AMD packs Cyclos clock tech into Piledriver
A test chip, created in 2009, was enough to convince AMD to add Cyclos' resonant clock mesh technology to Piledriver.
Making the announcement at the International Solid-State Circuits Conference (ISSCC) in San Francisco late yesterday, Cyclos claims that the x86 processing core design destined for Piledriver processors will include its resonant clock mesh technology.
The technology, invented by Cyclos, uses on-chip inductors to create an electric pendulum known as a 'tank circuit,' formed by the large capacitance of the clock mesh in parallel with the inductors. The result is a system which 'recycles' the clock power instead of watching it dissipate on every clock cycle as with traditional systems.
In other words: it's regenerative braking for processors.
'High-performance processors have used clock mesh designs for years, but with growing emphasis on power reduction in both servers and mobile PCs, the traditional approach has become too power hungry,' claimed Linley Gwennap, principal analyst of market watcher The Linley Group, in a statement released by Cyclos to coincide with its ISSCC presentation. 'This announcement proves that the Cyclos resonant clock mesh technology provides meaningful power savings in real-world products. We expect other processor designers to adopt the Cyclos technology in applications where power reduction is important.'
To back its claims of power savings up, Cyclos has some real-world figures. Based on Piledriver-based x86 processing cores running at 4GHz and above, the resonant clock mesh technology drops clock distribution power by up to 24 per cent at peak and between five and 10 per cent on average in the company's testing. Clock-skew, a serious issue in high-speed processors, is claimed to be unaffected by the drop in power draw.
Those figures are impressive, but a fair way away from the company's ARM926-based test chip created in 2009. On that platform, which ran at 250MHz as a proof of concept, the technology boasted a 25 to 35 per cent drop in power draw depending on the currently running application and its particular needs.
'We were able to seamlessly integrate the Cyclos IP into our existing clock mesh design process so there was no risk to our development schedule,' explained Samuel Naffziger, AMD corporate fellow, of the partnership. 'Silicon results met our power reduction expectations, we incurred no increase in silicon area, and we were able to use our standard manufacturing process, so the investment and risk in adopting resonant clock mesh technology was well worth it as all of our customers are clamouring for more energy efficient processor designs.'
AMD has confirmed that Cyclos' resonant clock mesh technology will be a part of the x86 portions of Piledriver on the desktop and in the server room. Cyclos itself, meanwhile, has announced plans to target system-on-chip (SoC) companies like Texas Instruments, Nvidia, Qualcomm and Intel with the same technology in order to drive down the power draw of mobile processors.
If you're wondering just how the resonant clock mesh technology works, there's a great write-up over on SemiWiki.
Previous Article
25 Comments
Discuss in the forums ReplyTrinity replaces Llano. Vishera replaces Bulldozer. If anything improved power consumption and TDP would mean Vishera will probably clock higher. At least from what I gather IPC improvements will be there although how high they will be is unknown. In fact the main improvements of the Phenom II were slightly improved IPC,together with higher clockspeeds and much better power consumption.
Anyway,we will soon find out how the Trinity CPU section performs compared to Bulldozer(which has L3 cache).
No, you have fundamentally misunderstood what it means.
By reducing power consumption the CPU designer can either increase die size (increase performance) or increase clock speeds (increase performance) and still stay within design power bands (45w/65w/95w/125w/140+w).
It means that AMD can maximise the performance of their CPU design as much as possible within their power limits.
This has no effect on whether or not it will beat or be competitive with Intel. Time will tell.
Erm, that's surly not a good thing? What about 2-2.5Ghz?
How many times do AMD have to come out publicly and say "We can no longer compete at the top end, we are focussing more on low and mid ranged solutions that offer value for money" ?
Half of the problem with Bulldozer was, one again, expectation. I do lay some of the blame for that on AMD for being a one man band and running around with their bass drum making lots of "oompah" sounds and blowing their horn, but any one who actually thought it could take on Sandybridge was crazy.
