Thursday, June 19th 2014
AMD Details Plans to Deliver 25x APU Energy Efficiency Gains by 2020
AMD today announced its goal to deliver a 25x improvement in the energy efficiency of its Accelerated Processing Units (APUs) by 2020.1 Details including innovations that will produce the expected efficiency gains were presented today by AMD's Chief Technology Officer Mark Papermaster during a keynote at the China International Software and Information Service Fair (CISIS) conference in Dalian, China. The "25X20" target is a substantial increase compared to the prior six years (2008 to 2014), during which time AMD improved the typical use energy efficiency of its products more than 10x.
Worldwide, three billion personal computers use more than one percent of all energy consumed annually, and 30 million computer servers use an additional 1.5 percent of all electricity consumed at an annual cost of $14 billion to $18 billion USD. Expanded use of the Internet, mobile devices, and interest in cloud-based video and audio content in general is expected to result in all of those numbers increasing in future years.
"Creating differentiated low-power products is a key element of our business strategy, with an attending relentless focus on energy efficiency," said Papermaster. "Through APU architectural enhancements and intelligent power efficient techniques, our customers can expect to see us dramatically improve the energy efficiency of our processors during the next several years. Setting a goal to improve the energy efficiency of our processors 25 times by 2020 is a measure of our commitment and confidence in our approach."
"The energy efficiency of information technology has improved at a rapid pace since the beginning of the computer age, and innovations in semiconductor technologies continue to open up new possibilities for higher efficiency," said Dr. Jonathan Koomey, research fellow at the Steyer-Taylor Center for Energy Policy and Finance at Stanford University. "AMD has steadily improved the energy efficiency of its mobile processors, having achieved greater than a 10-fold improvement over the last six years in typical-use energy efficiency. AMD's focus on improving typical power efficiency will likely yield significant consumer benefits substantially improving real-world battery life and performance for mobile devices. AMD's technology plans show every promise of yielding about a 25-fold improvement in typical-use energy efficiency for mobile devices over the next six years, a pace that substantially exceeds historical rates of growth in peak output energy efficiency. This would be achieved through both performance gains and rapid reductions in the typical-use power of processors. In addition to the benefits of increased performance, the efficiency gains help to extend battery life, enable development of smaller and less material intensive devices, and limit the overall environmental impact of increased numbers of computing devices."
Moore's Law states that the number of transistors capable of being built in a given area doubles roughly every two years. Dr. Koomey's research demonstrates that historically, energy efficiency of processors has closely tracked the rate of improvement predicted by Moore's Law. Through intelligent power management and APU architectural advances, in tandem with semiconductor manufacturing process technology improvements and a focus on typical use power, AMD's expects its energy efficiency achievements to outpace the historical efficiency trend predicted by Moore's law by at least 70 percent between 2014 and 2020.
Architecting for Energy-Efficiency Leadership
Like advances in computing performance, advances in power efficiency have historically come along with new generations of silicon process technology that shrink the size of each individual transistor. AMD expects to outpace the power efficiency gains expected from process technology transitions through 2020 for typical use based on successfully executing three central pillars of the company's energy efficient design strategy:
"The goal of an energy-efficient processor is to deliver more performance than the prior generation at the same or less power," said Kevin Krewell, analyst at TIRIAS Research. "AMD's plan to accelerate the energy-efficiency gains for its mobile-computing processors is impressive. We believe that AMD will achieve its energy efficiency goal, partially through process improvement but mostly by combining the savings from reducing idle power, the performance boost of heterogeneous system architecture, and through more intelligent power management. With this undertaking, AMD demonstrates leadership in the computing industry, driving innovations for a more energy-efficient future."
Worldwide, three billion personal computers use more than one percent of all energy consumed annually, and 30 million computer servers use an additional 1.5 percent of all electricity consumed at an annual cost of $14 billion to $18 billion USD. Expanded use of the Internet, mobile devices, and interest in cloud-based video and audio content in general is expected to result in all of those numbers increasing in future years.
"The energy efficiency of information technology has improved at a rapid pace since the beginning of the computer age, and innovations in semiconductor technologies continue to open up new possibilities for higher efficiency," said Dr. Jonathan Koomey, research fellow at the Steyer-Taylor Center for Energy Policy and Finance at Stanford University. "AMD has steadily improved the energy efficiency of its mobile processors, having achieved greater than a 10-fold improvement over the last six years in typical-use energy efficiency. AMD's focus on improving typical power efficiency will likely yield significant consumer benefits substantially improving real-world battery life and performance for mobile devices. AMD's technology plans show every promise of yielding about a 25-fold improvement in typical-use energy efficiency for mobile devices over the next six years, a pace that substantially exceeds historical rates of growth in peak output energy efficiency. This would be achieved through both performance gains and rapid reductions in the typical-use power of processors. In addition to the benefits of increased performance, the efficiency gains help to extend battery life, enable development of smaller and less material intensive devices, and limit the overall environmental impact of increased numbers of computing devices."
