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

[tbone8ty]

nice stilt, very informative


so if they cancel whatever is after Vishera AM3+, would they consider a 6 or 8 core APU on 28nm? would 28nm give them enough die area? i can maybe see a 6 core but 8 core would be pushing it.

regardless i hope Jim Keller is working overtime to get them in the right direction and polishing the crap out of steamroller for 28nm

[Oese]

How are the expectations here about non-APU desktop-produkt after Vishera?

We know, Kaveri will only come to desktop as APU. What do we know about server/workstation opterons/FXs after Vishera?

[cezar]

We'll see AMD business-class "Trinity" APUs on September 11:

http://www.cpu-world.com/news_2012/2...the_works.html

[AbortRetryFail?]

We'll see AMD business-class "Trinity" APUs on September 11:

http://www.cpu-world.com/news_2012/2...the_works.html

In the States at retail for $480US we already have the HP Pavillion P7-1235 with 65w Trinity A8-5500 APU 3.2/3.7GHz Turbo and HD7560D APU graphics engine.

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

Not sure if you are asking Llano<-->Vishera but AMD has noted significant increases in efficiency for AM3+ Piledriver. I'm not going to say for fear of putting the whammy on the them, but it looked really nice.

[Oese]

I wish good luck, really. I like the APU strategy. Lets hope AMD can leverage HSA enough...

[cezar]

"... available in late Q3 2012" - Nice

[tbone8ty]

R there gonna be lower piledriver models? Or r they gonna keep 8120/8150 for awhile

[cezar]

"27 processors will sell for less than before starting on August 27, 2012":
http://news.softpedia.com/news/AMD-A...r-288105.shtml

IMHO AMD is preparing market for upcoming Piledrivers

[sequoia464]

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

Come on now, you know you want to tell us everything you know.

Anything you have to say here will, probably, be held in the strictest confidence?

Seriously though, this is good to hear as there are rumors floating around that Vishera is going to be cancelled.

http://vr-zone.com/articles/better-l...#ixzz24Vr6N17f

[informal]

VR-zone is becoming a new fudzilla- inventing "news" when they have a slow traffic periods.

[FlanK3r]

We can be in relax guys...but remember one think, 32nm Zambezi and now 32nm Vishera. Its only light refresh at the same SOI-HK process.

[radier]

I sense disappointment in the air.

Tap tap tap - SGS 2

[FlanK3r]

SB and IB is refresh too...But If you have 32nm to 22nm, u have at high frequency working tranzistors and it help a nit in IPC (not much, but a bit yes). But still, if you shutdown turboboost, SB vs IB is only 3% IPC. So, do you think, it must be much higher at AMD refresh? No, it will be similar-only my logic. There are some new instruction sets hwich could help in some benchmarks and applications.

[informal]

I sense disappointment in the air.

Tap tap tap - SGS 2

What are your expectations then? It's just a tweaked Bulldozer. If you expected a 25+% performance jump out of it,sure ,you will surely be disappointed. The chip will offer 11% base clock uplift and (rumored) no Turbo clock uplift. Also it will offer moderate(~6-8%) IPC uplift. All combined ~15% performance jump over 8150,just about enough to beat 260K in application performance (not games though). It's not going to be a monster performer but a decent one for the money.

[radier]

If they cannot bring lower power consumption especially after OC, Vishera will be same crap like Zambezi.
In present price/performance factor BD don't sux that much but it's way too power hungry. It forces users to buy high quality motherboards and efficient coolers.

[FlanK3r]

in idle is great. Load? Depend at many factors....If you dont understand FX and tweaking, P-states settings, can not debate with me about this. Load of course is not ideal, but not bad. We talking about stock, not about OC. 90% of customers are non OC clients. You buying some CPU with some TDP and some specifications. Overclocking is only bonus and nobody does not guarantee some TDP after this. I think, for you must be clear, if Vishera has higher 100% load clocks +400 MHz and the same TDP limit....

[radier]

90% of customers are non OC clients.

We are at xtremesystems.org
Don't forget about it.

[BeepBeep2]

Bulldozer sucks too much power and wastes too much LN2.

25L does not go far with Bulldozer, even at idle, it boils like crazy.

[FlanK3r]

still is not so LN2 hungry as Gulftown ...I tested it...

[l0ud_sil3nc3]

still is not so LN2 hungry as Gulftown ...I tested it...

Agreed, but Gulftown is now a few iterations behind and it is still more efficient than the FX architecture. LN2 consumption I have no idea about as I have not tested a GT under cold.

[tbone8ty]

any more leaks yet besides coolar? looks like OBR has one for testing but hasn't commented on it. though maybe i dont want to hear what he has to say lol

[itsmydamnation]

What are your expectations then? It's just a tweaked Bulldozer. If you expected a 25+% performance jump out of it,sure ,you will surely be disappointed. The chip will offer 11% base clock uplift and (rumored) no Turbo clock uplift. Also it will offer moderate(~6-8%) IPC uplift. All combined ~15% performance jump over 8150,just about enough to beat 260K in application performance (not games though). It's not going to be a monster performer but a decent one for the money.

from an abstracted architectural standpoint i struggle to even call it a tweak. From a physical implementation point of view its far more then a tweak. Also the thing to remember is this piledriver core is a different one to trinity (likely just a new steeping) with the big change being additional memory instructions for the AGLU's, but dont be surpised if there is a little bit more improvement in clock scaling then trinity.

also even by your maths its 19% perf improvement not 15%.

[informal]

Average clock improvement over 8150 will be 5.4% ( geometric mean of stock/turbo clock of FX8350 divided by the same for FX8150: 4.1Ghz/3.88Ghz~=1.054 or 5.4%). IPC uplift will be smallish,let's take AMD's 5-6% (they stated a 3rd of 15% of BD->PD perf. uplift is IPC so ~5%). So if we take a higher number,~6%, for both IPC and Turbo improvements and combine them,we end up with : 1.06x1.06=1.12. Windows 8 scheduler may bring another 1-2% : 1.12x1.02=1.146 or pretty close to what AMD's slide tells us
2d073e0585e19760585b309ba1624838.jpg

[EniGmA1987]

from an abstracted architectural standpoint i struggle to even call it a tweak. From a physical implementation point of view its far more then a tweak. Also the thing to remember is this piledriver core is a different one to trinity (likely just a new steeping) with the big change being additional memory instructions for the AGLU's, but dont be surpised if there is a little bit more improvement in clock scaling then trinity.

also even by your maths its 19% perf improvement not 15%.

I thought the biggest change was the new resonant clock mesh design?




The thing that sucks most though is that if we get the promised 15%, and we get an additional 15% each year with the new models then Intel will probably be 30-40% faster still by the release of the 2014 model. Intel isnt just going to be sitting still so AMD can catch up. So while I enjoy a lot of AMD products, we really need something major in Steamroller or AMD will be behind until middle of 2015 at the very earliest.

[tbone8ty]

hence this is why they got back Jim keller