View Single Post
Old 2020-05-05, 19:38   #65
kriesel's Avatar
Mar 2017
US midwest

456710 Posts

Originally Posted by chalsall View Post
/some/ would value having some sort of sanity check on the GPU hardware. As said before, some of us [use] these things for more than just TF'ing (where determinism (or, at least, sanity) is coveted). There would be value in being able to determine if our "kit" was sane while running mfakt[c|o] (as opposed to a GPU based DC test).
Especially since issues can reside within specific circuitry or microcode that affects one computation type and not another, or for speed and low user involvement. Issues can be very localized, yet significant.

Remember the pentium fdiv bug; wrong output only for specific instruction and operand combinations. Various software workarounds were developed. The initial mass produced chips were wrong by design.
kriesel is offline   Reply With Quote