mfakto: an OpenCL program for Mersenne prefactoring - Page 122

airsquirrels · 2016-01-01, 22:49

I figured out the unexpected PCIe bandwidth usage/poor performance on slower slots.

http://www.mersenneforum.org/showpos...0&postcount=37

I'm not sure if this was oversight or some rationale for using host memory I wasn't aware of. Changing the buffer types to GPU memory and updating mfakto across my systems made a significant (19.4%) increase in overall throughput.

It is possible some of the other benchmarks/performance of parameters or kernel selection may not compare the same once bus bandwidth is removed as a bottle neck.

kladner · 2016-01-01, 23:15

Quote:

Originally Posted by airsquirrels

I figured out the unexpected PCIe bandwidth usage/poor performance on slower slots.

http://www.mersenneforum.org/showpos...0&postcount=37

I'm not sure if this was oversight or some rationale for using host memory I wasn't aware of. Changing the buffer types to GPU memory and updating mfakto across my systems made a significant (19.4%) increase in overall throughput.

It is possible some of the other benchmarks/performance of parameters or kernel selection may not compare the same once bus bandwidth is removed as a bottle neck.

Bravisimo!

I am purely nvidia, but I am delighted at such discoveries on whatever front.

lalera · 2016-01-22, 17:52

i have a question
what are the optimal settings for a R9 390X (Hawaii XT) ?
i played around with some different settings
but i can not reach 880 GHz-days (source: mersenne.ca)
only about 600 GHz-days
i do tf exponents around 200 millions - bitlevel 67 to 76 running windows x64

ET_ · 2016-01-22, 20:54

Quote:

Originally Posted by lalera

i have a question
what are the optimal settings for a R9 390X (Hawaii XT) ?
i played around with some different settings
but i can not reach 880 GHz-days (source: mersenne.ca)
only about 600 GHz-days
i do tf exponents around 200 millions - bitlevel 67 to 76 running windows x64

Hi

The throughput of a factoring program often depends on the chosen range: higher (> 150M) and lower (<1M) exponents usually perform worse and give a lower throughput.

Luigi

lalera · 2016-01-23, 10:16

Quote:

Originally Posted by ET_

Hi

The throughput of a factoring program often depends on the chosen range: higher (> 150M) and lower (<1M) exponents usually perform worse and give a lower throughput.

Luigi

thank you !

lalera · 2016-01-23, 10:22

can do somebody please make a binary of mfakto v0.14 for windows x64
with the modifications of airsquirrels ?

kracker · 2016-01-23, 17:47

Quote:

Originally Posted by lalera

can do somebody please make a binary of mfakto v0.14 for windows x64
with the modifications of airsquirrels ?

.

Rodrigo · 2016-01-23, 22:07

Great work!

For some reason, though, when I extracted the mfakto.ini file, there seemed to be no carriage returns .

Rodrigo

wombatman · 2016-01-23, 22:15

Notepad++ is a handy tool. It recognizes returns of all types.

lalera · 2016-01-23, 22:30

Quote:

Originally Posted by kracker

.

thank you very much !

lalera · 2016-01-23, 22:38

Quote:

Originally Posted by Rodrigo

Great work!

For some reason, though, when I extracted the mfakto.ini file, there seemed to be no carriage returns .

Rodrigo

the mfakto.ini file is a linux text file type
you can open it with wordpad(for windows) and then save it
after this you can open it with notepad (dos-textfile)

2016-01-23, 22:07	#1339
Rodrigo Jun 2010 Pennsylvania 2×467 Posts	Great work! For some reason, though, when I extracted the mfakto.ini file, there seemed to be no carriage returns . Rodrigo Attached Thumbnails

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
gpuOwL: an OpenCL program for Mersenne primality testing	preda	GpuOwl	2718	2021-07-06 18:30
mfaktc: a CUDA program for Mersenne prefactoring	TheJudger	GPU Computing	3497	2021-06-05 12:27
LL with OpenCL	msft	GPU Computing	433	2019-06-23 21:11
OpenCL for FPGAs	TObject	GPU Computing	2	2013-10-12 21:09
Program to TF Mersenne numbers with more than 1 sextillion digits?	Stargate38	Factoring	24	2011-11-03 00:34

2016-01-01, 22:49	#1332
airsquirrels "David" Jul 2015 Ohio 11·47 Posts	I figured out the unexpected PCIe bandwidth usage/poor performance on slower slots. http://www.mersenneforum.org/showpos...0&postcount=37 I'm not sure if this was oversight or some rationale for using host memory I wasn't aware of. Changing the buffer types to GPU memory and updating mfakto across my systems made a significant (19.4%) increase in overall throughput. It is possible some of the other benchmarks/performance of parameters or kernel selection may not compare the same once bus bandwidth is removed as a bottle neck.

2016-01-22, 17:52	#1334
lalera Jul 2003 13×47 Posts	i have a question what are the optimal settings for a R9 390X (Hawaii XT) ? i played around with some different settings but i can not reach 880 GHz-days (source: mersenne.ca) only about 600 GHz-days i do tf exponents around 200 millions - bitlevel 67 to 76 running windows x64

2016-01-23, 10:22	#1337
lalera Jul 2003 13·47 Posts	can do somebody please make a binary of mfakto v0.14 for windows x64 with the modifications of airsquirrels ?

2016-01-23, 22:15	#1340
wombatman I moo ablest echo power! May 2013 13·137 Posts	Notepad++ is a handy tool. It recognizes returns of all types.