S3 GammaChrome Specs

[swaaye]

Older titles often end up with problems because fixes for new titles break old things....

I'm not much of an S3 fan at all. They should have just changed their name entirely, IMO. S3 isn't much of a notable brand name. Savage 3D was slow and extremely buggy. Savage 4 was very buggy, and still is even with its final released drivers. Savage 2000 was a bad , broken joke. Virge was ok only for 2D and would lock up Supermicro 440FX-based boards (I had to get a Mystique instead). S3 Vision 968 even had 2D issues...etc etc..

S3's only major good contribution was introducing texture compression thru S3TC, IMO.

 

[Ailuros]

S3's only major good contribution was introducing texture compression thru S3TC, IMO.

While I can't disagree with the rest of the former post, that last sentence appears extremely harsh to me.

S3's engineers (former and remaining) have contributed quite a lot more than just that during the years to 3D. IMO too.

 

[KimB]

Like what?

 

[Ailuros]

Like what?

Like fast trilinear amongst others. DeltaChrome - despite it being a budget offering - sets an example with it's non-angle dependent AF.

 

[KimB]

Considering trilinear filteing has a significant performance hit today (and for good reason), you're going to have to do a lot better than that. And if the DeltaChrome was a compelling product, having good anisotropic filtering might mean something.

 

[Rolf N]

Angle-independent or not, DeltaChrome's AF really isn't all that hot. There's actually significant crawling going on. I know that the latest and greatest offerings from both of the big two have texture filtering issues, too (be it because of hardware or software), but I won't accept that as an excuse. I personally find this to be absolutely annoying and unusable. I somewhat hope this is unintentional, a hardware bug, because however bad that may sound, it allows me to hope that successors will be fixed and offer reasonable quality.

It really might sound like a bad joke, but I like the AF done by my Radeon 9800Pro much better than what the DeltaChrome does (i.e. I don't immediately feel the need to turn it off).

Trilinear OTOH is superb, way beyond R360. And I know it can't do "brilinear", which offers some extra peace of mind when compared to my FX5900XT

 

[Ailuros]

Considering trilinear filteing has a significant performance hit today (and for good reason), you're going to have to do a lot better than that. And if the DeltaChrome was a compelling product, having good anisotropic filtering might mean something.

DC would be decent enough for notebooks (which I'd figure might have been it's initial target).

"Better" is relative if the accelerator results in doing less than half the work on default. I personally was much happier with NV2x/3x class of AF and that irrelevant of performance.

zeckensack,

There's actually significant crawling going on.

Something like a LOD related issue or some sort of (MIPmap?) banding?

 

[Rolf N]

zeckensack,

There's actually significant crawling going on.

Something like a LOD related issue or some sort of (MIPmap?) banding?

No banding, no lod cheat. It is true aniso, it samples from the expected mipmap level(s), and it properly combines with trilinear. It simply takes too few samples for a given surface inclination, or at least so it seems.

I can only get rid of the crawling by pushing the lod bias up to +1.5 ~ +2.0 but that obviously sucks as a solution.

I've seen this in D3D and OpenGL, with app-controlled aniso and with forced-by-driver aniso, with all drivers ever released for the chip since June :?

 

[Ailuros]

I wasn't refering to any sort of "cheats", rather asking if there's something wrong with LOD values in general.

It simply takes too few samples for a given surface inclination, or at least so it seems.

That one sounds reasonable enough; I wonder if the nature of the algorithm is too limited or if it's an intentional performance optimisation (driver specific) after all.

I can only get rid of the crawling by pushing the lod bias up to +1.5 ~ +2.0 but that obviously sucks as a solution.

Sounds pretty familiar to the LOD issue for NV40's for the first six months or so (not that extreme that +1.5-2.0 would be needed, exept maybe Battlefield 1942). Since the negative LOD bias slider appeared (I think it was 67.02), most of the problems went away.