Reading the thread it looks like an expensive Pi5....
My problem with Pi4 is surfing on the internet of today. L2 cache like 1024 kB is not much.
L3 cach like 3-4 MB and private L2 cache like 512kB.
Any CPU like A76, A77 or A78 could be OK
I am reading a test in BYTE May 1986. They are comparing 68010, 68000 and 8088 with VAX 11/780.
Real time for sieve of Erathostenes in compiled C:
Vax 11/780 .............1.7
AT&T 68010............2.4
PCXT 8088..............8.2
TRS80-16B 68000..6.0
Multitasking (1 and 6 processes)
Vax 11/780...
Is he interested in games?
I have some versions of Raspberry Pi
All of them have Python2 and 3 as standard.
There are some games with code in Python.
This is much like Basic code in the DOS age.
RPi also have C,C++ and Java.
Take a look at Rosettacode.org and find something to test!
Raspberry Pi 5 is not announced yet but what would you like to have?
I guess both ARM32 and ARM64 are demanded so X2 and X3 are not any alternatives.
How much cost and how much RAM?
Have you ever seen any Atom beating any Core i on any program?
I tested this c++ code:
long_prtimes.cpp from rosettacode.org
It goes up to 64000 but you can go to 128000 or 256000 with small changes.
My Goldmont Plus at 2.7 GHz is faster than Core i7-8700 at 4.3 GHz (64bit)
For Raspberry Pi 4...
How big are MacBook and IMac today?
No doubt they could make a better CPU than Core i7 but how much can it cost?
Volume is everything. Something like a server CPU?
I have tested all CUDA alternatives for Fedora but my Linux USB-stick is too small (only 32 GB).
So I got a 512 GB SSD as "live USB-stick" and dowloaded Fedora 31. But to make it a complete Linux is not possible. It is asking for a new HD to install. This was not the problem with the 32 GB...
I have found some interesting code for CUDA so I would like to test the nvcc with my GPU (Geforce 2070).
However my Fedora Linux want something very big dev system like 80 GB (or Nvidia want it with Fedora).
Any smaller alternatives in Linux?
OK branch-and-bound is very fast but not the same on ARM and X86_64 with gcc compiler.
With ARM (RPi 3) I have no problem with optimizing -O1, -O2 or -O3 and it works up to 18x18.
With core i7-8700 it works with 22x22 but all optimizing result in segmentation fault....
Thank you for your answer!
I'm using Raspbery Pi 3 and it is slow but the problem is to use long long for some variables.
long is the same as int with my compiler.
paths long long
Not working!
Number of paths must be around 6000 mega but I get 1900 mega.
Some other long long variable must be used?
Today AMD is small compared to Intel. EPYC has less than 5% of the server market. But if they grow up will TSMC have capacity enough or will they have to go back to 12 nm at GF?
TSP_brute
This code is using a 5x5 plan and all is very fast for a modern cpu.
I have increased the distace plan up to 12x12 and it works fine with 12! paths.
But at 13x13 and 14x14 it doesn't work...not 13! or 14!
Paths must be declared as long instead of int but still it doesn't work.
Any ideas?
Not ARM as far as I can see. I would like to have it for Raspberry Pi.
Free Pascal is an alternative for Raspberry Pi.
Python is standard as C and C++ in Raspberry Pi
Texas Ti-83 or TI-84 for schools are using Z-80 at 6 or 15 MHz. Some BASIC code you can find on rosettacode.org. As most 8-bit CPU you have no multiplication or division so it is up to add or sub.
This is matrix multiply for some cards (FP32):
Matrix multiply
It is very much up and down like a switchback. I guess the cache in a GPU is not as sophisticated as in a CPU.
This is Tesla V100:
Tesla with HBM2
Totally out of budget for most of us but it could be intersting to see which benchmarks are depending of BW and which are not.
Anyone testing some graphic cards like this:
GPU timing comparison
Xeon 12T/6C of Nehalem type is compared with Tesla 2070 GPU on matrix multiply. This is FP64 but FP32 should be more interesting.
No opt is used as 90 sec for a 3 GHz Xeon running 2000x2000 is really slow.
For a very long time FP32 and FP64 were the only alternatives. Most compilers even today are not ready for FP16. If we look at GPUs from AMD and Nvidia only few are faster on FP16 than FP32. But machine learning is on the way....
FP16 is of interest for machine learning but is it good enought for weather forcast?
http://spec.org/cpu2006/publications/CPU2006benchmarks.pdf
They use FP32 in 481.wrf but do they really need it?
More about FP16:
FP16 on Wiki
My idea about weather forcast is that all instrument data are not...
You have some chess engine rating here: CE rating 40/40
You need something like 300 games to be on the list. But you have to force some kind of variation or...?
If CE 1 starts with white and CE 2 starts next time it could mean 150 identical games.....not very good!
So are they using some kind...
Valgrind is an analyzing tool for your program.
Valgrind analyzing tool
It will only follow your program without optimation and it will slow down it like 1:15.
You can see memory errors, cache miss, branch prediction and more.
I have tested Raspberry Pi and Celeron J4005 so far.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.