Hacker News new | past | comments | ask | show | jobs | submit login

I thought the premium in quadros these days is in double operations speed. Since I need floats, not doubles I opted for two GTX 680s. They are screaming compared to Q6000, but I could always use more.



If an application uses floats in lieu of doubles might that be evidence that accuracy is not at a premium?


Indeed, 53 significant bits (doubles) vs 24 significant bits (floats). In my case float performance was what I needed. There are always other options, depending on your (precision) needs and needs in general, such as binary coded decimals, fixed point arithmetic...




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: