Slashdot Mirror
Carmack on NV30 vs R300
Nexxpert writes "John Carmack has posted his thoughts on the NV30 vs R300 (featured via www.bluesnews.com. Highlights some of the shortcomings of Nvidia's next step as well as pointing out what they've done right. Interesting read." In particular the arb2 vs nv30 path differences mean that it's not as simple as saying "ATI roX0rs nVidia" or vice versa.(update: sorry bout the misspelling, don't know how I missed that)
Probably, but only for about 6 months until that $400 video card turns into a $200 video card and the Mobo becomes the $99 special.
God I love this business.
I've been running ATI cards on my desktop since the mach64 chip days. When I got my 9700 in August, I NEVER thought I'd still have a chip that was competitive with nVidia's best offering in 3D. I never bought ATI cards because they were best in 3D or driver quality - they never were better. ATI did have superior 2D quality (to my eyes) and Video/DVD playback. Given I spend 90% on my time on a desktop, ATI had the right mix of features. Now they finally are competitive with nVidia's 3D.
After we started to get benchmarks showing matched performance, the remaining questions were left to DX9 and the more complex shaders. From Carmack's comments and the shadermark tests that are showing up, it appears that ATI is anywhere from competitive to superior in the DX9 2.0 shaders, as well. It does look like NV30 can indeed run deeper/higher precision shaders, but we will have to wait to see if games ever do show with shaders deeper than the LCD between NV30 and R300.
Carmack does mention that nVidia promised that "compiler improvements" will increase the NV30 shader performance. (Better scheduling of parallel pipes?)
The astounding bottom line is that as of Jan 2003, the 9700 is not shown to be inferior in any way to an as-yet unreleased flagship product from the king of 3d on the mainstream desktop.3 Cheers for ATI.
high end isn't where the money is at.. its the
mid to low end that generates most of the revenue.
You need not worry until Nvidia starts losing
OEM business to ATI.
on getting 15 FPS at 1280x1024 with all options turned on with a gf2.
*MAYBE* in an empty room with very few shadows and no combat..
On my 9700 Pro running on a 2.4 GHz Xeon, at that resolution, I normally got ~40-45 fps, but when anything happened, it dipped under 10. Sometimes in the 1-2 fps range -- when I was shooting, and being attacked.
No way in hell any gf2 is going to push 15 fps when anything is going on. It'd be a freaking slideshow slower than the powerpoint presentation at the meeting I just got out of.
No - this is decidedly not what he said. What he said was that the ATI precision mode that is used doesn't correspond to a precision that NVIDIA uses on the NV30, and the ARB2 precision mode corresponds to what ATI is using, but is "between" two for NV30. So the NV30 has to render it at its highest precision (rather than rendering it at a lower precision and artifacting the hell out of the thing) which slows it down.
The thing is that the renderer didn't want the higher precision, so the excess precision is mostly wasted. It's like calculating out 1.5000 * 1.5000 and getting 2.2500, and then truncating it back to 2.2. You could've just truncated it to 1.5 and 1.5 initially, and gotten 2.2 as well.
So, I doubt that NVIDIA actually looks better in the ARB2 mode. It just runs slower, because it's calculating things that don't need insane precision to insane precision.
Hence the reason that Carmack said that NVIDIA is confident there's a lot of room to improve, if it can realize which of the three precision modes is most ideal and shift to it.
I don't think you understood what I said. The thing is that the rendering path only asked for 96 bit precision - however the NV30 had to render it in 128 bit because it doesn't have a 96 bit mode. You're not going to get "free" improved image quality simply by calculating things out to more precision.
It's like in freshman-level science classes - you don't take the numbers out to more significant figures than you start with, because the added precision is meaningless. Carmack was talking about fragment path programs, for which added precision probably wouldn't pan out to added image quality.