| 02:07.56 | *** join/#debianppc donofrio (~donofrio@host-232.subnet-17.med.umich.edu) |
| 06:11.34 | *** join/#debianppc foolsh (~quassel@192.121.245.135) |
| 12:59.01 | *** join/#debianppc gustavold (~gustavold@189-69-97-199.dsl.telesp.net.br) |
| 14:58.37 | *** join/#debianppc hartb (~hartb@32.97.110.52) |
| 16:01.52 | *** join/#debianppc warai_otoko (~warai_oto@179.43.152.50) |
| 16:06.02 | *** join/#debianppc warai_otoko (~warai_oto@179.43.152.50) |
| 16:11.16 | *** join/#debianppc scottt (~scottt@user-0ccssgt.cable.mindspring.com) |
| 16:25.13 | *** join/#debianppc johnjay (~pi@38.120.10.99) |
| 21:07.51 | *** join/#debianppc kingsley (~kingsley@97-113-185-229.tukw.qwest.net) |
| 21:08.23 | kingsley | Hi |
| 21:08.55 | kingsley | I'm considering buying POWER9 hardware. |
| 21:10.34 | kingsley | I love how IBM released its instruction set architecture under a permissive license, and how Raptor Computing Systems has open source computers that actually work. |
| 21:12.39 | kingsley | The main reason I'm writing is to ask for help finding out how fast POWER9 CPUs can simultaneously use all their cores in parallel. |
| 21:13.11 | kingsley | Specifically, I'm interested in |
| 21:13.18 | kingsley | 1.) rendering video and |
| 21:14.20 | kingsley | 2.) other computationally intensive applications like AI and machine learning. |
| 21:16.32 | kingsley | If it would be convenient, comfortable, and all those good things, maybe you could share any thoughts you happen to have on how I might find someone willing to run a benchmark or 2 on a POWER9 system. |
| 21:18.45 | kingsley | Benchmark #1.) A nice, small, 15 line python script that's fully commented. It basically tries to measure how many multiplications are done per second by varying a.) how many cores are used and b.) the total number of multiplications, which I suspect interacts with how much cache is on chip. |
| 21:19.11 | kingsley | It's at http://loaner.com/multiplications_vs_speed_experiment.py |
| 21:19.46 | kingsley | It'd be great if someone could run it on a POWER9 computer with the python interpreter named "pypy3", |
| 21:21.00 | kingsley | If the results are relayed to me, I should be able to chart them in a format like those for an ol' 2 core AMD at |
| 21:21.12 | kingsley | http://loaner.com/multiplications_vs_speed_and_pythons_on_AMDs_Athlon.png |
| 21:22.02 | kingsley | I wonder if python's multiplication speed increases linearly with the number of cores under pypy3 on POWER9. |
| 21:22.15 | kingsley | If it does, great! |
| 21:22.19 | kingsley | Why? |
| 21:23.26 | kingsley | Python is one of, if not the, most popular programming languages for math and science. |
| 21:26.24 | kingsley | Since AMD's open source ROCm support for off loading math intensive applications to GPUs isn't ready for prime time yet, at least on POWER9, I'd like to find out how fast pypy3 and POWER9's AltVec multiply on multiple cores. |
| 21:30.13 | kingsley | Benchmark#2.) The second benchmark renders video. It uses a fractal generator built into the "ffmpeg" command. It is... |
| 21:30.19 | kingsley | $ time ffmpeg -f lavfi -i mandelbrot=s=1920x1080 -frames 50 /tmp/mandel.webm |
| 21:31.00 | kingsley | So, do my above comments make sense? |
| 22:15.12 | *** join/#debianppc psydroid (psydroidma@gateway/shell/matrix.org/x-dvveefgsylezietn) |
| 22:35.45 | *** join/#debianppc psydroid (psydroidma@gateway/shell/matrix.org/x-toxjkeuivofppanb) |
| 23:23.44 | *** join/#debianppc hartb (~hartb@cpe-70-124-159-19.austin.res.rr.com) |
