it should clock the same or even better.

I agree, but sometimes the final performance isn't known this early. That said I do think they have more than a clue about IPC on Piledriver at this point, just that you can't be certain. I also believe that it's mostly clock speeds they don't know at this point. And I hope that there will be a bit larger headroom for Piledriver, even if Bulldozer seems to have some promising overclocking potential.

[Manicdan]

So assuming projections = "we have no clue", which I personally do believe they do,

projections does mean WE have no clue what can really happen, and thats not the same as what THEY know and can guess on.
then factor in how that 10% is measured and again, WE have no idea what the end result will be.

[GenTarkin]

Only if AMD had marketed the octo core as an 8 threaded processor w/ 4 cores... all these benchmarks, speculations and so forth would make WAY more sense.
I knew from the the initial announcement of its release projected to be april this year...as soon as they started detailing the uarch...I had a feeling it should be marketed as a quad core rather than a freakin 8 core.
Cuz, as an 8 core it looks weak...but you change perspective to a quad core w/ 8 threads and the thing becomes a monster.

bout sums it all up!
Damnit AMD market them as quads...not freakin octo's!

[hyc]

Personally I think that Piledriver slide is pure BS. There's no way AMD would talk about that to anybody outside of AMD, before BD has even been released. Not even to big partners like Cray, it would be utterly moronic.

[rog]

Personally I think that Piledriver slide is pure BS. There's no way AMD would talk about that to anybody outside of AMD, before BD has even been released. Not even to big partners like Cray, it would be utterly moronic.

You can be sure OEMs are informed about product strategys for next year or maybe even 2013.

[Opteron146]

Hope that 10% performance increase is per clock cycle. If it's with frequency i doesn't look to good.

Should be, they are talking about the core-architecture, i.e. not about the whole chip in that paragraph. I just wonder what "digital media workload" is. Cinebench? That would mean a score around 6.60. If they could also up the clock a bit, they could reach ~7.00. Sounds good, but the chip won't be here before ~H2/12. Intel will have IB until then and they could probably clock that thing crazy @22nm.

Apart from that the whole slide seems amateurish, even for an AMD slide.

[Opteron146]

So, "bdver2" is indeed Piledriver, not Steamroller, then... (?)
Dresdenboy was wrong here, it seems: http://citavia.blog.de/2010/10/21/si...llano-9726240/

Yes and that was already known for quite some time. AMD changed plans. The roadmap from Dresdenboy's blog is simply outdated. Remember the Bulldozer / enhanced BD and BD Next Generation names, from the Bulldozer introduction slides?

20101110amd6.jpg

That's basically BDv1, BDv2 and BDv3, or if you want to use the fancy codenames: Bulldozer, Piledriver and Steamroller. Lots of names for the same .. marketing at its best.
Don't forget that these are just the names for the core architeture, the real chips have then names, too. Trinity for the APU with Piledrivercores, etc. etc.

[Assimilator87]

Only if AMD had marketed the octo core as an 8 threaded processor w/ 4 cores... all these benchmarks, speculations and so forth would make WAY more sense.
I knew from the the initial announcement of its release projected to be april this year...as soon as they started detailing the uarch...I had a feeling it should be marketed as a quad core rather than a freakin 8 core.
Cuz, as an 8 core it looks weak...but you change perspective to a quad core w/ 8 threads and the thing becomes a monster.

bout sums it all up!
Damnit AMD market them as quads...not freakin octo's!

It would be awesome if they released 5 or 6 module processors, possibly with Piledriver.

[dess]

Yes and that was already known for quite some time. AMD changed plans. The roadmap from Dresdenboy's blog is simply outdated.

The blue text is Dresdenboy's guesses.
Sure that there are big changes, really? Piledriver is still a moderately enhanced Bulldozer, not an entirely new generation one.

Remember the Bulldozer / enhanced BD and BD Next Generation names, from the Bulldozer introduction slides?

20101110amd6.jpg

That's basically BDv1, BDv2 and BDv3, or if you want to use the fancy codenames: Bulldozer, Piledriver and Steamroller.

I remembered this slide, but thought the Enhanced one is bdver1.1, or something like that...
(I'm using the bdverX format, instead of BDvX, because the former is used in some include files for a certain compiler.)

Lots of names for the same .. marketing at its best.

Well, they are not entirely the same... Indeed bdver3(?), that one seems to be much more than some little enhancements.

Don't forget that these are just the names for the core architeture, the real chips have then names, too. Trinity for the APU with Piledrivercores, etc. etc.

I won't forget, I promise! (I know, of course.)

[rog]

Should be, they are talking about the core-architecture, i.e. not about the whole chip in that paragraph. I just wonder what "digital media workload" is. Cinebench? That would mean a score around 6.60. If they could also up the clock a bit, they could reach ~7.00. Sounds good, but the chip won't be here before ~H2/12. Intel will have IB until then and they could probably clock that thing crazy @22nm.

Apart from that the whole slide seems amateurish, even for an AMD slide.

FX-8150 is scoring around 6,40 with Turbo enabled.

[Olivon]

Rog seems to be very well informed

[dess]

Rog seems to be very well informed

Rog? Wasn't it hyc who believed he himself is? There are official roadmaps and many slides out there...

And, knowing that Piledriver will be available also on AM3+ is actually a selling point for current AM3+ systems, as this way it won't be so shortlived.
F.ex. originally I was going for Zambezi, but recently thought better wait for FM2 and Trinity. Now I'm in a dilemma, again.

[rog]

Rog? Wasn't it hyc who believed he himself is? There are official roadmaps and many slides out there...

The official Slide with a 3,6 Ghz Cinebench result in it? This is without Turbo.

[informal]

Turbo won't kick in fully MT fp heavy workloads...