AMD Realizes Significant Reduction in Power Consumption by Implementing Cyclos Resona

[TakeNoPrisoners]

Why do I have a feeling this will be BD 2.0.

A new article comes out saying it will be super fast. Just like the months before BD came out. Super cereal this time guys! :whiste:

 

[wlee15]

Decent, although this still won't match Ivy Bridge.

I still don't see it as a reason to cancel the Bobcat successors though. Sure, 17W is good enough for the ultrathins (aka netbooks) market, but the bobcat line went even lower power (~5W) and more importantly SHOULD have been super cheap to produce with its die size.

They just skipped the 28nm shrink of the Bobcat presumably to free up more wafers for the GPU , Bobcat's successor Jaguar is still on the map.

 

[CTho9305]

That dude has flip-flops named after him!

Is it the one on page 7 of this document? Pulse latches are cool, but it's important to consider metrics such as hold time and not just setup+clk2q.

It's a really neat idea. Looks like they presented it today at ISSCC 2012. I'll look forward to reading the paper.

Do you have a link to any more-technical information?

 

[Dadofamunky]

Hope this helps. I really want AMD to succeed, since even though I'm an Intel guy healthy competition is really needed in this field. Also hoping Windows 8 has a beneficial effect.

 

[Abwx]

http://www.hardware.fr/news/12149/cyclos-aide-amd-reduire-consommation.html

The variable TDP reduction of 5 to 10% of total TDP
is due to frequency dependant efficency of the LC
equivalent circuit whose resonnance and likely higher
efficency is at F = 1/2.Pi.sqrt(LC) , L being the inductance
value and C the equivalent capacitance of the relevant grid.

 

[Skurge]

All of this is over my head. Could someone explain this to me in layman's terms?

 

[Vesku]

Nice to see AMD making progress on the power draw side of their CMT design.

 

[denev2004]

I do hope they can do it well. AMD scheduled to use 32nm to beat Intel's 22nm. So it is gonna be better to see new tec come into it.

 

[GaiaHunter]

All of this is over my head. Could someone explain this to me in layman's terms?

In probably a non very correct way.

One needs to send a single high frequency signal from a single source for the chip to work, what is known as clock.

Part of that energy is wasted.

Using an inductor and a capacitor in parallel, the objective is store the energy that otherwise would be lost to use it to generate a low-clock signal.

 

[krumme]

All of this is over my head. Could someone explain this to me in layman's terms?

It means faster and less power hungry processor.

Discussed here also:
http://semiaccurate.com/forums/showthread.php?t=6145&page=9

Whitepaper here:
http://www.cyclos-semi.com/pdfs/time_to_change_the_clocks.pdf

I have no doubt this will have a huge impact on BD, as it is a high-speed architecture from the start. But we havnt seen the effect of this architecture. This could be an important step of releasing more of the effectiveness of the high-speed design. We will have high clock with low power.

 

[Arkadrel]

Even if this on avg. cuts power usage down by 10%, and in some cases upwards of 24%....
we all know it wont matter.

Intel performance/power ratio is more than 10% or even 24% above AMDs (if you look at BD vs SandyBridge).


AMD's fab abilities (since their fabless) are so far below Intels, they have to try new stuff, come up with new ways of competeing.
Thats the 1 good thing out of this, it forces AMD to innovate with designs.

Implementing Cyclos Resona, sounds good, anything that get them abit closer to intel power usage has to be a good thing.

 

[Ancalagon44]

Why do I have a feeling this will be BD 2.0.

A new article comes out saying it will be super fast. Just like the months before BD came out. Super cereal this time guys! :whiste:

Yeah, thats what I'm thinking.Its definitely going to be faster clock for clock, JFAMD said. Definitely. Like no doubt. Like guaranteed. Faster than Thuban clock for clock, is what was promised.

Anyway... I'll wait and see.

 

[GaiaHunter]

AMD's fab abilities (since their fabless) are so far below Intels, they have to try new stuff, come up with new ways of competeing.

Only since?

 

[frostedflakes]

Even if this on avg. cuts power usage down by 10%, and in some cases upwards of 24%....
we all know it wont matter.

Intel performance/power ratio is more than 10% or even 24% above AMDs (if you look at BD vs SandyBridge).


AMD's fab abilities (since their fabless) are so far below Intels, they have to try new stuff, come up with new ways of competeing.
Thats the 1 good thing out of this, it forces AMD to innovate with designs.

Implementing Cyclos Resona, sounds good, anything that get them abit closer to intel power usage has to be a good thing.

To be fair, Intel's performance/power ratio was way ahead of Phenom II, and tons of people still bought them. AMD doesn't need to be as efficient as Intel, they just need to have acceptable efficiency, which BD clearly didn't for most consumers. This new technology along with modest IPC improvements (I thought I read like 5-7% somewhere) and new steppings/refinements to GloFo 32nm should help a lot. I was skeptical about AMD's claim of 15% perf/watt improvement with Piledriver, but it's looking more like they might be able to hit this target pretty easily. Time will tell, though, we'll just have to wait and see to be sure.

 

[Ancalagon44]

Sheesh for Bulldozer we need a 100% IPC improvement and a 50% reduction in power usage, at the same time. 5-7% wont do squat.

