Alright guys, Cut it out, I'm tired of the personal attacks, Muya said once to settle down. If you guys dont we'll start enforcing the rules.
First off the NV3x cannot do FP16 only the NV40 can. Your thinking of 16bit precision. 16bit is only used when it does not affect IQ (which is most of the time). Based off of what you said I could make the argument that the NV40's 32bit gives better IQ than ATI's 24bit which is not true.
You are thinking of Floating Point Blending, Which is only supported by the NV4x in FP16.
Fp16 (64 Bit precision) Is fully supported by the NV3x. And 128 Bit (Fp32) is also fully supported by the Nv3x. In Any Event, Your right. he Geforce FX series are fully compliant DirectX 9.0 cards (with a few minor exceptions) They run shader 2.0 code. Albeit slowly, especially in Fp32 mode. But they can usually run partial precision adequately fast in some cases.