But does it do full bilinear in non-upsampled mode with only a little performance hit? 
It's not really the lack of filtering that's making my skin crawl with bad memories of mid-nineties gaming, it's the heavy aliasing, low texture res, texture warping, all kinds of visual anomalies. It just looks plain bad.
Your software renderer probably produces a much neater result (provided the source art is good enough), but I still have difficulty seeing software rendering as anything but a nice way to experiment a bit and do stuff that is difficult, and/or impractical in hardware.
I was much intrigued when you said your engine shades faster than hardware, but we really should not be that surprised. A modern FPU on a P4 or Athlon probably has a lot more resources dedicated to it, and it is clocked way way faster too of course... Not sure how many transistors are locked up in the FPU, but from looking at die shots, it's certainly a lot more than the ALU. Of course, a bunch of that will be due to stuff that isn't needed when doing 3D rendering, but raw hardware performance is certainly in favor of the FPU today.
Anyway, I don't want to sound as if I'm dissing your work; really, you're probably a programming wiz and your renderer is likely top-notch! The limit is rather with the hardware you have to work with, ie, the CPU itself...
Maybe you should seek employment with one of the big hardware designers and write low-level GPU description code instead?
It's not really the lack of filtering that's making my skin crawl with bad memories of mid-nineties gaming, it's the heavy aliasing, low texture res, texture warping, all kinds of visual anomalies. It just looks plain bad.
Your software renderer probably produces a much neater result (provided the source art is good enough), but I still have difficulty seeing software rendering as anything but a nice way to experiment a bit and do stuff that is difficult, and/or impractical in hardware.
I was much intrigued when you said your engine shades faster than hardware, but we really should not be that surprised. A modern FPU on a P4 or Athlon probably has a lot more resources dedicated to it, and it is clocked way way faster too of course... Not sure how many transistors are locked up in the FPU, but from looking at die shots, it's certainly a lot more than the ALU. Of course, a bunch of that will be due to stuff that isn't needed when doing 3D rendering, but raw hardware performance is certainly in favor of the FPU today.
Anyway, I don't want to sound as if I'm dissing your work; really, you're probably a programming wiz and your renderer is likely top-notch! The limit is rather with the hardware you have to work with, ie, the CPU itself...
Maybe you should seek employment with one of the big hardware designers and write low-level GPU description code instead?