R700 Inter-GPU Connection Discussion

[waver_01]

sexiest card ever?

 

[neliz]

sexiest card ever?

You do know the cooler-housing is shaped like a phallus?

 

[AlNom]

You do know the cooler-housing is shaped like a phallus?

The black makes it more believable.

*runs*

 

[waver_01]

The black makes it more believable.

*runs*

I lol'd

 

[fellix]

No heat-pipe sticking out?

Maybe this is a dummy assembly -- the holder plank is not even cut off from the PCB, yet?

 

[MfA]

What is the turn around time for DRAM fabs to switch from one form of memory to another? With 3 suppliers and the relative success of ATI I doubt there will be supply constraints for GDDR5 for very long. They're desperate for anything with a decent margin

 

[ShaidarHaran]

Bull? Go run some tests. No GPU rejects more than one triangle per clock. That is my point.

You're entirely wrong in thinking "most modern hardware is very good at tossing out invalid verticies." It's still only one per clock, and since G80, that hasn't been any faster than setup on any GPU.

Call me stupid - but what about Z culling? Doesn't this help with overdraw in determining which pixels can be rejected due to their position in screenspace?
Can't the same technique(s) be used to reject unneeded polys, or is this a much more complex task because polys can affect numerous pixels?

IHVs are always going on about new methods (or am I mistaken?) achieving improved performance in this area. If we could apply these same techniques to the application of poly-culling, I would think poly perf per frame would increase dramatically (effectively).

 

[ChrisRay]

The black makes it more believable.

*runs*

lol...

 

[coredump]

Call me stupid - but what about Z culling? Doesn't this help with overdraw in determining which pixels can be rejected due to their position in screenspace?
Can't the same technique(s) be used to reject unneeded polys, or is this a much more complex task because polys can affect numerous pixels?

IHVs are always going on about new methods (or am I mistaken?) achieving improved performance in this area. If we could apply these same techniques to the application of poly-culling, I would think poly perf per frame would increase dramatically (effectively).

Essentially, this is an extension of the clip planes that the hardware will already test against. Mintmaster's argument is that the GPU will have to process all the geometry regardless. i.e. to truly double the framerate in all cases then all points along the pipeline (including the CPU and application itself) need to process at twice the rate.

I was making the argument that most games are fragment limited, while Mintmaster was pointing out that the input to the fragment shaders have a large amount of geometric dependencies (shadow map creation as an example).

 

[Miksu]

So it seems that reviewers are just receiving the cards or may even have them already. Who knows, maybe we get some new leaks this weekend?

 

[Humus]

Well, I was thinking about a full shadowing solution, where your shadow frustum(s) must enclose the view frustum. Otherwise, you have the chance of missing some shadows.

Most games don't do that though. It tends to be rather costly for little gain. Shadows in the distance generally don't contribute that much to the overall impression so they tend to fade away at a distance.

 

[digitalwanderer]

You do know the cooler-housing is shaped like a phallus?

Yes it is, they actually modeled it after mine. I feel quite flattered.

 

[neliz]

Yes it is, they actually modeled it after mine. I feel quite flattered.

I heard Kirk takes one everywhere he goes..

So it seems that reviewers are just receiving the cards or may even have them already. Who knows, maybe we get some new leaks this weekend?

I don't think so.. it sounds a bit too early seen how tight the release schedules were over the past month.

 

[Miksu]

I don't think so.. it sounds a bit too early seen how tight the release schedules were over the past month.

At least Sampsa Kurri from Muropaketti told yesterday that his copy of R700 is in the mail. I would guess that he's not the only one who is soon visited by a courier

 

[Lukfi]

So, guys, what have you discovered about the R700 CF-sideport interconnect, besides that the cooler is shaped like digitalwanderer's phallus?

 

[AlexV]

So, guys, what have you discovered about the R700 CF-sideport interconnect, besides that the cooler is shaped like digitalwanderer's phallus?

That one is shaped like Digi's uretra :|

 

[digitalwanderer]

That one is shaped like Digi's uretra :|

SHHHH! Next they'll all be after me for picts to try and analyze!

 

[I.S.T.]

Yes it is, they actually modeled it after mine. I feel quite flattered.

 

[neliz]

At least Sampsa Kurri from Muropaketti told yesterday that his copy of R700 is in the mail. I would guess that he's not the only one who is soon visited by a courier

then I eat humble pie.

4 weeks then? launch in early august with better supply than the 4870

 

[sebbbi]

Most games don't do that though. It tends to be rather costly for little gain. Shadows in the distance generally don't contribute that much to the overall impression so they tend to fade away at a distance.

We actually do full shadowing on all light sources (each object blocks light from all light sources correctly), but majority of our light sources have rather limited distances (we have only distance limited spherical point lights and spot lights), so shadows are not that long usually. Meaning that majority of the shadow geometry is culled off nicely before sending it to graphics chip.

With unlimited shadow distances, all the objects have to occlude light (be "shadow casters"), or you get ugly looking artifacts from shadows that "go though geometry" (for example opponent behing a small hill on outdoor scene if the hill does not block the light correctly, or uses some alternative "fake" shadowing technique).