 

[grimpr]

To be fair, Intel's performance/power ratio was way ahead of Phenom II, and tons of people still bought them. AMD doesn't need to be as efficient as Intel, they just need to have acceptable efficiency, which BD clearly didn't for most consumers. This new technology along with modest IPC improvements (I thought I read like 5-7% somewhere) and new steppings/refinements to GloFo 32nm should help a lot. I was skeptical about AMD's claim of 15% perf/watt improvement with Piledriver, but it's looking more like they might be able to hit this target pretty easily. Time will tell, though, we'll just have to wait and see to be sure.

As with any first time implementation of new technology, the approach was somewhat conservative. We believe savings as large as 20% are possible with more aggressive design and analysis.

Comment from the eetimes post.

http://www.eetimes.com/electronics-news/4236592/AMD-counts-on-low-power-clock-IP

Still this is a nice thing to have, AMD licenced the tech, prices goes around 1 mill, and will use it extensively throught their entire CPU and GPU chips.

 

[Chiropteran]

To be fair, Intel's performance/power ratio was way ahead of Phenom II, and tons of people still bought them. AMD doesn't need to be as efficient as Intel, they just need to have acceptable efficiency, which BD clearly didn't for most consumers.

Bulldozer uses less power than phenom 2, except for 8150 in high load situations. (and overclocking). Idle power usage is way down, and for example peak 8120 power usage is less than similar phenom X6 CPU power usage in the same situations.

Lack of performance was the problem, not power usage, despite what many would have you believe.

 

[pelov]

To be fair, Intel's performance/power ratio was way ahead of Phenom II, and tons of people still bought them. AMD doesn't need to be as efficient as Intel, they just need to have acceptable efficiency, which BD clearly didn't for most consumers. This new technology along with modest IPC improvements (I thought I read like 5-7% somewhere) and new steppings/refinements to GloFo 32nm should help a lot. I was skeptical about AMD's claim of 15% perf/watt improvement with Piledriver, but it's looking more like they might be able to hit this target pretty easily. Time will tell, though, we'll just have to wait and see to be sure.

Well with Phenom/Thuban AMD priced their chips accordingly, so what you were giving up in terms of efficiency you were gaining in money saved when putting together the platform. As underwhelming as BD's performance is, the price of the chips is what's inexcusable. Hell, I'd buy an 8150 if it were ~$180-190. At that price it really wouldn't be that bad a buy, particularly if you needed the extra integer cores.

My guess is they increased IPC by ~10% and Trinity will be within spitting distance of Llano in that respect. Any performance improvements upon Llano would be derived from the higher clock speed and any new instruction sets. If those PCmark Vantage benchmark figures are true and those high stock clock speeds come to fruition, I think we might even see AMD beating their own estimates they gave earlier last year. But by now I think we've all learned to heavily scrutinize anything they or JF-AMD give us =P Guess we wait and see.

The GPU side is a completely different animal. There's no doubt in my mind that it's going to be awesome

 

[frostedflakes]

Bulldozer uses less power than phenom 2, except for 8150 in high load situations. (and overclocking). Idle power usage is way down, and for example peak 8120 power usage is less than similar phenom X6 CPU power usage in the same situations.

Lack of performance was the problem, not power usage, despite what many would have you believe.

Power consumption gets pretty crazy, especially when you OC them to 4.5GHz+. Not a ton worse than Phenom II X6, they could get pretty power hungry with high overclocks as well, people love to compare Bulldozer power consumption to Sandy Bridge to prove how "crappy" it is but sometimes seem to forget the fact that Phenom II weren't power sippers either and used a lot more power than SB, which is just a ridiculously efficient CPU. But being 32nm you'd expect BD to offer quite a bit better performance per watt than Phenom II, unfortunately it doesn't, in that efficiency test Tom's did it barely edges out Phenom II X6 in heavily threaded tasks.

But BD is a new architecture, uses a new 32nm process vs a very mature process with Phenom II, etc. There's definitely a ton of room for improvement here, and Piledriver is sounding promising.

 

[Riek]

Sheesh for Bulldozer we need a 100% IPC improvement and a 50% reduction in power usage, at the same time. 5-7% wont do squat.

🙄 If you believe Bulldozer needs 33% higher ipc, higher clocks, more cores and lower TDP than IvB to compete you are basically making a fool of everybody that buy's a pc.

 

[sm625]

It would take a power reduction of more than 60% to get BD to be competitive.

 

[denev2004]

Are those HCK Trees being the "resonant clock mesh technology"?

 

[Arkadrel]

If this tech works on CPUs.... any chance AMD can use it on their graphic cards as well?

Maybe 8xxx series of card? or the 9xxx ones?

 

[Arzachel]

It would take a power reduction of more than 60% to get BD to be competitive.

It needs a more mature process, less cache with tighter latencies, clocks somewhere around 4ghz and not bouncing p-states like crazy, if it still does that. The reworked threading in Win 8 would also be nice.

 

[blckgrffn]

Not really, AMD needed 2x cores vs. the Xeons to be competitive as well. It'll get them in the range of competitiveness, that's it. Piledriver isn't looking bad though.

I guess it all depends if things keep going the way of BF3 or not. It's all fine and dandy until you don't have enough cores for the needed threads.

That said, the number of people on dual cores, let alone quad cores has to be huge and anyone that would toss quads under the bus has to realize that is going to limit their market.

But, if a given piece of software is threaded to the point where all cores are really being used it would make sense that even an Intel quad might fall behind a well done AMD octocore. Haha, 2x the cores to level the playing field 🙂 btw, isn't that what cinibench etc. show right now?

In the end, I think we can all agree that a (more) competitive AMD is better for all of us 🙂