Thursday, June 25th 2020

AMD Exceeds Six-Year Goal to Deliver Unprecedented 25 Times Improvement in Mobile Processor Energy Efficiency

AMD today announced it has exceeded its moonshot 25x20 goal set in 2014 to improve the energy efficiency of its mobile processors 25 times by 2020. The new AMD Ryzen 7 4800H mobile processor improves on the energy efficiency of the 2014 baseline measurement by 31.7 times1, and offers leadership performance2 and extraordinary efficiency for laptop PCs. Greater energy efficiency leads to significant user benefits including improved battery life, better performance, lower energy costs and reduced environmental impact from computing.

"We have always focused on energy efficiency in our processors, but in 2014 we decided to put even greater emphasis on this capability," said Mark Papermaster, chief technology officer and executive vice president, Technology and Engineering at AMD. "Our engineering team rallied around the challenge and charted a path to reach our stretch goal of 25 times greater energy efficiency by 2020. We were able to far surpass our objective, achieving 31.7 times improvement leading to gaming and ultrathin laptops with unmatched performance, graphics and long battery life. I could not be prouder of our engineering and business teams."
Energy efficiency for processors is determined by the amount of work performed per unit of energy consumed. To achieve the 25x20 goal, AMD focused improvements on developing a highly integrated and efficient system-on-chip (SoC) architecture; improved, real-time power management features; and silicon-level power optimizations. AMD reduced average compute time for a given task by 80% from 2014 to 2020, while also achieving an 84% reduction in energy use.3 That means an enterprise that upgrades 50,000 AMD laptops from 2014 models to 2020 models would achieve five times more computing performance and reduce associated laptop energy consumption by 84%, which over a three-year service life amounts to saving approximately 1.4 million kilowatt hours of electricity and 971,000 kg of carbon emissions, equal to 16,000 trees grown for 10 years.4

Achieving its 25x20 energy efficiency goal not only delivers a stronger user experience but also further solidifies AMD's leadership in sustainability. AMD was the first semiconductor company to have its climate protection goals, including 25x20, approved as a "science-based target" by the Science Based Targets initiative—deemed aggressive enough to help mitigate the impacts of computing on climate change. 5

"Stewardship of our planet can go hand-in-hand with developing powerful technology and helping our customers achieve their aims," said Susan Moore, corporate vice president for corporate responsibility and international government affairs at AMD. "The power efficiency improvements we have made within our laptop processors make a difference in the world around us. Reporting publicly each year on progress and ultimately achieving, and surpassing, the 25x20 energy efficiency goal reflects AMD's commitment to sustainable technology."

Industry Perspective
The performance improvements and reduced power consumption of the Ryzen 7 4800H processor outpaced the historical efficiency trend predicted by Koomey's Law—a Moore's Law analog describing energy efficiency improvement trends—by 2x from 2014 and 2020.6

"Six years ago, AMD challenged itself to dramatically improve the real-world energy efficiency of its mobile processors," said Dr. Jonathan Koomey, an industry expert on energy efficient computing. "I have reviewed the data and can report that AMD exceeded the 25x20 goal it set in 2014 through improved design, superior optimization and a laser-like focus on energy efficiency. With a chip 31.7 times more energy efficient than its 2014 predecessor, AMD has far outpaced in real-world efficiency gains what would be expected from a traditional Moore's Law pace as embodied in Koomey's Law."

"AMD undertook an audacious and public engineering goal to improve its processor efficiency by 25 times, vastly outpacing historical averages, and in doing so, achieved an industry-leading position in mobile processors," said Kevin Krewell, principal analyst at TIRIAS Research. "The 25x20 goal required major changes to architecture, design and software, and did not rely solely on silicon process technology advancements. Achieving and surpassing this challenging goal is a testament to the hard work the AMD team put into its products and it catapulted AMD to a leadership position in mobile processors."

"AMD's goal of improving the power efficiency of its laptop processors 25 times by this year may have seemed like a bit of an abstract, arbitrary target when they first announced it six years ago," said Bob O'Donnell, president of TECHnalysis Research. "But as they surpass that impressive goal, it's becoming significantly more meaningful. Lower energy consumption has never been more important for the planet and the company's ability to meet its target while also achieving strong processor performance is a great reflection of what a market-leading, engineering-focused company they've become."

