News for hackers. Stuff that matters.
Re:The prospects for 128 bit processors (1995) by hajile on Wednesday February 23, 2022 @10:38PM (#30449927)
> Memory addressing

Current 64-bit CPUs generally don't go past 52-bits of memory addressing. That's 4.5 Petabytes and they still have 12 doublings to go. It is also possible to have memory addressing wider than your CPU. 8-bit chips almost universally did this. Even the P4 had 36 bits on a 32-bit CPU.

> Multi-byte processing

You can have wider vector load instructions without affecting the main CPU width

> Gaming/simulation

Legit point. IBM POWER has 128-bit IEEE floats and decimal hardware coprocessors, but I believe the rest of the CPU is still 64-bits wide. The easiest solution here is to allow those 256/512 vector units to do 128-bit floats in addition to 64-bit floats.

> Hashing

If you have a hashtable that needs 128-bits due to collisions, it is a MASSIVE table. It definitely won't be fitting into cache and probably not into RAM.

A few extra instructions (or even a hundred extra instructions) to do the math with two 64-bit segments will still be nothing compared to the thousands of cycles reaching out to RAM or the millions/billions of cycles reaching out to the disk.

Re:The prospects for 128 bit processors (1995) by gnufx on Thursday February 24, 2022 @11:19AM (#30456474)
> Legit point. IBM POWER has 128-bit IEEE floats and decimal hardware coprocessors, but I believe the rest of the CPU is still 64-bits wide.

Yes (POWER9+, I think), but also 128-bit "IBM" format. Transitioning to an IEEE ABI for ppc64le GNU/Linux has caused some pain in GCC and elsewhere recently. (I don't know how software long double is typically compiled and how fast the results are.)