vainfo floating point error on systems with nvidia prior to series 8

Hardware:
Intel i686 prescott
nvidia 7400 Go
Software:
Gentoo 32 bit (x86)
nvidia proprietary driver 304.137
VA-API 1.2.0
libva 2,2.0+ (a patched version)

This problem probably can be extended to other machines. I googled a similar problem on AMD r600

vdpau doesn't seem to support card earlier than series 8 at all, neither do nvdec/nvenc.
nouveau supports the cards, but isn't exactly fast with old cards.

The card is classed legendary, Which is why I use propritary driver version 304.137 which is the most up to date for this series.
VA-API (vaapi) / libva seems to be the only way to support this configuration. And it does indeed accelerate the scaling and decoding.
Problem is vainfo reports floating point error (seen someone mention this problem on AMD r600 as well). Because of this error avidemux doesn't start unless I disable the VA-API support for this package.

glxgears seems to run 60 fps in low and full window. I guess I got vsync on and it might affect the result. Screen is 1440x1050. Which is why hw video scale is important. I tested it can decode 1080p and even HEVC realtime too, so sound like hardware decode is in play and up to date.