Moore's Law states that the number of transistors capable of being built in a given area doubles roughly every two years. Dr. Koomey's research demonstrates that historically, energy efficiency of processors has closely tracked the rate of improvement predicted by Moore's Law. Through intelligent power management and APU architectural advances, in tandem with semiconductor manufacturing process technology improvements and a focus on typical use power, AMD's expects its energy efficiency achievements to outpace the historical efficiency trend predicted by Moore's law by at least 70 percent between 2014 and 2020.
Architecting for Energy-Efficiency Leadership
Like advances in computing performance, advances in power efficiency have historically come along with new generations of silicon process technology that shrink the size of each individual transistor. AMD expects to outpace the power efficiency gains expected from process technology transitions through 2020 for typical use based on successfully executing three central pillars of the company's energy efficient design strategy:
- Heterogeneous-computing and power optimization: Through Heterogeneous System Architecture (HSA), AMD combines CPU and GPU compute cores and special purpose accelerators such as digital signal processors and video encoders on the same chip in the form of APUs. This innovation from AMD saves energy by eliminating connections between discrete chips, reduces computing cycles by treating the CPU and GPU as peers, and enables the seamless shift of computing workloads to the optimal processing component. The result is improved energy efficiency and accelerated performance for common workloads, including standard office applications as well as emerging visually oriented and interactive workloads such as natural user interfaces and image and speech recognition. AMD provides APUs with HSA features to the embedded, server and client device markets, and its semi-custom APUs are inside the new generation of game consoles.
- Intelligent, real-time power management: Most computing operation is characterized by idle time, the interval between keystrokes, touch inputs or time reviewing displayed content. Executing tasks as quickly as possible to hasten a return to idle, and then minimizing the power used at idle is extremely important for managing energy consumption. Most consumer-oriented tasks such as web browsing, office document editing, and photo editing benefit from this "race to idle" behavior. The latest AMD APUs perform real-time analysis on the workload and applications, dynamically adjusting clock speed to achieve optimal throughput rates. Similarly, AMD offers platform aware power management where the processor can overclock to quickly get the job done, then drop back into low-power idle mode.
- Future innovations in power-efficiency: Improvements in efficiency require technology development that takes many years to complete. AMD recognized the need for energy efficiency years ago and made the research investments that have since led to high impact features. Going forward many differentiating capabilities such as Inter-frame power gating, per-part adaptive voltage, voltage islands, further integration of system components, and other techniques still in the development stage should yield accelerated gains.
"The goal of an energy-efficient processor is to deliver more performance than the prior generation at the same or less power," said Kevin Krewell, analyst at TIRIAS Research. "AMD's plan to accelerate the energy-efficiency gains for its mobile-computing processors is impressive. We believe that AMD will achieve its energy efficiency goal, partially through process improvement but mostly by combining the savings from reducing idle power, the performance boost of heterogeneous system architecture, and through more intelligent power management. With this undertaking, AMD demonstrates leadership in the computing industry, driving innovations for a more energy-efficient future."
49 Comments on AMD Details Plans to Deliver 25x APU Energy Efficiency Gains by 2020
[humour]
This also seems to reflect the classic mindset that dug a large hole for AMD in the first place. A single minded focus on what would become K8 and Bulldozer whilst almost totally ignoring the competition and expecting Intel to persevere with Netburst and Core respectively. Intel might be wedded to x86, but that doesn't mean it's their sole focus - they do have an ARM architectural licence, and the IP deal (rare for Intel) with Rockchiptends to point to some diversification in processor strategy.
@GhostRyder
You keep dreaming those dreams son. The naïveté is refreshing. Last time I checked, Intel was built across quite a few product lines - and even taking CPUs in isolation, they basically own the x86 pro markets. HSA is all nice and dandy but at some stage it has to progress to actual implementation rather than a PPS decks and "The Future IS..." ™. For that to happen, AMD need to start delivering. They won't have IBM on board, and Dell, Cisco, and HP are all firmly entrenched in the Intel camp.
Seriously, how do you think AMD plans to contend with CPUs like the C2750? It's like an i5, without an iGPU, twice as many cores, and the PCH put onto the CPU. It's everything you could ever want from a low power CPU with the exception of half-decent graphics, but Intel already knows how to play that game with the Iris Pro and if the consumer market ever demanded it, I'm sure Intel would deliver and it's important to remember that Intel's iGPUs aren't as crappy as they used to be (most people don't game, keep that in mind too.)
AMD should take all this PR funding and put it into R&D because pandering to the masses isn't going to make their hardware any better than it already is. I don't see Intel making claims like this nearly as often as AMD does when it comes to PR.
