G80 rumours

looks like Oct. for the luanch to be a bit more then likely

http://www.forbes.com/2006/08/18/nvidia-0818markets10.html?partner=yahootix

Interesting about the margins, its a bit vague but maybe the g80 isn't on .9?

 

They dont say Oct "ahead" could be Nov too. Though I want it asap.

 

Well I know but speculation r600 was nov. dec at least what the intial rumors were, now its dec- jan, so just basing it on that.

 

Pre-vista delay, now

 

Vista is supposed to be ready soon actaully wouldn't be suprised if it comes out Jan, Feb, RC1 for betaing is coming out soon end of this month.

 

I meant the delay of Vista from Q4 06 into Jan/Feb next year. The delay could have slightly altered the release schedules of the DX10 gpus.

 

Ah ok, miss understood , could be.

 

http://www.hardware-infos.com/news.php?news=559

english translation

 

I'm still sitting on 1st 1/2 of November in the B3D pool!

 

The german-to-english translation was looking fairly good..., at least until this little gem came along:

The article itself looks like another bit of pure speculation to me.
Although, of course, it is evident that the raw number of pixel and vertex shaders (32/16) is a bit irrelevant, as it's obvious we can't compare a Direct3D10 pipeline to a DX9c *equivalent*.

 

Weihnachtsglocken

 

And with that geo's horse passes Uttar's and Razor's ...

 

32/16 huh? Maybe if we knew what 1 of each could do we would have a clue about what the numbers mean for performance

 

I'm still going for 26th September

 

oh what the hell mid Oct for me!

 

now Nov for sure

 

32 PS with 3 full ALUs per PS and 16 unified VS/GS plus 32 TMUs pool available for PS and VS/GS. Looks interesting (But i'm pretty sure that that is a misinformation )

 

LOL yeah read the article just made a bad joke bit late over here.

 

why 3 per PS?
What if these 16 VS are not only VS/GS but capable of being PS also?

I always thought that if a chip has PS-only pipes and unified PS/VS/GS pipes in 2:1 or 3:1 or whatever is appropriate ratio, it will give benefit: (for example lets assume 48PS + 16 unified)
-smaller die size (48 PS vs 48 unified)
-capable of high peak PS ops (64)
-enough VS/GS power

 

Why increase the ALU/TEX ratio? Well, it's simple, really. Core clock speeds and memory clock speeds are scaling roughly in lock-step. But GPU cores are getting more and more dense. Thus it makes sense to dedicate more and more transistors to operations that don't require external memory accesses. In other words, ALU's.

Moving up to a 3:1 ALU:TEX ratio in nVidia's next architecture sounds like a natural progression from their current setup. They may even go for a 4:1 architecture if they end up reducing the number of texture units (which is still plausible, though I would personally question the wisdom of such a move, it may possibly end up being an overall performance win).