mersenneforum.org Interpretation of results
 User Name Remember Me? Password
 Register FAQ Search Today's Posts Mark Forums Read

2020-11-15, 17:12   #1
diocupro

Nov 2020

216 Posts
Interpretation of results

Hi,

I had run a benchmark test and I have problems interpreting it because I have no idea what the different values say.

This is one row of my output:
Quote:
 Timings for 2048K FFT length (6 cores, 1 worker): 2.42 ms. Throughput: 413.85 iter/sec. Timings for 2048K FFT length (6 cores, 6 workers): 18.79, 19.54, 18.55, 18.68, 18.50, 18.08 ms. Throughput: 321.19 iter/sec. Timings for 2048K FFT length (6 cores hyperthreaded, 1 worker): 3.07 ms. Throughput: 325.58 iter/sec. Timings for 2048K FFT length (6 cores hyperthreaded, 6 workers): 32.95, 19.79, 20.43, 17.03, 23.78, 17.67 ms. Throughput: 287.22 iter/sec.
Why does the throughput decrease as the FFT length increases?
What do the milliseconds mean?

 2020-11-15, 18:17 #2 Uncwilly 6809 > 6502     """"""""""""""""""" Aug 2003 101×103 Posts 2·4,583 Posts The bigger the FFT size, the more work that the processor needs to do per iteration. The milliseconds are the time per each iteration. The smaller this value, the faster the testing.
2020-11-15, 18:53   #3
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

2·32·269 Posts

Quote:
 Timings for 2048K FFT length (6 cores, 6 workers): 18.79, 19.54, 18.55, 18.68, 18.50, 18.08 ms. Throughput: 321.19 iter/sec.
Six workers, six corresponding average times per iteration at the stated fft length. The corresponding iterations/sec for each worker are 1000ms/sec / (average iteration time in ms). The total throughput is the sum of those six figures. Generalize from six workers to N.
1000/18.79 = 53.22 iter/sec
1000/19.54 = 51.18
1000/18.55 = 53.91
1000/18.68 = 53.53
1000/18.50 = 54.05
1000/18.08 = 55.31
Sum of six = 321.20 iter / sec (0.01 /sec difference is probably due to 2-digit roundoff)

Work required for an iteration is roughly exponent * log (exponent) * log( log(exponent)) and fft length is a nearly linear function of exponent, while the processor's rate of work is fairly constant. See for example the last two attachments of https://www.mersenneforum.org/showpo...19&postcount=5, right columns; constant within +-20% over 2M-64M fft length. (Numerous processor types have been exhaustively benchmarked and posted in that thread.) Large multiprecision multiplication is so for some rather fundamental reasons; see Donald Knuth, Seminumerical Algorithms or https://www.mersenneforum.org/showpo...21&postcount=7
If this still doesn't make sense that iteration time is dependent on fft length or exponent, time yourself for each of squaring a one-digit decimal number; a 4 digit, and a 10-digit.

What is most efficient on a given system depends on system and processor details and fft length. The optimal number of workers can change versus exponent or fft length. Dual-Xeon systems do MUCH better with 2 workers or more than with one; single-worker throughput on the Knights Landing I'm benchmarking now is positively dreadful with one worker (less than 10% of maximum in some fft lengths).
Hyperthreading usually is not an advantage in fft-based multiplication, but in some cases provides an advantage.
Benchmarking them is the right thing to do.

Welcome to the forum. And the learning curve.

Last fiddled with by kriesel on 2020-11-15 at 19:54

 2020-11-15, 18:56 #4 Runtime Error   Sep 2017 USA 21110 Posts The "best" setting based on your benchmarks is (6 cores, 1 worker) because it has the most throughput measured by iterations per second. Note 1: Running multiple workers usually won't help because the fast Fourier transform (FFT) size for candidate exponents is too big to fit into the processor's cache. RAM is needed to hold the information, so RAM speed often becomes the limiting factor instead of processor speed. Having multiple workers will only lead to greater RAM-bottlenecks. Note 2: Hyperthreading basically means the operating system can schedule two tasks for each core to perform because there is usually downtime between working on each task. This does not help for Prime95 since that task will fully utilize the core.
 2020-11-15, 19:17 #5 diocupro   Nov 2020 2 Posts Thanks for your explanation.
 2020-11-17, 23:09 #6 ZFR     Feb 2008 Bray, Ireland 2×3×19 Posts That's some cool homework you got.

 Thread Tools

 Similar Threads Thread Thread Starter Forum Replies Last Post bhelmes And now for something completely different 0 2019-10-21 16:10 kriesel Probability & Probabilistic Number Theory 1 2019-05-22 22:59 esakertt PrimeNet 3 2012-11-14 20:03 Unregistered Information & Answers 3 2010-07-26 00:49 __HRB__ Information & Answers 3 2009-10-22 21:50

All times are UTC. The time now is 15:49.

Fri Jan 15 15:49:54 UTC 2021 up 43 days, 12:01, 0 users, load averages: 1.66, 1.79, 1.76

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.