http://wccftech.com/amd-polaris-revisions-performance-per-watt/
>AMD has reportedly started rolling out new revisions of its Polaris 10 & Polaris 11 GPUs with a 50%+ improvement in performance per watt.
>50%+ improvement in performance per watt.
>bring down the typical board power from 150W as is found in the RX 480 to less than 95W
>95W
THANK YOU BASED AMD
>>57067687
I'm sure early adopters are gonna be very happy.
>buying amd anything
Enjoy your unfinished housefires.
>>57067687
>50%
>from 150 to 95
Maths is hard
>>57067687
so not an software upgrade but an hardware upgrade? That's a shame tho.
>>57067836
>50%+
>+
Says it'll be going into embedded first, then laptops. No word on desktops.
>>57067687
That's assuming the performance is the same, pleb
>>57067836
How does it feel being retarded?
The RX480 has a 150w power target, meaning everything on the board tries to maintain that figure as a nominal draw.
At stock settings this is divided into 110w for the ASIC itself, and 40w for the GDDR5.
The voltages for the ASIC's pstates are only as high as they are to ensure yields were as high as possible. They wanted to ship out as many good working dies as possible so they had wide margins.
8 1GB GDDR5 chips replaced by 4 2GB chips would cut the memory power draw in half.
Another 30w saved by lower voltages and tweaks to the IMC in how it handles the memory's power states could be easy to accomplish.
A lot of current cards can underclock just the ASIC and save 20w without losing performance. Its just a matter of learning the nuances of Wattman's behavior in addition to setting pstates individually.
>>57068825
Wouldn't it be better to actually use a single 8GB chip?
Or does it make it actually slower because only 1 piece of hardware is being addressed instead of asynchronously getting them from 2 or more.
Or would it actually be a case of production cost for not using only 1 chip, I don't know how expensive GDDR5 chips per GB actually are.
>>57068841
There are no chips in production with a density that high.
>>57068825
You can't tie 2 32bit memory controllers into a single GDDR5 memory chip m8.
>>57069358
>I don't know fucking anything: The Post
Hey, child, you tech illiterate shit eating subshuman faggot. Guess what? You you don't know anything about memory controllers and memory PHY.
Go look at the PS4 revisions. The original had 8 1GB GDDR5 chips. The revised models have 4 2GB chips. Guess how they can do that?
Protip: Its because the PHY can address the layers in the chips in the exact same way. There is no access bottleneck.
Tech illiterate retards like you should just kill themselves.
Great, I just bought RX480 Nitro two weeks ago.
>>57069395
you should buy some rope then.
>>57067879
A 50%+ improvement suggests at least a 50% reduction you idiot.
>>57067836
The power saving is for the GPU alone. The GPU is not the only power consumption on the board. If the GPU was drawing 110W of the 150W of the RX480, then a 50% reduction on the GPU would result in 95W power draw for the card.
>>57069643
I'm glad not everyone on this board is an idiot like op.
>>57069672
I am often an idiot, I just happen to not be in this thread.
>>57067836
>math is hard
50% higher performance per watt, not 50% lower power consumption. If it consumed half as much power for the same performance, that would be 100% higher performance per watt.
>>57069411
We won't see the RX485 until next year senpai
AMD are a bit late to the effiency game.
Let's see how the GTX 2k series of Nvidia will perform.
I guess it will be released at the same time around these AMD refreshes.
>>57067836
dumb niggerposter
I know 14nm is a new process not yet matured, but still
>50% PpW increase in one shot
Logging this under "I'll believe it when I see it"
>>57070185
The Ellesmere die is not drawing an usual amount of power at all. Vs the 32CU 28nm Tahiti, die the 36CU Ellesmere draws half the power at higher clocks.
Higher clocks, more CUs, more transistors per CU, higher perf per clock, and radically lower power per clock.
>>57070157
>>57067687
It's pretty nice.
Can't wait to see how their GPUs perform
Does this mean i have to wait a few months for stores to stock them? And will this apply to custom designs like sapphire and msi?