The Scottish Boinc Team

Benchmarks, smenchmarks?

I upgraded two systems to Ubuntu 17.10 -that comes with the 7.8.3 client- and my integer benchmarks exploded:

AMD Athlon 5350 (Socket AM1), running Ubuntu 16.04, BOINC-client 7.6.31:

di 28 nov 2017 02:13:50 CET | | Running CPU benchmarks
di 28 nov 2017 02:13:50 CET | | Suspending computation - CPU benchmarks in progress
di 28 nov 2017 02:14:22 CET | | Benchmark results:
di 28 nov 2017 02:14:22 CET | | Number of CPUs: 4
di 28 nov 2017 02:14:22 CET | | 2444 floating point MIPS (Whetstone) per CPU
di 28 nov 2017 02:14:22 CET | | 8932 integer MIPS (Dhrystone) per CPU
di 28 nov 2017 02:14:23 CET | | Resuming computation

updated this system to Ubuntu 17.10, BOINC-client 7.8.3:

Tue 28 Nov 2017 10:01:31 PM CET | | Benchmark results:
Tue 28 Nov 2017 10:01:31 PM CET | | Number of CPUs: 4
Tue 28 Nov 2017 10:01:31 PM CET | | 2343 floating point MIPS (Whetstone) per CPU
Tue 28 Nov 2017 10:01:31 PM CET | | 56329 integer MIPS (Dhrystone) per CPU
Tue 28 Nov 2017 10:01:32 PM CET | | Resuming computation

What is it that makes such a big difference?

scole of TSBT wrote: ↑Wed Nov 29, 2017 12:27 am What is it that makes such a big difference?

Better programming I hope. Do we know of (sub)projects that rely heavy on integer performance? I may be able to test it out....

Dirk Broer wrote: ↑Wed Nov 29, 2017 11:55 pm

scole of TSBT wrote: ↑Wed Nov 29, 2017 12:27 am What is it that makes such a big difference?

Better programming I hope. Do we know of (sub)projects that rely heavy on integer performance? I may be able to test it out....

WEP/Wanless maybe?

Primegrid

I just have my Raspberry Pi B+ on steroids -it thinks it can do a whopping 9162 integer MIPS (Dhrystone)- I will make it WE(e)P/Helpless...

First guesstimate: 11:47 hours. I'm not impressed..
My Pi-2's can do it in a little over 50% of that guesstimate, my Pi-3's in a little over 33%