Jim Conyngham's comment in the section below the article summarizes it:
"Many of the claims in this article are simply impossible on their face."
You cannot do away with roundoff error and also claim that the size of the
number influences its accuracy. We already do a few of the things that
posits claim to do in pmemd, using a known size of the number to represent
in order to go more precision out of the int32 format than would be
possible with fp32. But the notion that 32-bit posits, which devote some
bits to telling us how many bits to have in the exponent before they even
get to the fraction, would approach the accuracy of double-precision
numbers with more fraction bits than the posits have bits is ridiculous.
This sounds like a very credulous reporter.
Dave
On Fri, Jul 19, 2019 at 1:00 PM Matias Machado <mmachado.pasteur.edu.uy>
wrote:
> "Gustafson told us that a 32-bit posit can replace a 64-bit float in
> almost all cases"... "he thinks posit-based arithmetic would deliver a
> two-fold to four-fold speedup compared to IEEE floats."
>
> ** WT#! Please, somebody tell Redhat and Canonical not to drop the 32-bit
> support yet :-P **
>
> "Perhaps the largest opportunity for posits is in machine learning, where
> 16-bits can be used for training and 8-bits for inference."
>
> ** AI with a pocket calculator?? **
>
> Matias
>
> ------------------------------------
> PhD.
> Researcher at Biomolecular Simulations Lab.
> Institut Pasteur de Montevideo | Uruguay
> [http://pasteur.uy/en/labs/biomolecular-simulations-laboratory]
> [http://www.sirahff.com]
>
> ----- Mensaje original -----
> De: "Bill Ross" <ross.cgl.ucsf.edu>
> Para: "AMBER Mailing List" <amber.ambermd.org>
> Enviados: Jueves, 18 de Julio 2019 0:29:52
> Asunto: [AMBER] Posits preparing to pounce on Floating Point
>
> Another important advantage to the format is that unlike
> conventional floats, posits produce the same bit-wise results on any
> system, which is something that cannot be guaranteed with the IEEE
> standard (even the same computation on the same system can product
> different results for floats). It also does away with rounding
> errors, overflow and underflow exceptions, subnormal (denormalized)
> numbers, and the plethora of not-a-number (NaN) values.
> Additionally, posits avoids the weirdness of 0 and -0 as two
> distinct values. Instead it uses an integer-like twos complement
> form to encapsulate the sign, which means that simple bit-wise
> comparisons are valid.
>
>
> https://www.nextplatform.com/2019/07/08/new-approach-could-sink-floating-point-computation/
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Jul 19 2019 - 11:00:02 PDT