With all of this said, I still love my AMD graphics cards but I'm glad I decided to get an i7.
But then again I expect nothing less from you hence why I rarely care anymore what you have to say. Keep posting I have nothing to say to you.Maybe, no one is saying an i7 is not as good as anything amd has on the table. They have the best performance right now and it's not going to change for awhile.
Right now iris pros main advantage is that ram built into the chip. Depends on how far they take it, but I could dig it either way if offered at a decent price.
Now then I'm done with you, and I'll leave on a nice Mark Twain quote which i should heed. I'm sure your response will be equally hilarious but I would rather not drag this thread any further off subject than this.
Looks like somebody 'roofied' you.
See what I did there?
You flip-flop faster than a politician caught red handed with a rent boyDidn't you say that last time out? Oh, yes! You did:rolleyes: o_OSo what? Sales mean f___ all if it doesn't translate into revenue.
The fact is Intel has benefited greatly from the innovations AMD has brought to the x86 & general computing realm whilst the single biggest gift they've received from Intel in the last decade has been the bribes to OEM's circa 2006, in other words a stab in the back ! Also Nvidia is embracing HSA with CUDA 6 (software only atm) so what I see from your post is ignorance for one & secondly you (perhaps) think that Intel is pro-consumer when in fact they're virtually the exact opposite & their actions, like unfairly blocking overclocking on non Z boards just recently, over the last many years certainly proves this point !
and this is not confirmed, this feature will allow the CPU and GPU to share system memory, which should boost performance of heterogeneous applications.
Which heterogeneous applications would they be ? Would these be future applications or applications actually available?And? What has that got to do with AMD's strategic planning?Yep. That's Intel.Not strictly HSA, it's unified memory pooling and isn't Nvidia part of the OpenPOWER consortium rather than the HSA Foundation?If you think OpenPOWER, Intel's UMA, and HSA are all interchangeable on a software level I think you're going to have to show your working before you start bandying around terms like ignorance.Maybe you should stop ascribing conclusions based on comments that haven't been made.
I'm a realist, and I see what the vendors do, how they achieve it, and the outcomes. Noting the facts doesn't imply anything other than noting the facts.Looks like you're just looking for an excuse to vent because this has absolutely no correlation to anything I've commented on.
Looking for an argument that Intel isn't an abuser of its position? You won't find one here. Intel's modus operandi is fairly well known. Intel's failings as a moral company don't excuse AMD's years of dithering, changing of focus depending upon what others are doing, saddling themselves with a massive debt burden by paying double what ATI was worth, selling off mobile IP for peanuts, dismissing the mobile market in toto, and a host of missteps.
You want to talk about ignorance? Blame Intel's bribery of OEM's (particularly Dell) to keep AMD out of the market? Know why the settlement wasn't bigger? AMD - thanks to Jerry "Real men have fabs" Sanders were too proud to second source foundry capacity. Bribes from 2006? Sure there were....AMD also couldn't supply the vendors it already had. Think that was a blip? Analysts were warning of AMD processor shortages yearsbefore this ever became acute. AMD complaining that Dell didn't want their processors was offset to a degree by OEM's complaining that AMD chips weren't available in quantity (so, 2002, 2006, and this from 2004 - see the trend), so AMD waited until vendors were publicly complaining* (and Jerry had been put out to pasture) before AMD struck a deal with Chartered Semi....and even then used less than half their outsourcing allocation allowed under the licence agreement with Intel.
Sometimes the truth isn't as cut-and-dried as good versus evil.
*Poor AMD planning causes CPU shortages: ....But European motherboard firms, talking to the INQ on conditions of anonymity, were rather more blunt about the problem. One described the shortages as due to "bad planning".
With all of that said, it pisses me off when people like you think that writing concurrency code that scales is easy when it's not.
For it to scale, most of it needs to be parallel, not just a tiny bit of it and I can't even begin to describe to you how complex that can get.
So if 50% of your workload is parallel, you'll benefit from two cores basically. HALF of your work needs to be parallel just for a 2.0 speedup... and you want code to run on how many cores again?en.wikipedia.org/wiki/Amdahl's_law
Also people (including but not limited to developers) do need a push to get things done more efficiently, for instance how many browsers were using GPU acceleration before Google (chrome) pushed them into obsolescence ? How many browsers still don't use SSE 4x or other advanced instruction sets, this isn't just you I'm talking about but it also is not a blanket statement targeting every software/game developer out there, since I clearly put emphasis on most !Irrelevant since I didn't mention the type of workload & thus you shouldn't try to sell Amdahl's law as an argument in such case.
