AMD "Piledriver" refresh of Zambezi - info, speculations, test, fans

[Oese]

mhm even though in coolaler forum people insisted this is piledriver, "more and more folks" point to just c0 bulldozer with some enhancements.This would fit with only 5-7% more perf/clock for the c0 version that was determined by means of the coolaler benches. If so, we can hope for a later AM3+ piledriver upgrade.

[Alex-Ro]

Actually if this is not piledriver,and just another bulldozer revision,i think that amd will launch piledriver together with a new chipset under a different socket...

[Oliverda]

Actually if this is not piledriver,and just another bulldozer revision,i think that amd will launch piledriver together with a new chipset under a different socket...

No new socket for Vishera.

[Olivon]

Better late than never ? AMD Steamroller, not Piledriver, to be the 'saviour' - VRZone

Now, there's no 'Vishera', the Piledriver-based 4-core block (8-core in AMD speak or 4-core multithreaded in Microsoft Task Manager) replacement for the infamous 'Zambezi' Bulldozer desktop out yet, and the desktop Trinity flavours, some of which our guys have in samples, are also postponed without clear, if any, release date.

[FlanK3r]

mhm even though in coolaler forum people insisted this is piledriver, "more and more folks" point to just c0 bulldozer with some enhancements.This would fit with only 5-7% more perf/clock for the c0 version that was determined by means of the coolaler benches. If so, we can hope for a later AM3+ piledriver upgrade.

Remember 2 things:
1)NEVER!!! could be refresh of architecture better than about 5% clock per clock. And at the same process. Look at Sandy Bridge and Ivy Bridge. Ivy Bridge is refresh at lower process and still, clock per clock without turbo is IB better only about 3-5% only. Higher improvement could be only at new based cores as Steamroller or so.
2) Vishera is Pildedriver based core with L3+next instruction

[Oese]

imo that author doesnt know about the ES floating around??

[Oese]

http://www.planet3dnow.de/cgi-bin/ne...?id=1345294802

here it stated that piledriver actually is c0 bulldozer.. so yeah, thats it^^

[Evantaur]

if piledriver is a complete failm, I'll have to think about converting my gaming rig fully evil . Some day i come home from work and turn my computer on and it's like "i'm sorry Evan but i can not do that" and attempts to kill me

so yeah please be atleast better than thuban

[FlanK3r]

No, no, Vishera (codename in CPU-Z) is CPU with Piledriver CPU cores, Zambezi is CPU with cores "Bulldozer". Simply.

[demonkevy666]

I'm not buying that as "piledriver" cores.
The gain's aren't enough over current FX.

[FlanK3r]

There is maybe confusing, because reference design Vishera is the same as Zambezi.

http://translate.google.cz/translate...%26start%3D360

[Oliverda]

+10% IPC & +10% clock. It means approximately +20% over FX-8150.

[radier]

In your dreams. This architecture is a crap and nothing will change that.

Tap tap tap - SGS 2

[informal]

In your dreams. This architecture is a crap and nothing will change that.

Tap tap tap - SGS 2

Thank you for your "valuable" input.

[FlanK3r]

In your dreams. This architecture is a crap and nothing will change that.

Tap tap tap - SGS 2

this is not true. If you think, u have the best CPU in your signature, dont have it (dont be proud!). I have many,many CPUs, example i7-3930K and others, and I dont say, Bulldozer uniqarchitecture is crap. Integer performance is very good, some multithreads aplications are awesome with FX or minimal average. There is only worse FP unit and x87 operations example.

[sdlvx]

In your dreams. This architecture is a crap and nothing will change that.

