Polygon throughput - Accurate?

[MrWhiteUK]

Looking through various tech articles on the net I noticed there seems to be massive variation in the polygon throughput figures for hardware accelerators and the modern consoles.

Firstlty there seems to be no mention of textures/lighting used to derive said figure, surely this would affect the outcome greatly? Does 83 Million Triangles per sec mean single texture & 1 light? Does it include texture compression or several light sources?

Secondly, this applies only to the PC accelerators, is the final figure a theoretical maximum for the card alone or was it achieved in a PC, what about the spec? Can modern PC's actually supply these cards with enough data to achieve this maximum? This leads me onto my next question.

Are PC's becoming more and more of a bottleneck to the latest accelerator cards, is the agp port being pushed to it's maximum? If not now, soon?

Does each sucessive nVidia card improve on the last by an equal amount or is it more? If more is this why the cards get more expensive each release?

Have there been any tests to investigate the above? Is there such a thing as an accurate polygon throughput figure?

Can software be written soley to test a machines total throughput, has it been done?

Or is the Polygon per/sec a meaningless figure?

Just for kicks, anyone make an educated guess to the realistic throughput of the fastest consumer pc in polygons per sec? (textured and single light)

Many thanks

Dom.

 

[Sahakiel]

IMO, polygon count for video cards is the same as MIPS for CPU's.

 

[kazeakuma]

Been a while since I looked at this stuff, but here's an attempt at answering some of your questions.



<< Firstlty there seems to be no mention of textures/lighting used to derive said figure, surely this would affect the outcome greatly? Does 83 Million >>



Usually, the number quoted is for unlit and unshaded polygons (textures don't come into it). Add lights and textures and you can get huge performance drops(Lights especially). Eg. When the PS2 was about to be released Sony was throwing round a figure in excess of 40million pps. From memory the true output of the PS2 is somewhere near 10million pps with effects turned on.



<< Are PC's becoming more and more of a bottleneck to the latest accelerator cards, is the agp port being pushed to it's maximum? If not now, soon? >>



Yes and no. Take a look at the latest anandtech article on the new P4. In CPU intensive games (Comanche 4) the CPU is most definitely the bottleneck. However most games are not that CPU dependent so in those cases the video card can be the bottleneck. The AGP port isn't pushed to it's maximum yet, but for high polygon output it will become the bottleneck. Hence the move to AGP 8X.






<< Does each sucessive nVidia card improve on the last by an equal amount or is it more? If more is this why the cards get more expensive each release? >>



The performance gains are variable. While they are always there, not necessarily by the same amount. New features (GF3) can distort this somewhat too. As for the cards getting more expensive that is most likely a locational thing. I know that where I live, they are actually coming down in price again. They did go up excessively (TNT2 Ultra cost me $330, a GF2 Ultra would have set me back $1200) they have long since started dropping in price (GF4 Ti4600 is around $800).

There is a sort of standard benchmark for video cards (SPEC) which is usually used to judge professional 3D performance, not gaming. Usually however, it's best to ignore polygon output ratings as they are 99.9% meaningless unless it's with all effects turned on. Then you could judge (to a degree) relative performance. Even then it's extremely iffy (a tile based renderer could have half the poly count of a conventional renderer and match it).



<< Just for kicks, anyone make an educated guess to the realistic throughput of the fastest consumer pc in polygons per sec? (textured and single light) >>



Wouldn't have a clue :