The V5-6000 will be out....

[RoboTECH]

"horrendous amount of texture duplication...suck the bandwidth dry"

you and Ben are full of that FUD.

besides, it has 11.7 GB/s of throughput. you could drop a full 4GB and still have as much as the GTS-U.

"Plus this board has no T&L or bump mapping."

actually all cards (well, almost all) can do multipass embossed bump mapping. GLQuake 1.14 for example.

"The GF2 Ultra costs $100 less, has more features and will almost certainly be faster."

it'll be about $100, has more features, and will almost certainly be slower.

"I'm sorry but I just can't see why anybody would pay $600 for slightly better FSAA and to play Unreal. "

Neither could I, except that it'll have FAR better FSAA (go on Ben, start up with that Techno-babble of yours, you know what looks better IN PRACTICE bro! <G>) and will play UT, Unreal, Deux Ex, Vampire, ALL NFS's and ALL EA sports games much better, much faster, much smoother, and will look significantly better.

I will say that I see the GTS-U being faster in MDK2, Evolva and Q3 @ 1600x1200 w/TC enabled. oh yeah, and 3dMark2000.

 

[SmackdownHotel]

3dfx deliver on time? ...nah.

 

[BFG10K]

RoboTECH: you're really into those smileys today, aren't you?

actually all cards (well, almost all) can do multipass embossed bump mapping.

Is it rolling bump mapping that Carmack was talking about in Doom 3? I think it is because the V4/V5 can't do this.

 

[BenSkywalker]

Arrgh, I had a post typed up and hit the wrong button

Robo-

Where did you get the updated V5 6K specs? Last I saw had them at 1.33Gpix/GTex, about 50% more for the GF2U in GTex, which 3dfx has told as repeatedly is the most important

For bandwith, 4X FSAA could require sixteen reads of each texture per frame, that 4GB/s edge could easily be overwhelmed.

Far better FSAA? I'd say marginal at best. Have you heard about the latest Q3 points release? Texture compression disabled by default(as it should be), so that puts the boards back on equal footing for default configuration in image quality, with 3dfx or ATi users being able to match/best the comparable nV boards in speed while clearly losing in image quality by enabling texture compression.

UT looks much better and plays extremely smoothly on a GF board now, and the issue you had(4.02 is now 4.28 support) has been dealt with(email me if you want details, article should be up later today).

I think debating this is somewhat amusing, I know I wouldn't think about dropping this kind of cash on either of the boards, not with G800 and NV20 around the corner, but if we didn't have something to argue about, what would we do

 

[RoboTECH]

<< RoboTECH: you're really into those smileys today, aren't you? >>


:| :Q :disgust: :frown:

Ben:

<< Where did you get the updated V5 6K specs? >>

183 MHz core/mem speed. A little bird told me.



<< Have you heard about the latest Q3 points release? Texture compression disabled by default(as it should be), so that puts the boards back on equal footing for default configuration in image quality, with 3dfx or ATi users being able to match/best the comparable nV boards in speed while clearly losing in image quality by enabling texture compression. >>

indeed, but a quick check of the newest wickedGL drivers makes that TC image quality loss almost completely non-existent. AS it was, it was clearly far superior. I'll still stand by enabling TC on the Radeon and 5500 and disabling it on the GTS. Also, there are some SERIOUS issues with 1.25 (Aside from it being beta). I wonder how long those changes will last.



<< I think debating this is somewhat amusing, I know I wouldn't think about dropping this kind of cash on either of the boards, not with G800 and NV20 around the corner, but if we didn't have something to argue about, what would we do >>

exactly!!!!!

 

[PistachioByAzul]

Anyone else think that 3dfx should diversify and start offering a wider range of products designed for men with small penises?

 

[Weyoun]

just a few things...

ben:

where do those 16 texture reads come from? i'd really like to learn more about FSAA and the diff methods... i kinda understood your conversation with dave from beyond 3d on another thread, but started to sway when he was talking about only using the one texture read for multisample AA. if yoou wouldnt mind, could you go over it with me? thx

enginenr9:

yes, 3dfx should have had a different solution to high end rendering for such a talented group of engineers, but your post is clearly here to kick up sh!t. we dont want it...

robotech:

i dont know if 3dfx could clock the v56k at 183, dont vsa-100's have trouble in getting this high? even if one in every two chips could make it this far, sli at this level effectively gives you a 4 times larger chance of not being able to run at that clock, since only one chip has to screw up and you're toast.

