The End of IEEE 754 (pdf)

[Shade]

Abstract: popular systems for representing numbers in a computer are considered. Serious shortcomings of the IEEE 754 system are revealed, and options for systems to replace it are proposed. It's shown that addition is enough to perform calculations with numbers.

Link to the article: github.com/shadenova/Nova/blob/main/The End of IEEE 754.pdf

[Sensei]

Quote

The End of IEEE 754

Not really.. Existing file formats use IEEE 754 inside their files.. If you want to maintain backward compatibility and readibility of these existing files, you must support IEEE 754.

IEEE 754 is supported by hardware, CPUs, FPUs, MCUs, GPUs. Trying to change the standard in the middle of the IT age would take a dozen years.

[swansont]

!

Moderator Note

Making threads to link to articles is not in keeping with our rules on advertising. If you want to discuss something, post the material here.

 

[OldChemE]

At one time, back in the 60's, I learned to program computers in machine language, and I delighted in the fact that I could actually know, step by step, how the computer was performing its operations.  Since then we have reached the stage in technology where the actual operations performed by the computer are complete buried in layers of code-- and the performance is vastly enhanced.  Sure-- someone who knows a lot about the esoteric details might conclude that IEEE 754 was not the best approach.  BUT  its the one that things are built on.  What you have in in the linked article is someone who sees the inefficiencies in the "wheel of choice" and wants to re-invent the wheel.  The question is, can they demonstrate a financial and sociological benefit to the user of computation devices to make the change.  "This is better" doesn't cut it.

[studiot]

28 minutes ago, OldChemE said:

At one time, back in the 60's, I learned to program computers in machine language, and I delighted in the fact that I could actually know, step by step, how the computer was performing its operations.  Since then we have reached the stage in technology where the actual operations performed by the computer are complete buried in layers of code-- and the performance is vastly enhanced.  Sure-- someone who knows a lot about the esoteric details might conclude that IEEE 754 was not the best approach.  BUT  its the one that things are built on.  What you have in in the linked article is someone who sees the inefficiencies in the "wheel of choice" and wants to re-invent the wheel.  The question is, can they demonstrate a financial and sociological benefit to the user of computation devices to make the change.  "This is better" doesn't cut it.

+1

Do you knoe which IEEE standard is it that says most IT change will be a downgrading ?

[Sensei]

@studiot@OldChemE

BTW, do you know that there is fast trick to do sqrt in single precision IEEE 754

https://www.google.com/search?q=fast+sqrt+single+precision+ieee+754

https://en.wikipedia.org/wiki/Fast_inverse_square_root

 

CPUs/GPUs/MICs designers should make it instruction in hardware to have ultra-fast sqrts..

 

Edited October 5, 2022 by Sensei