I would put winrar/winzip and other compression utilities in the category of workloads that are more easily paralleled than others because of the nature of what they're doing. Once again, this comes down to the workload argument. Archival applications and games are two very different kinds of workloads, it's a lot easier to make something like LZMA2 to run in parallel than something like a game which is incredibly more stateful than something like an algorithm for compression or decompression. This isn't a matter of tools, you could have all the tools in the world but that won't change the nature of some applications and how they need to be implemented. OpenCL doesn't solve all programming issues and it doesn't mysteriously make things that couldn't be run in parallel to suddenly able to be. These tools you talk about enable already parallel applications to scale a lot better and across more compute cores than they did before, it doesn't solve the problem of having to make your workload thread-safe without being detrimental to performance in the first place.
You complain about me nitpicking, but you're pointing out things that require that level of analysis and detail because problems like these aren't as easy to solve as you make them out to be.
One question, have you ever tried to write some OpenCL code and running it on a GPU? Try doing something useful in it if you haven't and you'll understand real quickly why only applications that are mostly parallel code in the first place use OpenCL. I get the impression that you haven't so you shouldn't talk about something if you've never done it. I am, because I have... trust me, it's not intuitive, it's hard to use, and it's only helpful in very selective situations. I would never use it unless I was working with purely numerical data that was tens of gigabytes large or bigger and only if the algorithm I'm implementing is almost completely stateless (or functional if you will). Games (other than the rendering part, which GPUs are already doing,) hardly fit any of those criteria. It's not that developers aren't using OpenCL, it's that they can't or it doesn't make sense to in most real world applications in the consumer market.
I do enjoy listening to you try to say what developers are and are not doing right when you're not in their shoes. Even as a developer I wouldn't presume to think I knew more about another developer's project than they do without even seeing the code itself and having worked with it. So I find it both amusing and disturbing that you feel that you can voice you opinion in such an authoritative way when not even I would make those kinds of claims given my own experience in the subject as I'm a developer professionally and I'm even working on a library that uses multiple threads.
Tell me more about why you're right.
**Edit** There is a lot of butthurt in this thread, over PR, nothing more. I am glad they have this goal.
How about we debate the power useage and how perhaps a embedded capacitor(s) that can provide the peak power required when firing up more cores or execute units could provide us with a 200Mhz CPU that clocks to 4Ghz instantly?
Decoupling capacitor built in anyone?
Honestly, I don't get it...That's a little childish I guess. Latest top i7 CPUs from Intel are also APUs.
I apologize, I didn't mean to completely ignore the topic of the article in the first place... But as a user of both AMD and Intel (actually an AMD user from Socket A up until my first Intel build at the release of Core 2) I certainly hope that they achieve these goals simply for the reason that any innovation from any team is always good for us. I can remember when AMD first mentioned Fusion, and adding the GPU to the CPU die... then low and behold, here comes Intel taking that idea and running with it and beating AMD to market initially with a crappy solution (though AMD still has the better iGPU today) and look where we are now with integrated graphics. I for one do appreciate the strides made with iGPU's for systems such as my Surface Pro. (I would really like to see an AMD APU version of one!) Or when AMD puts the memory controller on die with the Athlon 64, then here comes Intel with Nehalem doing the same thing. It really is too bad that they took such a step backwards with Bulldozer when they seemed to have good momentum going and hanging with Intel back in those days. I really hope that the day may come again when they are close and drive each other to really innovate. I've been an Intel user now since Core 2 and would love to feel like I have another option when building my desktops.... possibly by the time ill be looking to replace my upcoming Haswell-E system?
There are really maybe three situations that I feel are important for multi-threading:
A: When you know that you want, have everything you need to get it but is something that you don't need until later. (a form of speculative execution)
B: When you have a task that needs to run multiple times on multiple items and doesn't produce side effects. (I.E. Graphics rendering or protein folding)
C: A task that occurs regularly (every x seconds, or x milliseconds) and requires very little coordination.
As soon as you have side effects or have tasks that rely on the output of several other tasks, the ability to make something usefully multi-threaded goes out the window and people might not realize it, but games are one of the most stateful kinds of applications you can have and just "make it multi-threaded" as it doesn't solve problems. In fact if the workload wasn't properly made to run in parallel, making an application multi-threaded can degrade performance when overhead is most costly than the speedup that's gained from it or even make the code that's executing more confusing because of any locking or thread-coordination you may have to do.
I currently develop with Clojure, which is a functional language on top of the JVM among other platforms which I don't typically use (except for ClojureScript which is interesting)....andTo make a long story short, application state is what makes applications demand single-threaded performance and not managing it well is what reinforces that.
I, for one, would like one of these 0 new free cards. With a 25X improvement over the current 0 free cards, it should not be any issue for me to receive [RESULT UNDEFINED]Your post is invalid. Minimum butthurt level not met. Ignoring.