PrimeGrid
Please visit donation page to help the project cover running costs for this month

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise

Advanced search

Message boards : Project Staging Area : optimal WFS settings

Author Message
Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100017 - Posted: 19 Oct 2016 | 20:42:50 UTC

Weird, I thought wwwwcl would be most efficient when wieferich_blocks is a multiple of 16. But I got a 10% boost in peak throughput to over 73 Mp/sec in a single process with a GTX 760 during testing with

./wwwwcl -v -p 10000000000 -P 15000000000 -T Wieferich
by using
wieferich_threads=3
wieferich_blocks=9182

The number 9182 is totally nonintuitive, being 2 * 4591 (a prime). It was a typo for 8192.

I might have been punching below my weight in the latest challenge by using a power of 2.

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100022 - Posted: 20 Oct 2016 | 8:48:11 UTC - in response to Message 100017.

Weird, I thought wwwwcl would be most efficient when wieferich_blocks is a multiple of 16. But I got a 10% boost in peak throughput to over 73 Mp/sec in a single process with a GTX 760 during testing with
./wwwwcl -v -p 10000000000 -P 15000000000 -T Wieferich
by using
wieferich_threads=3
wieferich_blocks=9182

The number 9182 is totally nonintuitive, being 2 * 4591 (a prime). It was a typo for 8192.

I might have been punching below my weight in the latest challenge by using a power of 2.

I confirmed that this setting also works better on my rig with live PRPNet workload (4 processes). Upping threads to 4 in each process improves it slightly (with a slight hit on the BOINC workload).

It's easy to throw together a bash script that gets the best settings. Something like
for ((trials=1;trials<4;trials++)); do
for ((threads=3;threads<9;threads++)); do
for ((blocks=1024;blocks<10240;blocks++)); do
sed -i "s/^wieferich_threads=.*/wieferich_threads=$threads/;s/^wieferich_blocks=.*/wieferich_blocks=$blocks/" wwww.ini
echo `./wwwwcl -v -p 100000000 -P 150000000 -T Wieferich 2>&1 | grep "p/sec" | cut -d ' ' -f 7` $trials $threads $blocks >>results.txt
done; done; done; sort -n results.txt | tail -100
and come back in about 6 hours (more or less, depending on your GPU).

Interestingly, I peeked at the file part way through the process. Here's what I have so far:
40844322 1 3 4177
40884779 1 3 2389
41000993 1 3 3307
41069405 1 3 2819
41101492 1 3 2081
41101492 1 3 3847
41670280 1 3 2887
42378509 1 3 2053
43485056 1 3 3041
44010777 1 3 2621
Notice anything? Those numbers of blocks are ALL PRIME!

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100029 - Posted: 20 Oct 2016 | 17:02:33 UTC - in response to Message 100022.

The 100000000 -P 150000000 test range is done.
Prime wieferich_blocks are highly represented among the fastest settings.
Smooth numbers tend to occur among the slowest.

The 10 fastest trials

p/sec trial thr blks
45074779 2 3 2339
45043752 2 3 2894
45021832 3 4 1559
44938880 3 3 2818
44994650 3 3 2909
44914806 2 3 2858
44854733 2 3 2879
44847235 3 3 2962
44839740 3 3 2902
44796317 3 3 2423

The 10 fastest average settings
Avg.p/sec thr blks
43431435 4 2551
43228069 6 1621
42986109 7 1181
42837691 7 1523
42831624 8 1423
42813060 7 1439
42705003 6 1637
42631367 7 1459
42601576 6 1697
42432539 7 1801

Continuing that work for the 1000 fastest threads/blocks combinations and running them through the previous Wieferich range scaled up by a factor of 10.

for ((trial=1;trial<4;trial++)); do
sort -n results.txt | tail -1000 | while read speed xx threads blocks; do
sed -i "s/^wieferich_threads=.*/wieferich_threads=$threads/;s/^wieferich_blocks=.*/wieferich_blocks=$blocks/" wwww.ini
echo `./wwwwcl -v -p 1000000000 -P 1500000000 -T Wieferich 2>&1 | grep "p/sec" | cut -d ' ' -f 7` $trial $threads $blocks $speed >>results2.txt
done; done

Trials are in the outer loop to get similar tests widely separated in time.

The 1000000000 -P 1500000000 test range is done. That took 21 minutes.

The 10 fastest trials of the higher range are
p/sec trial thr blks prev p/sec
81260040 1 6 2011 42606654
81116224 1 7 1409 41834017
80695722 1 6 2083 41211978
80671732 1 6 1669 40744457
80559331 2 7 1753 40565188
80508269 3 6 1663 43124393
80504469 1 7 1753 42047753
80468388 3 6 1789 41719526
80391721 1 6 1597 40745076
80380085 2 6 1637 41972107

The 10 fastest average settings
Avg.p/sec thr blks
78070364 6 1597 40745076
77876765 5 2927 42334375
77499482 5 2851 42411334
77496457 7 1861 41843151
77478019 5 3571 41527737
77379951 7 2089 41520668
77364896 6 1951 41663162
77334062 6 2531 41522596
77055234 7 2741 40285609
77030912 6 2339 41362559

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100030 - Posted: 20 Oct 2016 | 17:17:44 UTC - in response to Message 100029.

You all know where this is going... there can be only one!

Does anyone care to replicate this experiment on their GPU?
I'd like to see how this differs for different combinations of CPU & GPU.

This chain of messages should be shuffled over to a new thread. Optimal WFS settings. Roger?

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100032 - Posted: 20 Oct 2016 | 19:17:36 UTC - in response to Message 100030.
Last modified: 20 Oct 2016 | 19:19:00 UTC

I've absolutely settled on using a prime number for wieferich_blocks.