Tap tap tap - SGS 2

If you're going to make a claim, please provide evidence. I think Bulldozer is a fantastic idea with half of the equation poorly implemented. Look at single core performance, it gets slaughtered in some situations. In others, it beats Sandy Bridge. This is because the module approach is extremely good and it lets Bulldozer extract a lot of performance out of horrible cores. Imagine if the core part of the module approach had the IPC of a Core CPU. The multi-threaded performance would be insane and it would destroy everything as long as the performance ratio between single core and multi-threaded stayed the same. The problem with Bulldozer isn't the module approach, it's the core design of each module. The module design is fantastic and if it wasn't for that, we'd have a regular quad with the IPC of Bulldozer, and it would be horrific.


Anyways, cpu-world claims Piledriver desktop is family 2xh
http://www.cpu-world.com/news_2012/2...rocessors.html

But look at the leak, family 15.
http://www.abload.de/img/536766_1727915561793652s89.jpg

Pretty clear this coolaler leak isn't a real piledriver unless cpu world is wrong. CPUZ reads CPU registers right? It can't be wrong about family number.

[informal]

Bulldozer idea is great but the reason for the "poor" core performance(in some workloads) is exactly the module idea-it's about making some (performance) sacrifices in order to make a product that is good for some workloads and not so good in other(not important to AMD workloads). The non-important workloads are the the ones in client segment where ILP is still very important while TLP is not so important. But this space is changing and more and more applications(games included) will become multithread aware. Single core performance will become unimportant since after certain point you only get diminishing returns. Sure, AMD (and intel) will try and squeeze out every little they can but they won't do it if it will worsen the perf./mm^2 or drastically increase time-to-market and/or production cost and product complexity. They are here to make money with as little investment as possible,which is logical approach in ANY business model.

[alpha0ne]

In your dreams. This architecture is a crap and nothing will change that.

Trying to increase your post count ???
It is better to hold your tongue and let ppl think you may be foolish, than to speak and remove all doubt

Thank you for your "valuable" input.

Beat me to it

[cezar]

Found this on the other forum:

Today someone asked AMD's Phil Hughes on twitter about last Vishera rumours. And this is his answer:

we don't comment on specific rumor/speculation but in terms of Vishera and other 2012 products, there have been no changes.

http://twitter.com/AMDphil/status/238044301588848643

[FlanK3r]

I said it...

[radier]

You don't expect that he would tell you the truth if there are some problems, do You?

[FlanK3r]

there are not problems with Piledriver cores or Vishera CPU, because...because I can not tell you more ... ,-)

[tbone8ty]

I think the most positive thing about vishera is gonna be power consumption........relative to bulldozer of course lol

[Miwo]

Will it be an improvement over Llano 32nm Power? Idle Power consumption is fantastic, but wow - load is terrible and exponentially bad once more volts goes into it

[The Stilt]

Bulldozer is not an exceptionally power hungry or hot chip..
At stock speeds and rated voltages.

Instead Bulldozer is a chip with a MASSIVE nominal leakage current.
I do not know if this is because of the botched design process by AMD or a generally very botched manufacturing process of a another company. Actually I don´t even care. However the highly varying leakage levels between the sequential production batches, plus the fact that a design previously proven good on another manufacturing process has similar issues.. It should give you a clue.

Bulldozer becomes hot and power hungry once the critical supply voltage level has been exceeded (varies between different silicon characteristics). The leakage current increases rapidly along with the temperature and a chain reaction begins: 1°C increase in the die temperature can increase the leakage current up to 2%. Since the increased leakage still increases the temperature and the leakage is highly temperature induced... You can figure out the rest.

Trinity APUs have excellent power efficiency in idle and load.
It is probably the most 'intelligent' chip ever designed. The "guardian angel" called Bi-Directional Application Power Management (BApm) is pretty damn smart. It is constantly seeking equilibrium between the maximum performance and optimal power consumption. E.G if you encode X264 while reading XS, BApm will allocate most of the thermal entity for CPU cores i.e they run at highest available performance state. While playing games the BApm will constantly balance the allocations per the requirements (i.e higher GPU or CPU intensivity required). See more from the slides.

And yes, there is still room for CPU & GPU overclocking in the TDP allocation