1 Testing by AMD Performance Labs as of 4/15/2020. Processors tested: AMD FX-7600P, AMD FX-8800P, AMD FX-9830P, AMD Ryzen 7 2700U, AMD Ryzen 7 2800H, AMD Ryzen 7 3750H, and AMD Ryzen 7 4800H. 25x20 program tracked against Energy Star Rev 6.1 8/12/2014 and 3DMark 2011 P-Score and Cinebench R15 nT. Results may vary with drivers and BIOSes. RVM-108
2 Testing by AMD Performance Labs as of 12/09/2019 utilizing an AMD Ryzen 4800H reference system and an ASUS Zephyrus M GU502GV system with Intel Core i7-9750H processor in Cinebench R20 1T and nT. Results may vary. RM3H-1
3 The normalized performance increase, based on a 50:50 weighted metric for Cinebench R15 and 3DMark11, is 5x higher from AMD's 2014 notebook processor to the 2020 design. This equates to one-fifth the average compute time for a given task. Annual processor electricity use (kwh), based on ENERGY STAR typical use energy consumption (TEC), in 2020 is 84% less than the 2014 amount. RM3H-42
4 Emissions reduction estimates for an enterprise upgrading 50,000 AMD laptops from 2014 to 2020 models are based on entering estimated electricity savings into the U.S. EPA Greenhouse Gas Calculator on March 23, 2020 (https://www.epa.gov/energy/greenhouse-gas-equivalencies-calculator). AMD estimated annual electricity savings based on ENERGY STAR typical use energy consumption between the 2014 notebook processor and power supply and the 2020 processor and power supply over a 3-year service life and multiplied by 50,000 units.
5 https://sciencebasedtargets.org/companies-taking-action/
6 AMD achieved a 31.7x increase in typical use energy efficiency from 2014-2020, or ~2x compared to what would be the historical rate of increase (doubling every 1.57 years) during the same timeframe of 14.1x. RM3H-43
Add your own comment

32 Comments on AMD Exceeds Six-Year Goal to Deliver Unprecedented 25 Times Improvement in Mobile Processor Energy Efficiency

#1
theoneandonlymrk
What was the start point again ,Kabini?/Renoir
Or we talking BD cores?
Posted on Reply
#2
HugsNotDrugs
I'm still happy that we have some competition in the x86 space, both desktop and mobile.
Posted on Reply
#3
Carillon
i dont buy it, 84% less energy used means either 2014 apus had a 250w tdp or 2020 apus have a 10w tdp
imho they are just measuring energy usage to complete a task, if thats the case thats all the efficiency gain is, 5x.
Posted on Reply
#4
$ReaPeR$
Guys, it's marketing with a pinch of truth. Chill.
Posted on Reply
#5
theoneandonlymrk
I'm not tense just intrigued.

And surely it's performance per watt , I don't doubt them.
Posted on Reply
#6
Vayra86
$ReaPeR$
Guys, it's marketing with a pinch of truth. Chill.
At least your cup is always full :)

Frog must be thirsty
Posted on Reply
#7
$ReaPeR$
theoneandonlymrk
I'm not tense just intrigued.

And surely it's performance per watt , I don't doubt them.
Most probably, but 84% sounds better. I hate marketing.
Posted on Reply
#10
xkm1948
TBH, everything looks way better than those construction crew CPUs from AMD
Posted on Reply
#11
$ReaPeR$
xkm1948
TBH, everything looks way better than those construction crew CPUs from AMD
Exactly.
Posted on Reply
#12
rrrrex


Stupid math. "Completing a computation task in 1/5th the time, and consuming 1/6th the typical energy". It's about the same, we shouldn't multiply it. CPU consumed less power cause it took less time for the task.
Posted on Reply
#13
Valantar
Carillon
i dont buy it, 84% less energy used means either 2014 apus had a 250w tdp or 2020 apus have a 10w tdp
imho they are just measuring energy usage to complete a task, if thats the case thats all the efficiency gain is, 5x.
Energy used is not the same (or even remotely close to) power draw or TDP. Energy used is the actual electrical energy (measured in Joules) to complete a given workload. This way you get a metric that is independent of both absolute power draw and absolute performance as it accounts for both. You can increase efficiency by drawing more power for short amounts of time if that lets you complete the workload faster and thus return to idle.

