zlacker

[return to "The largest number representable in 64 bits"]
1. firebo+vJ6[view] [source] 2023-11-27 23:42:56
>>tromp+(OP)
You just need posits! John Gustafson will teach you how. Amazing concept. It would be amazing to see it in more hardware. It's especially good for AI and numerous other applications. Puts floats to shame. Of course this guy starts with integers, which are great, but for floats posits are amazing especially if you can control the bits dedicated to each 'fraction'.
◧◩
2. chrisw+7Q6[view] [source] 2023-11-28 00:23:51
>>firebo+vJ6
Their range (excluding infinity) is slightly higher than floats, but still abysmally tiny compared to the BLC encodings used in the article.

PS: I still prefer type 1 unums, due to their fallback to intervals :)

◧◩◪
3. firebo+wN9[view] [source] 2023-11-28 22:01:46
>>chrisw+7Q6
Not really. I don't think you've read his papers. The implementations in hardware lack range, but soft posits can expand to ridiculous ranges, and comparing the math, posits are superior(imo, granted, they're more or less similar) to BLC, depending on how they're implemented, of course.
[go to top]