The Official S3 Discrete GPU Rumours & Speculation Thread

T-Buffer AA has programmable grid. RG/OG/sparse, ATi's "temporal" AA, stochastic sampling... all of that is possible

It's quite silly to criticise 3Dfx because of lacking HW AF. Enabling RGSS and setting adequate texture LOD results in better texture quality, than competitions would-be-AF could produce.

 

it's pretty damn unpopular to say that, but i agree 100%.

af isn't the end all be all especially when it still ain't perfect.

instead of af they could use higher res and uncompressed textures with rgss and adjustable lod mip bias..

athough i would like to at least have true trilinear instead of brilinear.

by the way, notice the perfect af shown in the g80 iq analysis. notice how it completely goes smooth all the way ahead, then notice how blurry oblivion's textures can be way straight ahead. there is a discernable difference between the g80's hq and the perfect af shown by the af tester.

 

OT I know,

Surely you meant "won't finish the story to fund his life"

Anyway, to bring it back "on topic", the T-buffer was just a new way of implementing an accumulation buffer with it's own new set of trade-offs.

 

Waiting for screens from a card other than NV30-G7x to show me what's wrong here.

No TC = lower detail textures and lower performance on various things, by the way. Inescapable. Cards would have to have more RAM and more internal cache size, among other things, to match doing things smarter w/ compression.

Screwing with LOD can cause lots of texture aliasing.

What does this mean exactly? If you don't force AF in the driver control panel, Oblivion doesn't use it. If you do force it, it works fine. The distance in Oblivion is heavily mip mapped and LOD'ed and will be blurry no matter what. This is a technical limitation caused by our computers not being 100x faster and having 100x more RAM.

 

Which is cleaned by RGSS and remains only sharp texture.

 

No.

 

Really?

 

Well by changing LOD you change where the mip maps come in. Without proper mip mapping, you get aliasing.

 

Because supersampling FSAA increases the texture sampling rate, you can use textures that contain higher frequencies (-> negative LOD bias) without aliasing.
Of course, this does not hold true for multisampling, where the texture sampling rate is not increased.

http://en.wikipedia.org/wiki/Sampling_theorem#Aliasing

 

Very much so. Have you actually tried it?On a surface that is rather prone to noise?Try it. And the magical "cleaning" efect of SS is considerably overstated. Unless you do ridiculous ammounts of SS(4X SS adds something along the lines of what 2X AF does). So, in short, no.

 

Gah, enough with this entirely tangential IQ discussion already! I need my S3 rumours and speculation, stat!

 

I speak about RGSS, which has double EER than oversampling. And yes, I have all officialy launched 3Dfx hardware and even Q3D HeavyMetal Mercury system. So I have quite a view of RGSS possibilities

 

I had a longer reply to this, but this sums it up nicely:what?

 

I was hoping that Cirrus Logic had gone ahead with their PC card based on the 3D engine of the M2.... yes, 3DO's / Matsushita's M2. the window for that was 1996/1997 when Voodoo1 and PowerVR PCX2 were the fasted consumer 3D cards around. M2 was equal if not slightly better in IQ and 3D performance. even with Voodoo2 coming out, an M2-based card woud've been nice as a single chip 3D accelerator. I'm assuming Cirrus Logic would have used the main M2 chip and not the PowerPC CPUs.

present day - I too would be happy if S3 came back strong in the low-end to mid-range segments. there's obviously not enough competition.

 

I can confirm 4x RGSS cleans much the texture aliasing. LOD -1.5 was a very reasonable setting. the bilinear filtering wasn't even a big issue as the mipmap boundaries were far away and close to each other, though of course it's far from perfect. with 16bit@22bit and compressed 32bit textures (if applicable) and a 800x600 res you had a pretty good compromise (on a 17" CRT) if the game wasn't too demanding.

I agree it's bad the thread has diverged, maybe not so badly. S3 reintroduced RGSS with deltachrome! will the new GPU support it? and is it safe to assume there's MSAA now. that GPU also looks awesome as far as we can tell from 3dMark scores (which I presume they're not nearly as meaningless as in the 3dmark03 days)

 

What are we talking about?Some recent title?With rather detailed surfaces, that are prone to noise?Or the stuff that was around when 3DFx actually existed?Because I'm quite certain you see the huge difference in terms of surfaces between then and now.

About S3...dunno, the Deltachrome and following Chromes also had nice 3DMark numbers being pushed by S3's marketing. They also sucked badly due to inept drivers and rather primitive architecture. Will this change?Maybe, who knows.

 

News...

S3 to volume produce new graphics chips by year-end

Chrome 460: DirectX 10, 90nm -> mid-range
Chrome 430: DirectX 10.1, 65nm -> entry

Maybe serious competition for ATi/NV in some price-segments?

 

Bumping this thread. Have we any new info on S3's new chips? Anyone who thinks they will be able to compete with AMD and NVidia's low-end offerings?

 

We have pictures of dies and names of SKUs:

D2(90nm) left, D3(65nm) right:


SKUs:
Chrome 460 (D2)
Chrome 450 (D2)
Chrome 440 (D3)
Chrome 430 (D3)

And some more details about 430:

http://www.3dnews.ru/news/videokarti_na_osnove_dx10_chipov_s3_graphics-270231/

But no infos about arch until now...

 

Small and simple boards, lower clocks then provious generations- I don't have much hope.

Wasn't Laguna lowcost incarnation of M2?