so, we are closing in on the 1/2 billion vertex mark, no?

[Megadrive1988]

are we not closing in on the half billion vertex/polygon per second mark?

with R300 pushing 1 million verts/polys per MHz. would that not mean that the 400 MHz and faster R300s are pushing that much more also? what is the fastest R300? 450-500 MHz, right? for every clock, 1M verts/polys, correct? even if it's all theoretical.

it seems to me, that in '03, we are set to shatter records, perhaps even approaching the 1 billion verts/polys per second barrier. we have:

*R350 that should clock at 400 MHz and over. even if it had the same number of vertex shaders as R300, it's still 1M per clockcycle. if it has more, then all the more polys/verts per sec.

*NV35 - will be out in 03 and if it follows Nvidia's NV20==>NV25 pattern, it could double the amount of verts per clock. whatever NV30 has in terms of vertex per clock, if NV35 doubles it, combined with a 50% increase in clock speed, we will be well on our way to 1 billion/sec. this should almost be expect from Nv. a large increase in vertex per clock performance, as this seems to be a serious weakness in NV30 compared to R300.

*R400 - could this be the first chip to shatter the barrier? being a whole new architecture. with ATI doubling the complexity of each new generation from R100 onward. no reason for this not to continue, right? now if you look at the jump from R200 to R300 in terms of poly/vert performance, it's something like 70M => 325M is it not? if there is a similar leap from R300 to R400, well, you can see how that would easily be over 1 billion

 

[KimB]

Re: so, we are closing in on the 1/2 billion vertex mark, no

are we not closing in on the half billion vertex/polygon per second mark?

Not for realistic applications, though perhaps in very limited, very specific scenarios.

with R300 pushing 1 million verts/polys per MHz. would that not mean that the 400 MHz and faster R300s are pushing that much more also? what is the fastest R300? 450-500 MHz, right? for every clock, 1M verts/polys, correct? even if it's all theoretical.

With nothing basic transform, the R300 can spit out one transformed vertex per clock, assuming no other limiting issues (bandwidth). That means at the default 325MHz clock, it can process up to 325 vertices, with absolute minimum levels of detail. Whether it can actually render that many is a function of the triangle setup engine, and in a real game, the vertex processor will be doing far more than just basic transform for pretty much any vertex.

In real world situations, expect the R300 to put out closer to 50-100 million vertices per second, possibly less with more advanced games (due to more advanced vertex program use).

it seems to me, that in '03, we are set to shatter records, perhaps even approaching the 1 billion verts/polys per second barrier. we have:

Um, okay. Not like this hasn't been done ever year for at least a decade in the computer market (Probably 2-3 decades).

Anyway, vertex processing power is no longer going to be about how many vertices you can transform, but how many operations you can perform. I think this will be particularly true for some of the less fundamental operations, such as sin/cos, ln, e^x.

In other words, due to the programmable nature, it's really becoming silly to attempt to pin down the processing power of the new graphics cards to something so simple as "vertices per second" rates.

 

[Megadrive1988]

ok true, things are not as simple as they used to be. and of course, peak vertex rate is not the same as sustained real-world use. I usually divide by 10 the marketing spec numbers of any given platform.

 

[Mintmaster]

Re: so, we are closing in on the 1/2 billion vertex mark, no

*R400 - could this be the first chip to shatter the barrier? being a whole new architecture. with ATI doubling the complexity of each new generation from R100 onward. no reason for this not to continue, right?

I don't think there is any point in increasing the triangle setup rate any more for a while now. It'll be mostly vertex shader speed that will be the focus of future vertex engines. I can tell you that as far as I know (not through rumours), R400 will be an absolute beast in this realm, due to some intelligent architecture. I can't say much more than that though...