do we need a Geforce2 and Geforce 3 comparison, I can get them 🤣

@obobskivich
Well done good work 😀
This is an interesting thread! In we change driver, we will see a some differences?

Was there older version of AF tester for DX8? That would be great for old 3d accelerators.

Note also that there are various driver settings that will affect texture filtering. And some may not be exposed to the end-user. Even different driver versions could affect the texture filtering quality.
Texture filtering is a great place to cheat and save bandwidth to artificially inflate performance.

My personal favourite is the Kyro II PowerVR. They were accused of cheating because coloured mipmaps didn't show up as one would expect for trilinear filtering. It looked like the cards did bilinear filtering instead.
But no, the PowerVR guys are just more clever than everyone else: What they did was: they grabbed a block of samples from the larger mipmap, and then generated the smaller mipmap data on-the-fly from these samples, rather than having to access two separate textures.
As a result, the trick of giving each mipmap different colours didn't work as expected, obviously. So you couldn't see the colours being filtered, even though it was actually doing trilinear filtering.

Hope this necropost doesn't unleash the wrath of our beloved mods... especially since plenty of the captures on this thread have been taken down.

It has also recently become a curiosity of mine how most nV cards starting from the GeForce256 have a tendency to render mipmaps in old games with a significant amount of aliasing, going to the extent where even entire textures are drawn without any mipmaps at all, a la Japanese PS2 game (i.e Look at Normandie's grass and road textures in NFS Porsche).

Rage 128 to ATI R400 and R500, on the other hand, appear to render much smoother and more cohesive mipmap levels, up to the point I could say building a rig for dx8-9 games with an ATI card could make for a way more enjoyable experience. I observed this issue no matter what degree of anisotropy I used in games.

Why is this still the case? Is nV forcing certain optimizations which are buried at a silicon level?
I was surprised to notice such a shocking difference switching back and forth from my X1950 XTX to a 8800 GTX and even my RTX3090.

I swear there was a newer thread dedicated to this but I seem to unable to find it.

I do know the Geforce2 and the FX have the same exact aliasing issues, but also i've seen the same aliasing happen in LLVMPipe Gallium FWIW. so if anyone wants to software renderer the geforce aliasing, then there you go!!! This particularly is obvious when there's adjacent texels with less color value differences, which can frequently occur on lightmaps, and as the filtering occurs on R G B separately you'll see some discoloration in between the alias blocks- 32-bit color won't help you there.

anyway, nvidia suck's

I'm thinking severe texture aliasing would be a game specific problem. Or maybe driver issue. Some old PC games don't use mip mapping or don't do it well. Per pixel mip mapping wasn't even supported by all cards until like 1999.

The earlier GeForce cards do seem to have some strange issues with banding in textures though and their 16bit rendering is pretty muddy. And of course the DXT1 precision issue. I think Nvidia was just very specifically spending transistors to achieve an overall performance and adequate quality goal. Their quality did lag behind but they were also usually fastest.

this thread looked at filtering differences
Image Quality of various old video cards (Quake 3 comparison)

i'd disregard the early dxt compressor as that's just a topic of disingenous argument. It's a driver component that already had improvements in 2001, and only idtech3 games that DON'T ship DXT textures use it (for Savage4/2000 in mind, enabling it for nvidia's a mistake to begin with). By the time MOHAA came out (which has a texture compression checkbox to compress TGA/Jpegs AND ship higher res DXTs), it's been fixed.