I have worked on the procedure to make it extensive, now everything happens in a loop and the user can specify how many loops to run. There is also a procedure to prepare the benchmarks directory.
Explanation of the procedure: the user clones gpuowl into a default 'gpuowl' directory with the following command:
Code:
git clone https://github.com/preda/gpuowl
After the user runs the following procedure to prepare the benchmarks directory:
Code:
#!/bin/bash
set -o xtrace
cp -r gpuowl/ gpuowl-bench/
And then run this procedure with two parameters: GPU name, and number of loops:
Code:
#!/bin/bash
set -o xtrace
# For each gpu model run this script changing the first argument to reflect hardware name;
# The loop number is the second argument: 10 = 1 million iterations per exponent;
# Keep gpuowl.log
# For amdgpu-pro a change is required: -use NO_ASM
cd gpuowl-bench
COUNTER=1
while [ $COUNTER -lt $2 ]; do
echo GPU PRP benchmark loop $COUNTER is running now.
echo ***********************************************
# First GIMPS discovered prime
./gpuowl -prp 1398269 -iters 100000 -log 5000 -cpu $1 -results bench-1398269-$1-$COUNTER.txt
# Last known prime
./gpuowl -prp 82589933 -iters 100000 -log 5000 -cpu $1 -results bench-82589933-$1-$COUNTER.txt
# Largest SEL computed exponent
./gpuowl -prp 332412937 -iters 100000 -log 5000 -cpu $1 -results bench-332412937-$1-$COUNTER.txt
let COUNTER=COUNTER+1
done