Mintmaster said:
Anyway, I don't think it's worth arguing much more. You are firmly in favour of NV30 due to its ability to execute fast shaders, which is true.
That's where you are wrong. Go read my posts when the R300 came out. There is this blanket assumption that if you are defending wrong-headed and unfair criticism that you are somehow "for" one over the other. When the R300 came out, I consistently defended it's "limitations" in shader ability, go check the threads. In fact, I even got accused of being pro-ATI at one point by Derek Smart. More than that, I actually have an R9700PRO in my system today.
I have been
adamant since day 1 that these two cards will be about on par and that they will each beat each other in some categories and fail in others and that there will be
NO OVERALL WINNER. It is the inability of people to process this and avoid turning everything into a damned horse race that leads to these bogus critiques. The attack on the size of the fan is the most absurd I've seen and the idea that people care about the extra slot is absurd. You won't see the same people complaining about USB or Audio connectors eating a slot. Today's systems have oodles of free slots, especially systems with integrated networking, audio, raid, scsi, etc.
Yes, I like NV30's (hypothetical) fast shader execution rate. I think that is the future. I also like ATI's AA, anisotropic filtering, and HW tesselation ability. I like that ATI went with a 256-bit bus, paving the way for others to make the leap. (although P10 and Parhelia beat them). I like ATI's multiple render-target support.
The fact is, both cards have things that I like and things that I don't like, and I wish there was a card that combined the best aspects of both.
As for the 8 pipeline issue, or the balance issue. You are still wrong. This is not a matter of being pro-Nvidia, it is a matter of being right or wrong. Fact is, the high clock allows the NV30 to both match the R300 in older game performance, but to gain a benefit in shader execution rate. It also means that they will have comparable stencil/shadow buffer rates. The card is simply "balanced" for compute-centric shaders.
The fact that NVidia went to .13 micron, yes, had an extra cost. However, like I said, both ATI and NVidia are going to pay this cost, and NVidia simply decided to pay it sooner. Nvidia needed .13um to squeeze 125M transistors into the process. Their die simply may not have been possible at .15um. Thus, .13um gave them a side benefit of increased clock scaling potential.
Moreover, since NVidia has paid this design cost, it will reduce the design cost of the NV31/NV35, since many of the kinks of the .13um design process have now been worked out.
The idea of a "balanced" card has to be taken in context of the use case for which the card is designed. We had this same discussion back in the days of 1, 2, or 3 TMUs per pipe. If the average game was a dominated by dual textured blending modes, then the 2 TMU card is more "balanced", since both TMUs will be non-idle most of the time. If games were dominated by single texturing, then the second TMU unit would be idle most of the time and the card would not be "balanced" in that it would have extra wasted bandwidth for the second TMU, which is not used 90% of the time.
Nvidia's whole strategy, right or wrong, is that they see the future as being compute bound. Even on short 16-32 length shaders, you need shader performance, and Nvidia has burnt transistors insuring this scenario is speedy.
Whether or not any "killer" DX9 games come out during the lifespan of the card is a separate argument. But the NV30 is balanced for the requirements for which it is designed. The extra pipelines exist, because, stencil/Z fillrate is important for future games like Doom3 with unified lighting.
If you were looking at future games like Doom3, Halo2, Splinter Cell, Unreal2+, etc today, you would design to maximize two things: z/stencil rate and shader compute rate. So, IMHO, a card designed for these scenarios is balanced.