If CPU A needs 20W continuously to perform the workload in 200 seconds, it consumes X Joules.
If CPU B needs 30W continuously to perform the workload in 100 seconds, it consumes 3/4X Joules.
Of course if CPU C needs just 20W continuously to perform the workload in 100 seconds, that is again just 1/2X Joules.
Etc., etc.
rrrrex


Stupid math. "Completing a computation task in 1/5th the time, and consuming 1/6th the typical energy". It's about the same, we shouldn't multiply it. CPU consumed less power cause it took less time for the task.
...isn't that efficiency? Isn't the only reasonable measure of power efficiency "power consumed to perform a given task"? As in "miles you are able to drive per gallon of gasoline" or "amount of light output per watt of power consumed"?
Posted on Reply
#14
rrrrex
@Valantar
He took point per watt and multiplied by spent time. But points per watt are already have time in it (less time = more points, or twice more time = twice less points). So his "energy" depends from t^2, not t.
Posted on Reply
#15
Valantar
rrrrex
@Valantar
He took point per watt and multiplied by spent time. But points per watt are already have time in it (less time = more points, or twice more time = twice less points). So his "energy" depends from t^2, not t.
Actually this is wrong. The number is based on two factors: (a) performance at a given power draw in a common workload, and (b) idle power based on Energy Star's ETEC standard. This is the two numbers they are multiplying together, with 5 (actually 5.02) being the increase in compute performance (a), and 6.33 being based on an energy consumption of 0.16 or 16% of the 1 unit of power consumed by the original reference system (b) (1/0.16=6.25 - I'm guessing 0.16 is rounded up, as 1/6.33=0.157978). While I can agree it's a bit of an optimistic formula, it isn't directly misleading, and it's the one they've been basing this work on from the very start of the program. In other words, the "25x" goal they first set for themselves was based exactly on calculations like these. Is 25x or 31x higher than the sum perceptible increase in performance and efficiency? That is of course impossible to quantify, and you are welcome to propose your own formula, but then you also have to adjust the goal they're being compared to alongside it.
Posted on Reply
#16
Patriot
Do that many people have problems with basic physics? This is always how efficiency has been measured.
Posted on Reply
#17
Mathragh
They set a goal, and reached it handily by their own set parameters. What's there to be negative about! Honestly in my opinion it's quite refreshing compared to most other (tech) news you hear these days.
Posted on Reply
#18
R0H1T
$ReaPeR$
Exactly.
Nope, the biggest handicap AMD had was the fabs & that's why Intel led the x86 market the last decade by a wide margin. Intel already moved to 14nm in 2014 & AMD was stuck at 28nm! A hint of competition in the fab space & Intel's fumbling like crazy :laugh:
Posted on Reply
#19
dont whant to set it"'
Before this turns into a math dispute in a "obscure" internet forum thread , I share with you this gem of a video which I'll rewatch now .
Posted on Reply
#20
$ReaPeR$
R0H1T
Nope, the biggest handicap AMD had was the fabs & that's why Intel led the x86 market the last decade by a wide margin. Intel already moved to 14nm in 2014 & AMD was stuck at 28nm! A hint of competition in the fab space & Intel's fumbling like crazy :laugh:
Although fabs do play a role, I do believe it's minor compared to the uarch. For example the FX 8xxx/9xxx never had a problem reaching 5ghz but they barely competed with i5s and i7s. So..
Posted on Reply
#21
dont whant to set it"'
$ReaPeR$
Although fabs do play a role, I do believe it's minor compared to the uarch. For example the FX 8xxx/9xxx never had a problem reaching 5ghz but they barely competed with i5s and i7s. So..
Much as they wanted to do the Zen arch before the the k11.5 one , the feasibility of a commercial product was not there at those times into actually manufacturing.
Posted on Reply
#22
wahdangun
Carillon
i dont buy it, 84% less energy used means either 2014 apus had a 250w tdp or 2020 apus have a 10w tdp
imho they are just measuring energy usage to complete a task, if thats the case thats all the efficiency gain is, 5x.
wtf, no, its proportional with performance. except if the performance stay same from 2014 to 2020 then you will be right
Posted on Reply
#23
R-T-B
Vayra86
Frog must be thirsty
I believe the figure in performance per watt on a select benchmark. Outside of that, gg guys.

I'm a realist, and never thirsty.
xkm1948
TBH, everything looks way better than those construction crew CPUs from AMD
I swear bob the builder was on the codename squad for a bit there...
Posted on Reply
#24
watzupken
I feel this is factual correct, but considering the bar was set very low with a Bulldozer architecture and 28nm, to the current Zen 2 and 7nm, just the die shrink itself will help improve energy efficiency. The improvements made from BD to Zen 2 is incredible, again due to the very low bar set by BD architecture. Still overall I think it is a great achievement since they were far behind Intel just about 4 years back. Now, they have overtaken them in most cases.
Posted on Reply
#25
rrrrex



So they took 5x perfomance boost and 1/6 of power power consumtion. But look at E, there are off, sleep and idle, no load. Is it true that 4800H consumes 6 times less power in idle than 7600P?
Posted on Reply
Add your own comment