RSX is a faster 7800GTX
- Thread starter sklaar
- Start date
- Status
Jawed said:
http://www.google.sk/search?hl=sk&a...ty+heads&btnG=Vyhľadať+v+Google&meta=
http://www.flipcode.org/cgi-bin/fcmsg.cgi?thread_show=5282
"Character models, are actually fairly low-polygon. They use bump-mapping to simulate much more geometry than they have. This allows for faster texture projection onto them, which is what I'm about to discuss.
Now to the meat of this post: the real-time lighting effects. We all know how great, how stunning lighting in environments can now be. You can have some really detailed stuff going. That is old news. New news is applying light through grates to your character, or things like that.
The lighting compiler will have to check if. in each section in the portal structure and such, a light casts a shadow onto geometry by means of any other geometry in the area. In this case, usually, a lightmap will be generated. However, this lightmap is to be inversed for transparent objects like a grid of metal and projected from the light. We'll get to that later.
Now, we have pre-calculated lightmaps for the entire sector we're in. We have a lightmap from a grid being projected fro ma light. This is where the lighting model "branches", depending on case.
Firstoff, the light is to project lightmaps onto the model through simplified geometry. After that, the model is to cast a shadow in the conventional method (render from lightsource, project z-buffer). IF the grid (which is a transparent polygon) is between the light and the model in question, said model is to have the inverted lightmap we mentioned before projected onto it also. But this needs to be done first, before any other rendering is performed, for speed purposes.
What about light between sectors? This will have to be done for every light, except that through-portal light can be pre-calculated even better, because you need to project many less lightmaps.
This only has us projecting roughly 3,000 polygons (500 polygon model * 6), which is really fast for a videocard nowadays, since you can project onto huge terarains with not that much slowdown on higher-end cards.
So now we've got spectacular lighting effects with realtime shadowing. This of course assumes these lights are static. But what about lights that move in a static way? Just go back to projecting their lightmaps on everything? This is what I think might be best. You don't have to project it onto all 10,000 polygons in a sector though -- you can precalculate which ones need to be projected onto."
... coz of stencil shadows.
Not for a shadow volume cast from that head it's not.Jawed said:
Fillrate complexity typically grows linearly with the number of polygons used to construct the volume (so long as the added polygons actually contribute to shape of the object).
And it's a linear function scaled with the number of lights affecting the object - so with 4 lights per object, doubling the polygons would mean ~ 8x more fillrate.
Shifty Geezer said:
It'd be flexible, perhaps, but from a performance perspective I'm not sure it'd be very desireable.
If what Kirk is saying is true, I don't think it a particularly good sign. What could be going so wrong on Xenos to encourage a dev to give up Xenos's greatest advantages (multi-task balancing and utilisation)?
I'm not sure you'd be gaining a whole lot in terms of pixel shading power over an "equivalent" dedicated architecture either, in such a scenario. A dedicated architecture already has a high proportion of its power invested in pixel shading, so while all of Xenos's power would be pumped into that, you're losing some efficiency within the shader (general vs dedicated)..how much, we don't know, but you'd have to hope your greater number of execution units overcomes that. You've also lost your utilisation advantage, and are taking a probable hit in terms of vertex performance and eating into available CPU performance for other tasks. Even if there was a pixel shading advantage over a dedicated architecture in that scenario - which I don't think is at all certain - I'm not sure if it'd be worth it in a lot of cases.
Of course, Kirk may well be telling porkies. And a lot of devs may be fine with it, which his comments don't preclude (but then some may be more performance sensitive than others too).
Alpha_Spartan said:Dude, get with the times! Anonymous sources are all the rage these days!!!gurgi said:
Heh. "I've even heard that some developers" doesn't even sound like he's talked to any of them, but somebody told him that somebody talked to a dev and that guy said....
To seriously try and gauge C1 performance based on public inverviews from Nvidia is pretty sad imo.
Shifty Geezer said:
http://www.beyond3d.com/forum/viewtopic.php?t=23937
Here's a thread on the topic and as ever, it's inconclusive!
Btw, this interview adds nothing that we already didn't know...
Jawed
Legend
Fafalada said:
Well, I still can't work out what you're talking about.
The silhouette of the head amounts to approximately the same area regardless of whether it's a cube or a sphere.
In general as the poly count increases, the average number of pixels per poly decreases. So the total number of pixels created in the silhouette would remain approximately constant, but be generated by more polys.
Jawed
seismologist
Regular
Jawed said:
From what I've read so far t doesn't seem that way at all. The more I learn about how the shaders work the more I agree with what David Kirk is saying.
Johnny Awesome
Veteran
Of course you do.
As others have stated: Judging Xenos performance from "anonymous" dev comments "relayed" by an NVidia PR guy seems pretty foolish.
Can we not have yet another thread where a bunch of Sony fanboys come in to debunk the performance of the X360 without anything to back up their claims but for their hatred of MS and Sony/Nvidia PR. It's ridiculous.
You don't see people like myself, QRoach, JVD, etc.. claiming that PS3 SPEs are useless just because we have a fondness for the X360. We don't take the word of J. Allard as gospel truth when it comes to GP performance as it concerns PS3 vs. X360. That would be ridiculous.
It would be nice if everyone learned their lesson from last generation and took a more balanced approach in the architecture wars.
As others have stated: Judging Xenos performance from "anonymous" dev comments "relayed" by an NVidia PR guy seems pretty foolish.
Can we not have yet another thread where a bunch of Sony fanboys come in to debunk the performance of the X360 without anything to back up their claims but for their hatred of MS and Sony/Nvidia PR. It's ridiculous.
You don't see people like myself, QRoach, JVD, etc.. claiming that PS3 SPEs are useless just because we have a fondness for the X360. We don't take the word of J. Allard as gospel truth when it comes to GP performance as it concerns PS3 vs. X360. That would be ridiculous.
It would be nice if everyone learned their lesson from last generation and took a more balanced approach in the architecture wars.
- Status
