AMD Vows 25-Fold Increase in Chip Energy-Efficiency by 2020

AMD Vows 25-Fold Increase in Chip Energy-Efficiency

AMD Accelerates Energy Efficiency of APUs, Details Plans to Deliver 25x Efficiency Gains by 2020

Advanced Micro Devices has made few big waves in chip technology since the past decade, when it introduced some important advances ahead of arch-rival Intel. But it is setting an ambitious target to push the state of the art again.

Mark Papermaster, AMD’s chief technology officer, on Thursday is promising a 25-fold improvement in the energy-efficiency of its products by 2020, using a series of design techniques that go beyond those that have historically come from shrinking transistors on chips.

“We believe it’s going to have a big impact on the industry,” Papermaster said
in an interview ahead of a speech he is giving on the topic in China.

The gains AMD sees aren’t just in reducing power consumption–an area where it is seen to have lagged Intel–but in boosting computing performance at the same time. That combination could help drive AMD’s chips, which are now mostly in laptop and desktop PCs, into more tablets and smartphones.

And laptops could see some pretty dramatic improvement. Sam Naffziger, an AMD researcher who holds the title of fellow, sees about a five-fold increase in computing capability for the typical laptop while drawing about one watt of power, down from five watts or so today.

That means typical battery life would go far beyond the current typical maximum of eight to ten hours. “We’d be talking several days of battery life,” Naffziger said.

AMD is just one of many companies pursuing such goals. Besides making more attractive products, technology vendors and their customers worry about the global impact of spiraling energy consumption of electronic devices in homes and data centers.

AMD cites estimates that three billion PCs in use worldwide use more than 1% of all energy consumed annually, while 20 million server systems use an additional 1.5%.

Other chip companies, like Intel, Samsung and Qualcomm, have more to spend on research than AMD. It no longer operates its factories or develops new processes for making computer chips; AMD relies on others for benefits from the transistor miniaturization race known as Moore’s Law, after Intel’s co-founder.

But Naffziger notes that AMD does have some formidable assets, too. It is a major supplier of chips known as graphics processing units, or GPUs, a status that helped win the job of supplying chips for the latest generation of gaming consoles from Sony and Microsoft.

The GPU circuitry can handle some kinds of computing chores–like image and speech recognition–at particularly high speeds. So AMD has been aggressive about combining GPUs on the same piece of silicon with conventional processors, as well as even more special-purpose circuitry for jobs like video encoding. It has also teamed up with other companies in an effort known as HSA, for Heterogeneous System Architecture, that helps companies develop and program such multi-function chips.

Naffziger cites other design techniques to reduce power consumption. One involves embedding a component called a micro-controller that manages power consumption, temperature and activity throughout a chip.
 
What connection, if any, to the rumored power to performance improvements in that upcoming part from AMD? I'm referring to the talk of a chip with the power of a R9 290 that sips power like a R9 280? That's my question and maybe we'll hear about that if/when the chip in question hits retail.
 
What connection, if any, to the rumored power to performance improvements in that upcoming part from AMD? I'm referring to the talk of a chip with the power of a R9 290 that sips power like a R9 280? That's my question and maybe we'll hear about that if/when the chip in question hits retail.

I believe the original rumor was something performing around a R9 280x levels with a Pitcairn TDP, ~180w.

And there doesn't seem to be much of a connection between the article and the rumor.
Article is mainly focused on APU/SOC and CPUs.
 
Last edited by a moderator:
I believe the original rumor was something performing around a R9 280x levels with a Pitcairn TDP, ~180w.

And there doesn't seem to be much of a connection between the article and the rumor.
Article is mainly focused on APU/SOC and CPUs.

Bold added by me. Thanks! I keep getting them mixed up; I'm used to thinking in terms of the HD 7970 and the HD 7870. I forget to properly convert them to the newer nomenclature.

But I was wondering if any of the technology or techniques used, were used by both. I guess it could just be a matter of this is the age of being compelled to spend resources on getting the most out of the existing process.

Different answers for different products, it's just the motivation that is the same, I suppose. But I'm hoping, because it would be cool, that AMD stumbled on some "magic smoke" that they can use across the board.
 
That sounds promising.

Indeed . I only Watched Half . As Intels Knights landing is basicly what HP is talking about . Intel photonics . Micron 3D stacked Dram . Where AMD talks APU intel talks AVX. I believe Intel has stated that they also have same memory pool for the Cpu /GPU just as AMD is doing. Knights Landing Went from 1TflopDP to 3TflopD at the same time power consumption dropped from 300 watts to 200 wats That about 4x improvement power/efficiency. Intels Knights Landing do out in 2015. I wonder who will be first NV . Amd Intel . Intel has announced a real part . So have the others I believe. As nice as the Silvermont cores make Knights landing . It doesn't really get interesting until Intel uses it Broxton chips in PHi. Broxton single core is suppose to be better IPC than amds best Cpu .Intel HAS stated that the Some xeon cores will come with the Omni Fabric and the Micron 3D dram along with intels MCDRAM interface . Intel Also said they would provide PCI-e 3 swithes for AMD and NV to use as interconnects Intel with its Chips will use photonics in the first generation to interconnect their hardware. Intel priority photonics. Intel also has the option of using QPI. Also. Rather than PCI-e 3 on dual sockets with phi in one socket and big core in the other . Of course NV can do the same But with an Arm Core
and NO qpi. Intels at least 3 years ahead in photonics. But 2015 has the right stuff to be a great hardware year from all players ;)
I would never buy a Knights Landing . But what comes after you bet I would . Phi at 10 nm Should have around 12Tflops DP. I don't completely understand intels plans for AVX . But its really starting to look like Intel is build BOTH a compute and Graphics monster. Did Intel DUMP larrabbee Completely. At 10nm Intel is likely going to AVX1024.
At 12Tflops and under 150 watts . I hope intel is secretly developing graphics for Phi and AVX. Cannon Lake on intels 10nm could easily be big little cores. Say 2 big cores combined with say 50 Phi cores at around 95Watts Should give a desktop All the compute power A home user can ever use and deliver at least 6Tflops DP. We will know more when we see the Direction that the generation 8 graphics is heading. All the players will have Great Stuff. Its getting exciting.

More in line with The Machine Link. But hopefully still related to The Subject . As AMD is talking Hugh gains so I looking at ways to achieve these gains.http://www.intel.com/content/www/us/en/research/intel-labs-silicon-photonics-research.html After reading the Page . I put Omni Fabric into The Intel Search . . If this info Interest you. Than do the same Intel Intel Makes bold statements about Omni Fabric But don't say alot about it. I am thinking photonics
 
Last edited by a moderator:
Back
Top