AMD's Fermi?

[k1net1cs]

I mean not in terms of ridiculous power consumption, but architecture-wise.
Apparently AMD decided to follow Nvidia's & Intel's footsteps on making future GPGPUs (and gfx cards).

http://www.anandtech.com/show/4455/a...ts-for-compute

Will be interesting to see how this'd turn out.
What do you guys think?

[Foamy4]

Looking forward to see massive parallel computing GPUs from AMD

[The Chubu]

So VLIW4 is going to last at most one year more only?

They've been with VLIW5 for like 4 generations (exept HD69xx)...

[Lane]

Originally Posted by The Chubu

So VLIW4 is going to last at most one year more only?

They've been with VLIW5 for like 4 generations (exept HD69xx)...

Only in low end GPU ( + APU with Bulldozer next year ) . If all work as intended it seems the new architecture will allready be launched this year on the next high end HD7xxx series cards ( low end HD7K will still use VLIW4 )


we have made allready 2 threads about it in news threads and AMD thread ( next AMD architecture ( AFDS - new AMD architecture unveiled ).

Not i dislike there's one more about it.

Anyway: - yes the next AMD GPU architecture is going to exploit massive parrallel and computing, it bring a lot of innovation too, on paper this architecture can look similar to Fermi, but in reality it seems goes a lot further in many aspect.


The andantech article is an excellent one,

i just add 3 mores for complete informations.

16 June article on PCper ( who was cover the Liveblog of the AFDS )
http://www.pcper.com/reviews/Graphic...PUs-and-Beyond

google translate from Pcwatch ( japan )
http://translate.google.com/translat...17_453941.html

Hardware.fr ( use a translate tools )
http://www.hardware.fr/news/11648/af...-gpus-amd.html

[deltatux]

Originally Posted by The Chubu

So VLIW4 is going to last at most one year more only?

They've been with VLIW5 for like 4 generations (exept HD69xx)...

You have to realize that VLIW4 and VLIW5 are only design methodology variants of the VLIW type of processor ordering. It's like CISC and RISC, actually they're all relatable.

VLIW = Very Long Instruction Word, all GPUs use the VLIW design methodology because it's designed for massive parallelization of data processing.

It's not an architecture, it's a design methodology.

From what I understand, VLIW4 is just VLIW done in 4 stages instead of 5 stages because AMD finds that the final stage is rarely used and could be removed so they can save die space.

deltatux