NV40 poll

[Geeforcer]

...

 

[Natoma]

You forgot 16x1.

 

[Geeforcer]

Sorry, forgot to hit "Add option"

 

[jvd]

its going to be a 1x16

 

[Geeforcer]

BTW, I know I didn't include it in the poll, but what about 8x2/16x1?

 

[Mintmaster]

BTW, I know I didn't include it in the poll, but what about 8x2/16x1?

Well, that really is the same as 16x1. When r300 is rendering a 2 texture polygon, it outputs 4 pixels per clock. It doesn't actually reconfigure itself to a 4x2, but that's the net speed. 2 clocks per pixel per pipe.

 

[MuFu]

What about 16x1/32x0, or better still, 2x(8x1/16x0)?

 

[Geeforcer]

BTW, I know I didn't include it in the poll, but what about 8x2/16x1?

Well, that really is the same as 16x1. When r300 is rendering a 2 texture polygon, it outputs 4 pixels per clock. It doesn't actually reconfigure itself to a 4x2, but that's the net speed. 2 clocks per pixel per pipe.

I know, but perhaps that notation would make more sense in the face of contradictory 8x2 and 16x1 rumors.

 

[MuFu]

It doesn't make sense because an 8x2 architecture can never output more than 8 pixels per clock.

 

[Geeforcer]

What about 16x1/32x0, or better still, 2x(8x1/16x0)?

2x(8x1/16x0)? As in, two chips? I personally find multi-chip approach extremely unlikely. Right now, I am leaning towards 8x2/16x0.75 configuration

 

[Geeforcer]

It doesn't make sense because an 8x2 architecture can never output more than 8 pixels per clock.

But what if it can work on more then 2 quads?

 

[kyleb]

then it has more than 8 pipes.

 

[Tim]

Well, that really is the same as 16x1. When r300 is rendering a 2 texture polygon, it outputs 4 pixels per clock. It doesn't actually reconfigure itself to a 4x2, but that's the net speed. 2 clocks per pixel per pipe.

I think the teori behind 8x2/16x1 uses the same trick as used in the NV31/NV34/NV36 where it only has two full pipes but the second texture unit can be used, with some limits, as an extra pixel pipe.

 

[duncan36]

Do people even realize that Nvidia failed at 6x2 with the Nv30? Given that people think they're going to do 8x2?

 

[Geeforcer]

Of course that would have requited then to aim for 6x2 in the first place.

 

[madshi]

http://www.theinquirer.net/?article=14872

NV40 is 16 pipelines part, sometimes.

If that is true, it sounds like 8x2/16x0 to me.

 

[Bouncing Zabaglione Bros.]

http://www.theinquirer.net/?article=14872

NV40 is 16 pipelines part, sometimes.

If that is true, it sounds like 8x2/16x0 to me.

The last three paragraphs of that article are most telling.

 

[Hyp-X]

Do people even realize that Nvidia failed at 6x2 with the Nv30?

No, and I don't buy that.

Just think about 6x2 - it can't execute 1.5 quads.
It might be 3 x 2x2 with the 2x2 part executing quads every other cycle.
It would be a stupid design probably more complicated than a 8x2. (2 x 4x2)

Actually executing multiple independent quads was a bold move from ATI with the R300, everything in NV3x is about that nV wasn't ready for a move like that.
Hell they weren't ready for 256 bit bus either - but that was relatively easy to add in a refresh.

 

[dan2097]

Damn after reading that inquirer story I want to change my vote to 8x2/16x0

 

[Arun]

For the last time... 16x1/32x0.
And it isn't even in the fucking vote options?! AAAAARGH!

Uttar