ARM, Synopsys Bringing Mali GPU to 20nm Processing

Status
Not open for further replies.

phate

Distinguished
Oct 23, 2009
149
0
18,680
"this implementation will enable designers to optimally implement ARM Mali-T600 family GPUs using Synopsys tools in sub 20-nm leading-edge process technologies."

"ARM said this combined functionality brings additional hardware complexity which is further compounded by the new double-patterning requirements introduced by 20-nm and below technologies."

The "leading-edge" process technology is a misnomer, as evidenced by the second statement. Though nominally on par with Intel, this is actually still a generation or so behind. IIRC, Intel doesn't need to use double-patterning for their interconnects, because they are already implement EUV lithography.

And the problem gets worse in the next-generation as ARM manufacturers will still be using 20nm double-patterned interconnects when their transistors hit 14nm. So power draw will drop, but density will still remain at 20nm even though they'll be marketed at 14nm due to transistor size.

Basically this story is a lot of fluff, trying to spin how their weaker, older process technology is somehow something to be proud of.
 

InvalidError

Titan
Moderator
[citation][nom]phate[/nom]Basically this story is a lot of fluff, trying to spin how their weaker, older process technology is somehow something to be proud of.[/citation]
ARM and Synopsys do not make the physical chips. ARM makes the HDL design and Synopsys makes synthesis software tuned to individual foundries' processes.

If TSMC, Global Foundries and other fabs actually had 14nm capabilities, Synopsys would license a version of their suite tuned for that too.
 

saturnus

Distinguished
Aug 30, 2010
212
0
18,680
[citation][nom]phate[/nom]Basically this story is a lot of fluff, trying to spin how their weaker, older process technology is somehow something to be proud of.[/citation]

ARM design chips. They do not manufacture anything. If there was a demand for much smaller process node designs than 20nm and beyond, ARM would provide those.

Intel's smallest process node for their mobility portfolio is currently 28nm although it plans to start 22nm production at the very end of this year (late Q3 or Q4)

Samsung however is right on Intel's tail, and plans 20nm production (which will probably use the design mentioned in this article) to start Q4 this year or Q1 next year.
 

antilycus

Distinguished
Jun 1, 2006
933
0
18,990
Im confused on why I would care about this more than ARM + Tegra? NVIDIA has been doing it for years. I guess POWER VR has too but I'd rather take NVIDIA than Power VR simply do the their understanding of the market
 

twist3d1080

Distinguished
Mar 20, 2010
40
0
18,530
[citation][nom]phate[/nom]"The "leading-edge" process technology is a misnomer, as evidenced by the second statement. Though nominally on par with Intel, this is actually still a generation or so behind. IIRC, Intel doesn't need to use double-patterning for their interconnects, because they are already implement EUV lithography.[/citation]

Nobody has implemented EUV in a production enviroment.

[citation][nom]antilycus[/nom]Im confused on why I would care about this more than ARM + Tegra? NVIDIA has been doing it for years. I guess POWER VR has too but I'd rather take NVIDIA than Power VR simply do the their understanding of the market.[/citation]

Tegra's GPU has never been an industry leader, ever. Power and Mali GPU's have out performed ULP Geforce GPU at every release. Thats why you would care.
 

ojas

Distinguished
Feb 25, 2011
2,924
0
20,810
[citation][nom]saturnus[/nom]ARM design chips. They do not manufacture anything. If there was a demand for much smaller process node designs than 20nm and beyond, ARM would provide those.Intel's smallest process node for their mobility portfolio is currently 28nm although it plans to start 22nm production at the very end of this year (late Q3 or Q4)Samsung however is right on Intel's tail, and plans 20nm production (which will probably use the design mentioned in this article) to start Q4 this year or Q1 next year.[/citation]
No, it's 22nm for Core and 32nm for Atom, Qualcomm uses 28nm for Krait.

Samsung's 20nm...all i remember about it is that it was due for a 2014-15 launch, could be wrong.
 
Status
Not open for further replies.