News - AMD plans to deliver 25x APU efficiency gains in next six years

[HEXUS]

Thanks to design optimisations, intelligent power management and HSA.

Read more.

[DanceswithUnix]

What now takes 250W you will be able to do in 10W? That doesn't make sense.

What you can now do in 2.5W you will be able to do in 100mW? Maybe, but with the backlight on who much cares.

[watercooled]

Well it's a fairly ambiguous claim - if they're referring to platform power, the industry has already made some significant improvements there over the past few years but there's still some room to go, especially on the desktop where power efficiency seems to be an afterthought at best on a lot of motherboards.

They could also be talking about specific areas like media, where hardware/GPU/DSP processing can be massively more efficient than CPU processing - the 2.5W to 100mW is already possible with stuff like video codecs, moving from software to dedicated logic. This sort of thing would tie into the HSA direction - maybe we'll see more of the sort of stuff we're used to seeing in mobile ARM SoCs start to appear in desktop/laptop processors?

[sam3]

"AMD has managed to improve the typical energy efficiency of its mobile processors by more than 10x"

It looks like their target will be 2.5x its current efficiency (as they are already on 10x - to make 25x total)

[xslavic]

..i guess it means that the ratio of the (CPU x performance + the Gpu x performance)/watt will be bigger by 25 times,so cpu maybe +10% + gpu y% total divided by the power consumed will have that effect and maybe twice than ARMS and Intels at that level.Thats what i understood form the sentence.

[wasabi]

Also got to remember Moore's law means you're in effect doubling the transistor count every two years anyway - or at least the competition will be. So you in effect are going to get a X8 performance boost IF power draw stays constant. So really they're talking about dropping power to 1/3 of current levels - then multiply that 1/3 by 8 to get 24X, if you see what I mean. Yes, it is back of envvelope maths.

[valhar2000]

What now takes 250W you will be able to do in 10W? That doesn't make sense.

Possibly. That would mean that, in six years, a low power APU will have the performance of 2 or 3 of today's high-end AMD APUs. An very ambitious goal, certainly, but not science-fiction.

[benegerton1985]

Well I certainly believe they can get close, the difference they made in efficiency between the Trinity and Richland APU's I own is fantastic, and they both use the same architecture! If I look at the p-states for both chips, a Trinity chip requires 0.9v to be at it's lowest p-state at 1.4Ghz and Richland requires 0.325v to be at 1.8Ghz. No idea about Kaveri however but I presume the difference between Kaveri and it's successor based on the same architecture would be similar again.

[Zak33]

2020 update :

AMD have exceeded that target from 6 years ago!

https://hexus.net/tech/news/industry/143740-amd-revisits-25x20-efficiency-goal-says-excelled/

[Corky34]

That's some serious memory you have there.

I was about to make a fool of myself and say the year in the article was wrong.

[Zak33]

I think its quite an achievment, and remember at the time thinking "go on then...crack on" as my own AMD experiences were of low power use (on cpu...)

[Xlucine]

What now takes 250W you will be able to do in 10W? That doesn't make sense.

What you can now do in 2.5W you will be able to do in 100mW? Maybe, but with the backlight on who much cares.

Didn't make sense in 2014, but they've managed it

The 9590 (OK, not the best example of power efficiency) is about half the speed of a 4800U in cinebench R15, and that's almost a 250 to 10 W comparison