FP on Ryzen 1700 and core i7-8700

[Greyguy1948]

I have both so i can compare most software. Normally 8700 is somewhat faster on integer but not on floating point. Some Open MP programs are confusing me.
Look at this:
http://people.sc.fsu.edu/~jburkardt/c_src/openmp/
http://people.sc.fsu.edu/~jburkardt/c_src/quad2d_openmp/quad2d_openmp.html

Compute_pi is comparing SEQ and OMP. Both are bad for 8700 even compared with Athlon 845....
FFT_openmp is normally faster on 8700.
OS Fedora 64-bit
gcc -O3 -fopenmp

You can use this to test how many threads to use:
export OMP_NUM_THREADS=1
./a.out
export OMP_NUM_THREADS=2
./a.out
export OMP_NUM_THREADS=4
./a.out
export OMP_NUM_THREADS=8
./a.out
export OMP_NUM_THREADS=16
./a.out

 

[Greyguy1948]

Compute_pi
N .... mode...... time (sec)
10^9 ... SEQ ...... 4.07 ..... 8700
10^9 ..... OMP.... 0.41 ...... 8700/ 12T
10^9 .... SEQ ...... 1.21 ..... 1700
10^9 ..... OMP ..... 0.18 ..... 1700/ 12T
10^9 ...... SEQ ..... 2.73 ...... 845
10^9 ..... OMP ..... 1.37 ...... 845/ 4T

 

[VirtualLarry]

Not really too surprising. Intel's chips, are "faster" FOR GAMING. For general-purpose and scientific compute, Ryzen CPUs really pack a punch.

 

[Greyguy1948]

Yes we should see more of Ryzen on SPEC. So far only RATE and no SPEED on EPYC. Next generation...?

QUAD2D_OPENMP
Time(secs)....Threads......CPU
1.16..............12................8700
0.40..............16................1700
0.42...............8................1700
0.75...............4................1700
1.28...............2................1700
2.53...............1................1700
2.93...............4.................845
3.54...............2.................845
5.86...............1.................845

 

[LTC8K6]

Probably something amiss somewhere for the 8700 to be doing so poorly.

 

[Cerb]

Have you tried to see if -O2 changes things, as a start? Or, try a matching or similar -march. The 845's score makes me think it's more of a code generation problem than the CPU's capabilities. I mean, Zen should be competitive , if not excellent, for x87 to SSE2, starting to get OK but mediocre with AVX. But, that's an awfully big difference, and Excavator has no business besting anything faster than an older Core i3, here.

 

[Greyguy1948]

Looking at COMPUTE_PI -Os is better than -O3 for 8700.
-O2 and O3 are very close. -mavx or -mavx2 and -O3 are the way for 8700....

Opt...............mode......time(secs)
-Os...............SEQ.........3.70
-Os...............OMP.........0.34
-O3...............SEQ.........4.08
-O3...............OMP.........0.42
-O3-mavx....SEQ..........0.92
-O3-mavx....OMP..........0.17