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

[FlanK3r]

Beep: I think the same ....

[wez]

Never heard about Cyclos before, so had to look them up.

http://www.cyclos-semi.com/news/firs...mentation.html

AMD’s 4+ GHz x86-64 core code-named “Piledriver” employs resonant clocking to reduce clock distribution power up to 24% while maintaining the low clock-skew target required by high-performance processors.

Now I have no idea how much power the clock distribution part in BD use, but everything counts I guess

[haylui]

I think AMD are just milking B2 with the pointless FX-4170 and FX-8140 releases. The extra 100 MHz won't be worth a single $ over the 8120. We'll never see a B3 stepping out of Bulldozer IMO between how hard it failed (if not in performance to some, performance per watt is dismal...), production capacity with APU/CPU balance and available wafer slots @ GloFo.

We can blame the manufacturing process all we want but it's been 6 months since retail production began for FX-8100 series, 9 months since AMD Llano retail production and apparently it has only improved 3%. To be honest, AMD and GloFo need about 10x that.

One reason I am reluctant to blame the manufacturing process is Llano's performance per watt. At a high vcore (1.4v+ stock vcore) it can still beat Phenom II performance per watt, memory bandwidth and performance per clock and include a GPU with significantly less package pins (not sure how they pulled that off!)... Bulldozer couldn't, at all.

I think it's right for AMD to skip a new stepping here and get the already reworked and finished Piledriver core on the market ASAP. I'm sure they already have early engineering samples of 8-core FX-8300 CPUs, and Trinity will probably ramp up retail production before the end of March.

i am agree with you. AMD better make Piledriver out asap...

[Tomasis]

Piledriver ready to buy under summer?

one never knows who plays with engineering sample 8300

Chew?!

[cesariuth]

http://www.techpowerup.com/forums/sh...7&postcount=14
Fake?

[FlanK3r]

yes, fake (bad making fake, look at clocks backround)

[cesariuth]

Maybe real information with bad image editing XD

[tbone8ty]

Hopefully cyclos implementation won't inhibit Overclocking

[LowRun]

Now I have no idea how much power the clock distribution part in BD use, but everything counts I guess

Cyclos said that it could be as much as 30% of the total power for a CPU and that they could shave 25% off of that but they weren't specifically talking about any CPU.

Then they said their tech had a 5 to 10% gain in power consumption for a Piledriver @ 4 GHz.

[haylui]

Cyclos said that it could be as much as 30% of the total power for a CPU and that they could shave 25% off of that but they weren't specifically talking about any CPU.

Then they said their tech had a 5 to 10% gain in power consumption for a Piledriver @ 4 GHz.

from where do you see the 5 to 10% power consumption increase @4GHz in Piledriver?

MD’s 4+ GHz x86-64 core code-named “Piledriver” is the first high-volume processor core to deploy the Cyclos RCM technology to reduce clock power while maintaining an ultra-low clock-skew target. The Cyclos resonant clock mesh relies on fully-integrated inductors to reduce average power consumption by 5% to 10% and improve peak-power-constrained performance with minimal area impact.

source:
http://www.cyclos-semi.com/technology/

[informal]

I guess LowRun meant that PD @4Ghz sees 5-10% power consumption decrease ("gain" in his post). The tech's efficency lowers when you push the frequency very high. Best results are achieved at much lower clocks which bodes well for Trinity ULV.

[tbone8ty]

4170, 6200 in the wild yet?

do you think transistor count will go down with piledriver as they refine the arch?

[BeepBeep2]

4170, 6200 in the wild yet?

do you think transistor count will go down with piledriver as they refine the arch?

Well, since the 1.2B and 2B numbers are both false, we will never know...but I don't think it will go down or anything outside of the possibility that they will add more L2 and eliminate/reduce L3.

Hopefully they get at least a "Llano" type boost (with some extra clock speed), 5-10% IPC increase and the rest in frequency.