The money for R&D just isn't there any more. The last time AMD competed at the top end they lost money on every CPU sold.
As much of an AMD fan boy as I am, I'm starting to loss faith. I thought Bulldozer might have done something big and I'm keeping an eye out for the Pile driver stuff, but I think my next big upgrade will be Intel based, if it all doesn't shape up to well. 990X sort of power and high clock speed would be good enough for me, but seem they are struggling to get to that level.
What "promises" are these? Please post some links.
Or are you referring to AMD fanboy expectations? A brand of troll whose delusions are matched only by those with blue or green tee-shirts.
agreed. amd isn't promising anything here, they're just stating that they're using a new-ish technology that will make a difference. they're not saying that this will be a deal sealer.
No, I don't think I have missed the point at all. TDP mainly affects clock speeds as you mentioned, but we learned years ago that a clock race just isn't enough to compete whether on low, mid or high end CPUs. You need to develop a good architecture that actually does something with the clock speed. Who cares if a certain AMD CPU can be overclocked to hit 4 or 5GHz when it is matched in terms of actual output performance by an Intel CPU running at stock speeds of less than half? Also, high overclocking also brings diminishing returns so the chips work out less efficient too.
The only thing AMD seems to announce these days is that their chips use less power while barely improving, and with Bulldozer, sometimes inferior performance.
As for the claims from AMD saying that they aren't trying to compete, it sounds like someone in a race who gets thoroughly beaten only to say that they didn't want to win and weren't actually trying.
I can't help wonder what benefit the Cyclos technology might have in the ARM architecture.
5-10% lower average power draw is nice, but it isn't all that much. Even a 25% peak draw reduction only puts it in the ballpark of Sandbridge equivelent power draw. Ballpark, not as low as, but ballpark. That doesn't even get in to over clocking where Bulldozer is drawing nearly twice the power of an overclocked sandybridge chip.
I want AMD to succeed, I want them to do that by providing value for money products that are competitive at their price point. A die shrunk X6 1100 would have made more performance sense than an FX don't you think?
Bulldozer was going to be a kicking cpu........... :-/
OpenCl is going to replace proprietary Cuda and Physx...... :-/
HSA is going to give their platforms an advantage as only they have both advanced gpu and cpu...... :-/ we'll see but I'm not going to hold my breath or buy into their promises until I see working adoption.
This new stuff might make it a normal chip.
Precisely. Most PCs are not the overclocked racehorses on our desks. Most are cheap office drones which even at basic spec are amply powerful to run the usual office applications. Now, what would companies using thousands of these be interested in? More power, or cheaper processors that use less electricity?
This. Much as I hate to see Intel aligning themselves as the processor maker to the enthusiast market, AMDs playing this as smart as they can. While I'm sure they didn't set out to create a proc that still hasn't quite matched what Intel had 2 generations ago, that's what they've got at the moment. Rather than racing for a market that doesn't bring in huge dollars, why not aim for the PCs under every desk in the corporate world? I guarantee they'd be happier to hear my company is going exclusively AMD on 100 machines than to hear I'm quite happy with my 1 lonely little desktop (or even the reality - 2 in daily use and another being built.) And my company is tiny potatoes - we have as customers companies that have triple the admin drones that we do employees total. You think AMD wouldn't be wise to target them? They've to do something - the more time goes on, the less they're able to compete. Their best bet at this point is Intel making a major mistake with IB or Haswell, which doesn't seem likely. Even the Cougar Point bug was barely a blip on the radar.
So what is an office going to pick? The computer that has the same price, but lower power bills, lower ambient noise and maybe just a bit faster in most applications that their employees are going to be running on the machines?
Or something that can play Crysis at low settings at good frame rates without needing a discrete GPU?
I would be interested to find out if offices care about the power envelope - tier 1 pc manufacturers rarely talk about power on desktops. So, unless your tech guy is power savvy, a lot of companies don't know. PC refresh cycles are often 5 years in business - they compare to what they've got now, and it's always a better deal for them whatever they buy...