 

[BenSkywalker]

Weyoun-

&quot;where do those 16 texture reads come from?&quot;

I'll try and keep this brief, this post assumes that you are familiar with at least the basics of the current boards FSAA implementation, if not then I can provide a few links that do a good job of going over them.

Because of the way the V5 6K's chips work, in tandem on the same frame, it is possible that each chip will need to read the same texture data to draw its' portion of the screen. This, without thinking about FSAA, could make the board read texture data four times instead of the single read a single chip solution would offer.

When you add FSAA into the mix, due to the method that 3dfx uses they draw a frame for each sample, ie 2X draws two seperate frames and 4x draws four. If the V5 6K runs into a situation where it needs to read texture data four times for a normal frame, it would need to read that same data sixteen times to utilize 4X FSAA.

The GF2U, OTOH, reads texture data only once and achieves its' FSAA by rendering that data at a higher resolution and then blending it down instead of using multiple frames at the native resolution and blending those images. When looking at FSAA and memory bandwith, the GF2U is significantly more efficient, though it also has quite a bit less raw bandwith.

The MultiSampling that Dave and I were discussing is a way of doing FSAA similar to that of how 3dfx does it now, just by only reading the texel data once. In this respect, it would save bandwith heavily over even the GF2 as it could be handled at the native resolution, but the problems I see, and what I was asking Dave about would be poor comparitive quality and also the need to process all pixel information in tandem. This would require, if my speculation is correct(and it is a guess based on what Dave described) a massive chip with four(minimum) or eight(better) pixel pipes with what probably should be three or four texturing units per pipe. That way you could handle FSAA with comparitively speaking very littler performance loss.

The problem is that it won't do anything for texture aliasing and probably wouldn't do much for pixel popping either. This is supposed to be countered by more sophisticated methods of texture filtering such as high tap anisotropic filtering and more advanced mip mapping in general.

I commented that L&amp;EAA would offer nearly the exact same results, which it should based on my take of what he is saying, without requiring the massive pixel pushing power that MS would require, and the bandwith requirements should be minimal at worst.

 

[Weyoun]

thanks for clearing that up ben, you're a lifesaver i just forgot about the effective bandwidth with 4 chips sli...

on what dave said, how does his method work, or should i get the thread again?
thanks again

 

[hans007]

i cant believe you guys havent gotten tired of debating this. Lets debate a new one, do nvidia card owners have bigger penise or 3dfx owners. What if you have both in one system? do you have 2 penises or is your penis twice as long? Its a mystery.

 

[RoboTECH]

EngineNr9:

<< Anyone else think that 3dfx should diversify and start offering a wider range of products designed for men with small penises? >>

Why? Do you feel left out?

Weyoun:

<< i dont know if 3dfx could clock the v56k at 183, dont vsa-100's have trouble in getting this high? >>

not at all. That's why most 5500 owners are up in arms about the pitiful benchmarks litering the web.

Ben, something to keep in mind WRT the 5500 and the 6000: when FSAA is enabled, the FSAA and rendering are split up as different jobs for each processor. in other words, one chip does the FSAA, the other chip does the rendering. So FSAA won't require the 4x texture reads.

hans: I am huge. Thank God, cuz I'm ugly, and I'd be quite single were it not for my manhood. I've had both a GTS and a 5500. HTH.

 

[BenSkywalker]

Robo-

&quot;Ben, something to keep in mind WRT the 5500 and the 6000: when FSAA is enabled, the FSAA and rendering are split up as different jobs for each processor. in other words, one chip does the FSAA, the other chip does the rendering. So FSAA won't require the 4x texture reads.&quot;

FSAA, as in the filter effect, is handled by the T-Buffer. Both chips are still utilized as rasterizers to generate frames, if not you FPS would be cut down to one eighth when rendering 4x FSAA. The blending process of FSAA is the simple part, obtaining the samples needed is the difficult one and one which both chips on the V5 5.5K, and all four of the V5 6K chips will be used for.