[haylui]

I guess LowRun meant that PD @4Ghz sees 5-10% power consumption decrease ("gain" in his post). The tech's efficency lowers when you push the frequency very high. Best results are achieved at much lower clocks which bodes well for Trinity ULV.

http://www.eecs.umich.edu/eecs/about...submission.pdf (page 11)
from this document, the optimal frequency, efficiency falls at 3.3GHz which is ~25% and at 4GHz it gets 20% efficiency.

[BeepBeep2]

I'd really like to see ~4.2 GHz @ 1.2v retail chips on 32nm out of PD as stated in that document within 125w TDP...maybe up to 4.4?
AMD could potentially save themselves in the midrange / high end with PD FX, even though it's still half a year away. APUs have a place for sure, intel is moving this way as well with their HD2/3K but don't have the graphics to compete.

I intend to buy Trinity Desktop A10 whenever it becomes available...

[LowRun]

from where do you see the 5 to 10% power consumption increase @4GHz in Piledriver?

I suppose i got lost in translation and yeah i did mean a decrease in power consumption.

[sergiojr]

http://blogs.amd.com/fusion/2012/03/...9D-generation/

The score for the 2012 AMD A10-4600M on the “Pumori” reference design for PC Mark Vantage Productivity benchmark shows an increase of up to 29% over the 2011 AMD A8-3500M on the “Torpedo” reference design. The AMD A10-4600M APU has a score of 6125 and the 2011 AMD A8-3500M APU scored 4764.
The score for the 2012 AMD A10-4600M APU on the “Pumori” reference design for 3D Mark Vantage Performance showed an increase of up to 56% over the 2011 AMD A8-3500M APU. The AMD A10-4600M APU has a score of 4352 and the 2011 A8-3500M has a score of 2788.

[TESKATLIPOKA]

sergiojr who knows how credible it is.

edit: it looks credible, but I think something is contributing to the really good score, like the scheduling patch or something.

[tbone8ty]

well they went from 1333 to 1600 memory so that helps.

Projections assume A10-4600M with AMD Radeon™ HD 7660 Graphics, 4G DDR3-1600 (Dual Channel) Memory and Windows 7 Home Premium 64-bit. The AMD reference design “Torpedo” consisted of the 2011 AMD A8-3500M with AMD Radeon™ HD 6620G Graphics, 4GB DDR3-1333 and Windows 7 Home Premium 64-bit.

if i was AMD id be testing with 1866 or higher all the time. lol

[tbone8ty]

cant wait for this the FX-8350 Vishera

http://www.fudzilla.com/home/item/26...oduction-in-q3

[Bo_Fox]

Looks like Trinity will have up to 35 Voodoopower at the most, while Kaveri with its 512 GCN shader processor units, 32 TMUs, 16 ROPs will have a LOT more (only really being hurt by the bandwidth, compared to HD 7750)!!!

EDIT: For an explanation of Voodoopower ratings, just take a look at my awesome new list, linked at the bottom of my sig! The fastest Llano is rated at 25 VP.

[haylui]

Trinity APU A10-5800K early benchmarks leak?

http://www.chiphell.com/thread-387958-1-1.html

[BeepBeep2]

Trinity APU A10-5800K early benchmarks leak?

http://www.chiphell.com/thread-387958-1-1.html

Hmm, that is interesting.

That puts Trinity 1M (assuming 4.2 GHz) @ 5 / 5.3 GHz around 19s /18s and 3DMark06 CPU score @ 3.8 GHz under stock Phenom II 955 BE levels. If true, I am not at all impressed D: but GPU improvement seems great.

[FlanK3r]

L3....Vishera will be better and we dont know, how Turbo worked at A10-5800 (CPU could run superpi at 3800 MHz or so). I think Vishera could run superpi1M at FX8150 clocks around 19-19.5s (FX has 20.5s Im thinking)

[Manicdan]

we were told that L3 was not really usefull client side
and for superpi i dont know how much it would help anyway considering its single threaded.

and its just superpi, who cares anyway