GMP 5.0.1 (0.2) (runs mainly from CPU cache, should scale well for per Ghz comparissons)
https://gmplib.org/gmpbench.html
Score/Ghz, CPU, Ghz, total score
1223 Core i5 (Haswell) 2.9Ghz 3556
735 Corei7 970(Nehalem) 2.7Ghz 1962
687 Core2 (Conroe) 2.1Ghz E6400 1466
684 Power7 3.55Ghz 2427
612 AMD E-350(Bobcat) 1.6Ghz 979
373 G5 2.7ghz 1013
356 Arm Cortex-A15 1.7Ghz 605
288 Arm Cortex-A9 1Ghz 288
269 Pentium4 2.6ghz 915
226 Atom 330 1.6ghz 362

1223/373 -> Haswell core is 3.2* faster than G5 core.
(I assume the benchmark does not utilize multithreading)

(I wonder if it would be possible to run GMP on modern G4 and PA6T)

It's easy to make mistakes when benchmarking, also with real life applications:
http://www.mactalk.com.au/10/109907-powermac-g5-faster-than-mac-pro-i7.html

Modern microprocessors (pretty old):
http://www.lighterra.com/papers/modernmicroprocessors/
"WARNING: This article is meant to be informal and fun!"

Single threaded performance.
(importance of multicore support)

(and OpenCL effect...)
(openCL and spreadsheets... ?? !! )
(one more)

(trying to look how much faster A57 might be...) (faster than G5/A15?, especially if SW is 64bit)

UPDATE: Power7 Ghz was wrong.
U2: (Macpro from G5 era to intel 2013 or so) (NOTE! No G5 was in that graph, so it shows only x64 Mac evolution) (source)
This x86 example shows how also we are stuck to low performance growth when we use only single core.
So, multicore use is mandatory unless:
This (x86 example) shows that any CPU is fine(irrelevant), if GPGPU is enable and used as math co-processor.

(IBM & OpenCL)

PortableOpenCL https://github.com/pocl/pocl

(Computer Architectures Overview PDF)
"Another interesting feature of the architecture is a virtual address system which maps all addresses into a 52-bit space. In this way applications can share memory in a "flat" 32-bit space, and all of the programs can have different blocks of 32 bits each."
With that kind of virtual addresses enabled, we could have AOS4.1.7 + apps in one space, MorphOS+apps in another space and multiprocessing, memoryprotection etc. aware apps in third "space".

Multicore programming.

x64 is rather a new ISA than continuation of x86, therefore the difference between 32bit and 64bit SW performance is bigger there than on PPC.

Last edited by KimmoK on 16-Jul-2014 at 01:31 PM.
Last edited by KimmoK on 16-Jul-2014 at 01:30 PM.
Last edited by KimmoK on 16-Jul-2014 at 10:03 AM.
Last edited by KimmoK on 15-Jul-2014 at 12:17 PM.
Last edited by KimmoK on 14-Jul-2014 at 07:22 AM.
Last edited by KimmoK on 14-Jul-2014 at 07:13 AM.
Last edited by KimmoK on 02-Jul-2014 at 06:13 AM.
Last edited by KimmoK on 01-Jul-2014 at 03:20 PM.
Last edited by KimmoK on 01-Jul-2014 at 03:10 PM.
Last edited by KimmoK on 01-Jul-2014 at 03:07 PM.
Last edited by KimmoK on 01-Jul-2014 at 03:02 PM.
Last edited by KimmoK on 01-Jul-2014 at 12:35 PM.
Last edited by KimmoK on 01-Jul-2014 at 12:35 PM.
Last edited by KimmoK on 01-Jul-2014 at 12:33 PM.
Last edited by KimmoK on 01-Jul-2014 at 11:49 AM.
Last edited by KimmoK on 01-Jul-2014 at 11:47 AM.
Last edited by KimmoK on 01-Jul-2014 at 11:46 AM.
Last edited by KimmoK on 01-Jul-2014 at 11:44 AM.
Last edited by KimmoK on 01-Jul-2014 at 11:42 AM.
Last edited by KimmoK on 01-Jul-2014 at 09:38 AM.
Last edited by KimmoK on 01-Jul-2014 at 09:37 AM.
Last edited by KimmoK on 01-Jul-2014 at 09:23 AM.
Last edited by KimmoK on 01-Jul-2014 at 09:21 AM.
Last edited by KimmoK on 01-Jul-2014 at 09:18 AM.
Last edited by KimmoK on 01-Jul-2014 at 09:14 AM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?