I figured we could use a hard benchmark to compare C8 performance against C8 pro. Having an engineering setup with full OS access, I grabbed a copy of the original DaCapo Benchmark Suite. The suite "consists of a set of open source, real world applications with non-trivial memory loads" and takes long enough to run to make a meaningful, if low tech comparison.
Because it's a headless environment, I had to skip a few failing tests from the suite. Here's the full script showing tests that ran and counted towards the final number:
date
java -jar dacapo-9.12-bach.jar avrora > /dev/null 2>&1
java -jar dacapo-9.12-bach.jar fop > /dev/null 2>&1
java -jar dacapo-9.12-bach.jar h2 > /dev/null 2>&1
java -jar dacapo-9.12-bach.jar jython > /dev/null 2>&1
java -jar dacapo-9.12-bach.jar luindex > /dev/null 2>&1
java -jar dacapo-9.12-bach.jar lusearch > /dev/null 2>&1
java -jar dacapo-9.12-bach.jar pmd > /dev/null 2>&1
java -jar dacapo-9.12-bach.jar sunflow > /dev/null 2>&1
java -jar dacapo-9.12-bach.jar tomcat > /dev/null 2>&1
java -jar dacapo-9.12-bach.jar tradebeans > /dev/null 2>&1
java -jar dacapo-9.12-bach.jar tradesoap > /dev/null 2>&1
java -jar dacapo-9.12-bach.jar xalan > /dev/null 2>&1
date
This is the script's output on a C8 pro:
$ ./run_benchmark.sh
Sun Jan 28 19:07:59 UTC 2024
Sun Jan 28 19:15:16 UTC 2024
$ cat /sys/devices/system/cpu/cpufreq/policy0/scaling_max_freq
2016000
Same on the C8:
$ ./run_benchmark.sh
Sun Jan 28 19:08:39 UTC 2024
Sun Jan 28 19:22:41 UTC 2024
$ cat /sys/devices/system/cpu/cpufreq/policy0/scaling_max_freq
1416000
Both hubs run at full speed, with C8 taking 14:02 to complete all benchmarks and C8 pro taking 7:17. That puts C8 pro's time at 51.9% of C8's time. That number is in line with our experience of C8 pro booting in about half the time of the C8.
Hopefully, this comparison helps!
