The radeon would be the better of those two options, maybe even a 9800 pro, as these are rediculously cheap if you look around (make sure it's 256bit though!). Worth the extra money? well, yes. You need to remember that that is a plain 5700 too. Not a 5700 ultra. With nvidia, there is often a pretty big change between plains and ultras (far more than the plain->pro change with Ati).

But, as mentioned, you'd be far better to go with a 6600 GT. The standard 6600 isn't really a good idea though. It's too cripled. The 6800 plain is more expensive than the 6600 GT, and faster, but compared to the 6800 GT, it's like a wounded puppy facing a large steam roller. It's somewhat of an oddity, and the 6600gt gives far more bang for buck.

In the very unlikly case that you are on a PCI-express Motherboard, then a x700 plain is a very good choice. smacks the 6600 plain with a rickety fence post and is cheaper.

If you an afford to wait, then you may wish to wait it out and see what happens when Ati release the R520 chip. From what's been leaked this thing is in some ways unlike anything we have ever seen, so may well shake things up quite nicly. And that always means price drops. (plus ati will start doing more AGP ports of chips like the x700, and x800 XL)

Quote: Original post by lc_overlord
Sort of, they do have that extention but it's really a hack(it renders two polygons for each you send to it), at least for the 9800(the x800 should have it though), while the fx 5700,5800,5900,5950 can do it the correct way.
This means they are over two times as fast during the stencil part.

the reason geforce fx's or 6x00s are 'twice' as fast when writing stencil is the pixel pipelines can effectivly be run twice per clock when doing stencil write + zlookup, or some other very uncommon cases (such as colour write and no z/stencil at all). Doom3 was pretty much designed to work like this (including the later) which is where nvidia gets their advantage, as most of the rendering is effectivly running at twice the fillrate. Dispite this however, the 6x00s really still arn't all that much faster than xx00s given that fact.
Two sided stencil really doesn't have much to do with it here. You get a boost if your geometry bound, but if your using stencil shadows it's almost 100% certain your not.

Quote:
True for the most part, it's not five times slower(the actual number depends on what you do with it), but most raedon cards have a severely limited amount of instructions for a single shader, this means that you can't use larger shaders on the 9800, so you have to fall back on slower multipass rendering.

In some cases, a FX card can be well more than 5x slower. An example I've used before would be this scene from my code (the map darkmeat). There are LOTS of lights there. Probably each pixel is accumulating maybe upto ~10 lights. Odly enough the stencil-shadows don't really make too much of an impact here. On my readon 9500 I can comfortably run that at 640x480 at upto 25fps. On my former flatmates FX 5600 ultra it was significantly slower at 320x200. This was with shaders written in Cg too. And don't get me started on the 5200s...

Quote: Original post by RipTorn
the reason geforce fx's or 6x00s are 'twice' as fast when writing stencil is the pixel pipelines can effectivly be run twice per clock when doing stencil write + zlookup, or some other very uncommon cases (such as colour write and no z/stencil at all). Doom3 was pretty much designed to work like this (including the later) which is where nvidia gets their advantage, as most of the rendering is effectivly running at twice the fillrate. Dispite this however, the 6x00s really still arn't all that much faster than xx00s given that fact.
Two sided stencil really doesn't have much to do with it here. You get a boost if your geometry bound, but if your using stencil shadows it's almost 100% certain your not.

I saw some doom3 benchmarks a few years ago when the 5800(or was it the 5950) came, that benchmark showed clearly what the two side stencil could do.
I ran some tests in my engine, and i did get a few fps more when doing double side instead of single side(i have the 6800ultra card).
the theoretical 32 pixel piplines when rendering stencil volumes allso help, so does the third part of the ultrashadow2 feature, but you have to code that so i don't know if it counts as mutch.

Perhaps i'll make a benchmark demo once i am finished with building my shadow volumes directly into VBO's, that tests the different features by tuning them on and of, and then meshuring the delta time after that.

FiringSquad's relesed a review in the last 2 weeks for the PowerColor X800 XL card, which gets as good, or better performance than the plain 6800 for $300.

The review says:

Quote:
Their answer? The RADEON X800 XL. The RADEON X800 XL takes advantage of TSMC’s smaller 0.11-micron manufacturing process, allowing ATI to cram all the features present in high-end RADEON X800 XT/X800 XT Platinum Edition into a smaller core, including their 16 pixel pipes. The real kicker though is X800 XL’s MSRP: $300. This is $100 less than GeForce 6800 GT!

Edit : I see you're South African, like me :)

I haven't seen the X800 XL available here yet. As I type I'm passing up a 1300 buck special for a X800 PRO (R3,080 on special, normal price R4,330) because I'd rather wait for the X800 XL and pay significantly less (hopefully) than 3 grand.

the problem is, for the mean time, the x800 XL is PCI-express only...

Quote: allowing ATI to cram all the features present in high-end RADEON X800 XT/X800 XT Platinum Edition

What Features? X800 is a minimal altered 3 year old R300 core with twice the pipelines.