NVIDIA: 11nm GPUs with 5000 Stream Processors Due in 2015
Bill Dally
NVIDIA’s Chief Scientist Bill Dally, said at the Design Automation Conference that computing is entering a world where performance increases are derived from parallelism and efficiency is determined by locality.
“Chip designers will need tool and techniques to optimize power, interconnect and locality. We are really looking to EDA (Electric Design Automation) to give us power tools. I want high-level tools that allow you to gain insights into power architectures very early in design,” Dally said.
Dally predicts that GPUs will continue to evolve rapidly in the following years, and until 2015, the GPUs will feature 5000 stream processing engines and about 20TFLOPs of computing power, based on 11nm process.
According to Dally, advanced EDA tools can enable performance leaps greater than those enabled by Moore’s Law.
July 31st, 2009 at 6:54 pm
So, it is okay to say this:?
- In 2015 expect mainstream using at least 2500 stream processors
- In 2015 expect low end using at least 1250 stream processor
The best case is he is talking about low end
.
July 31st, 2009 at 8:03 pm
Nice pedo-smile there, Bill Dally!
July 31st, 2009 at 8:07 pm
On 55 nm ATI has 800 stream processors. With 11 nm technology transistors would be 25 times smaller so 800 * 25 = 20.000 stream processors should be mainstream then.
I know this is incorrect ;p
July 31st, 2009 at 9:12 pm
come 2015 they’ll still be using the G92 chip -.-
August 1st, 2009 at 1:26 am
haha nice andy
BUT will it play crysis (laughs inside then dies a little there also)
August 1st, 2009 at 2:15 pm
I think the real question is will anyone ever start to innovate in their games? When I shoot a nade at a wall it should blow up the wall. When I repeatedly shoot a tank at a building, well there should be no building anymore. I really don’t care if this happens in today’s graphics or in super awesome future graphics.
August 2nd, 2009 at 4:53 am
I think they could easily beat moore’s law if the market simply required it… I mean just think of what the military might have right now
August 2nd, 2009 at 4:07 pm
can i get it for $99.00 cdn
August 2nd, 2009 at 9:37 pm
I think they are rather gonna have dual-core and quad-core GPUs than all those stuffed on one core.
August 3rd, 2009 at 1:20 am
Nah I don’t think there are the same kind of problems as with a cpu, in fact they can probably just shove stream processors in until the die is full.. this is the reason why gpu’s are much less flexible
August 6th, 2009 at 7:49 am
Piotrek, you maybe right. However seeing as Nvidia’s highest count of SPU at the moment is 240, and by your logic, 240 * 25 = 6000.
ATI achieved a “800 SPU” spec because they called them scalar processors. ATI breaks down their vertex processors into smaller groups to get their scalar processors, and that 800 ends up being 160(800/5) in vertex processors(Actual SPU).
A quote to explain more where the 5 came from:
“I think it’s disingenuous of AMD to say ‘scalar’ processors; they are vector processors. Why does it matter? Because each group of 5 ‘scalar’ processors can only process one vector at a time, even if that vector doesn’t have as many as five dimensions. ” -http://techreport.com/discussions.x/14344
It is just like saying the Core I7 is an octo-core when technically it is only a quad-core, the hyper threading only makes it act like an octo-core.
So with all of that said 160 * 25 = 4000 as opposed to the 6000 from the Nvidia figure. In the end, 5000 is the obvious middleman, and is the safest route to a “guess” number for Daily to say than looking like a dickhead to the public.