AMD's Real-time Visual Experience Goal: "Full Presence" at 1000 TFLOPS

Avatar image for scrollinglayers
ScrollingLayers

632

Forum Posts

0

Wiki Points

0

Followers

Reviews: 0

User Lists: 0

#1 ScrollingLayers
Member since 2015 • 632 Posts

It was in this presentation

I'd be happy if consoles in 2020 / 2021 gave us, oh maybe 20 TFLOPS.

At least it's good to know the semi-conductor industry wants to keep advancing forward like this, despite the challenges to making transistors smaller and frequencies higher.

If they succeed, that should be good enough for at least 3 more generations of PlayStation and Xbox consoles, and countless PC upgrades :P

Avatar image for ronvalencia
ronvalencia

29612

Forum Posts

0

Wiki Points

0

Followers

Reviews: 1

User Lists: 0

#2  Edited By ronvalencia
Member since 2008 • 29612 Posts

@scrollinglayers said:

It was in this presentation

I'd be happy if consoles in 2020 / 2021 gave us, oh maybe 20 TFLOPS.

At least it's good to know the semi-conductor industry wants to keep advancing forward like this, despite the challenges to making transistors smaller and frequencies higher.

If they succeed, that should be good enough for at least 3 more generations of PlayStation and Xbox consoles, and countless PC upgrades :P

Beyond Navi's 7 nm process tech, perhaps stack GPU with stack memory chips. CUBE shape GPU chip instead of near flat GPU chip.

At 40 TFLOPS per Navi 10 chip, you will need 25 GPU stack of chips to reach 1000 TFLOPS.

At 80 TFLOPS Fp16 per Navi 10 chip, you will need about 13 GPU stack of chips to reach 1000 TFLOPS.

Avatar image for ShepardCommandr
ShepardCommandr

4939

Forum Posts

0

Wiki Points

0

Followers

Reviews: 3

User Lists: 0

#3 ShepardCommandr
Member since 2013 • 4939 Posts

1000 tera flops? I doubt that is possible with silicon based transistors.As stands right now they can't even get 7nm working,let alone <5nm

maybe if graphene takes off