Fillrate, how much is enough?

[boxleitnerb]

My calculator says, 746,496 MPix/s is enough for 120 fps at 5760 x 1080. So, all we'd need is a chip with this fillrate, that can produce every pixel single-cycle. ;-)

Is it that simple? That is less then 1 GPix/s, current cards have 40x that. So you're implying pixel fillrate is completely irrelevant nowadays?

 

[DavidGraham]

The best thing to see if you are fillrate limited is to watch the behavior at lower resolutions, If the card gains so much performance (excluding memory shortage) , then it is fillrate limited , if not then the bottleneck is elsewhere .

 

[AlNom]

Is it that simple? That is less then 1 GPix/s, current cards have 40x that. So you're implying pixel fillrate is completely irrelevant nowadays?

Don't forget overdraw/transparencies, shadows, multi-pass/multiple render targets/post-processing.

 

[boxleitnerb]

The best thing to see if you are fillrate limited is to watch the behavior at lower resolutions, If the card gains so much performance (excluding memory shortage) , then it is fillrate limited , if not then the bottleneck is elsewhere .

I don't think this is necessarily conclusive. Look at AMDs cards. At higher resolutions they can use their higher shading power and bandwidth well, but at lower resolutions they are not as efficient (frontend?) and loose their advantage a bit. How can we tell that apart from a fillrate limitation?

 

[trinibwoy]

Yeah it's going to be difficult to separate ROP fillrate from shader fillrate in any non-trivial workload.

 

[swaaye]

My calculator says, 746,496 MPix/s is enough for 120 fps at 5760 x 1080. So, all we'd need is a chip with this fillrate, that can produce every pixel single-cycle. ;-)

If only sireric was still around to get your message things could be done.

 

[Homeles]

To bad you can't disable/enable ROPs and other components at will. It'd be nice if there was a real way to test this.

yep, all the way up to 11.

Couldn't you just have fillrate 10 be extra powerful?

 

[silent_guy]

Couldn't you just have fillrate 10 be extra powerful?

???
...
...
This one goes to 11.

 

[CarstenS]

Don't forget overdraw/transparencies, shadows, multi-pass/multiple render targets/post-processing.

All in one cycle! Else my equation won't work.

If only sireric was still around to get your message things could be done.

I'll forward this to him on Facebook

 

[keldor314]

My calculator says, 746,496 MPix/s is enough for 120 fps at 5760 x 1080. So, all we'd need is a chip with this fillrate, that can produce every pixel single-cycle. ;-)

Uh, try 746 MPix/s. (Unless you're using European notation...)

But yeah, unless you have some ridiculous amount of overdraw, you're not going to come any where near the fillrate limit. I suppose you can run into this with something like smoke particles, but with FLOPs increasing much faster than memory bandwidth, it may just be best to raymarch through a procedural texture for that stuff at some point.

 

[Homeles]

Uh, try 746 MPix/s. (Unless you're using European notation...)

He's from Germany, so...