However, using 6 threads is highly suspect, as each process utilizes at most 1.75 cores.

So by the highly dubious and unscientific process of survival of the fittest through increasingly larger test ranges, I decided to try a live PRPNet load with the settings

wieferich_threads=3
wieferich_blocks=1877


Running 4 instances of WFS with work from PRPNet and clients configured as above, and with 8 TRP loaded on BOINC, this is the same CPU loading I used during the challenge. I observe that the GPU is starved of CPU cycles at around 75% GPU utilization, but producing WFS results at the same rate I had before with 100% GPU loading. And no screen lag this time.

Previously I determined how much CPU was needed to fill the GPU to capacity. Now that there's more GPU headroom, I'll need to reduce BOINC loading.

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100034 - Posted: 20 Oct 2016 | 20:07:48 UTC - in response to Message 100032.

I've reduced the BOINC loading to 4 TRP workunits and using

wieferich_threads=4
wieferich_blocks=1877

keeps both the CPU and GPU humming at close to 100%.

I'm happy to report that my i7-5820's WFS throughput increased by 25% and is now at 54+/-1 M p/sec. This is up from 44M p/sec that I got in the challenge, and includes 4 BOINC TPS workunits (down from 8) running at the same time as 4 PRPNet clients.

Previously 48M p/sec was the best WFS throughput I could deliver on PRPNet workload using manual tweaking (with some multiple of a power of 2) and without BOINC loading.

This flies against my normal intuition about the best way to use hardware. I'll venture a guess that the way WFS is written creates contention for a resource inside the GPU, and that using a prime number for wieferich_blocks is a sly way of avoiding it.

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100035 - Posted: 20 Oct 2016 | 21:29:01 UTC - in response to Message 100034.

Yup, using primes and fiddling with threads and instances, this GTX760 produces maximum WFS throughput exceeding 59M p/sec. That's a 25% increase in maximum throughput over anything I saw previously.

See you next WFS or WSS challenge!

Profile Sysadm@NbgProject donor
Volunteer moderator
Volunteer tester
Project scientist
Avatar
Send message
Joined: 5 Feb 08
Posts: 1084
ID: 18646
Credit: 327,816,851
RAC: 98,788
321 LLR Gold: Earned 500,000 credits (636,742)Cullen LLR Gold: Earned 500,000 credits (506,692)ESP LLR Amethyst: Earned 1,000,000 credits (1,002,654)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,001,786)PPS LLR Amethyst: Earned 1,000,000 credits (1,158,608)PSP LLR Amethyst: Earned 1,000,000 credits (1,076,339)SoB LLR Gold: Earned 500,000 credits (637,941)SR5 LLR Amethyst: Earned 1,000,000 credits (1,027,811)SGS LLR Amethyst: Earned 1,000,000 credits (1,108,426)TRP LLR Amethyst: Earned 1,000,000 credits (1,321,197)Woodall LLR Gold: Earned 500,000 credits (500,901)321 Sieve (suspended) Bronze: Earned 10,000 credits (44,377)Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (4,967,439)Generalized Cullen/Woodall Sieve Amethyst: Earned 1,000,000 credits (1,818,632)PPS Sieve Double Bronze: Earned 100,000,000 credits (195,271,606)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Gold: Earned 500,000 credits (627,444)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (3,956,401)AP 26/27 Jade: Earned 10,000,000 credits (12,767,381)GFN Jade: Earned 10,000,000 credits (11,624,689)PSA Emerald: Earned 50,000,000 credits (86,755,160)
Message 100037 - Posted: 21 Oct 2016 | 12:00:39 UTC
Last modified: 23 Oct 2016 | 15:45:31 UTC

I started the first test-script

hardware:
- GPU: NVIDIA GeForce GTX 970
- CPU: Intel(R) Core(TM) i7-6700K (HT off)
work in parallel:
- boinc (3x rosetta@home on CPU, PrimeGrid/AP27 on GPU)
- some cmdline -interactions to monitor the system and the test

the ten fastest short tests

44590829 3 3 2184 44607139 1 3 1120 44649453 1 3 1066 44992387 3 8 3328 47222406 2 4 3822 47423569 2 4 3835 49318768 1 3 1232 49833593 2 4 3827 50204734 2 4 3833 53220285 1 3 1076


the ten fastest tests with the extended range
56207558 3 3 2450 35311148 56393396 2 4 1301 35981720 56592670 1 4 2467 34280215 56824036 2 4 2406 33914056 57421443 2 5 4882 33815742 57453495 2 4 1400 33720116 58267456 2 4 5394 38100757 62565651 3 7 2728 33975041 65504455 1 3 2255 35867716 70368519 1 4 2406 33878940

____________
Sysadm@Nbg
my current lucky number: 3299*2^1441747+1
PSA-PRPNet-Stats-URL: http://u-g-f.de/PRPNet/

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100048 - Posted: 21 Oct 2016 | 17:43:26 UTC - in response to Message 100030.

there can be only one!
Is no one is challenging that statement?
There might be a speed benefit to separate instances running with different (near-optimal) values for wieferich_blocks.

Profile Sysadm@NbgProject donor
Volunteer moderator
Volunteer tester
Project scientist
Avatar
Send message
Joined: 5 Feb 08
Posts: 1084
ID: 18646
Credit: 327,816,851
RAC: 98,788
321 LLR Gold: Earned 500,000 credits (636,742)Cullen LLR Gold: Earned 500,000 credits (506,692)ESP LLR Amethyst: Earned 1,000,000 credits (1,002,654)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,001,786)PPS LLR Amethyst: Earned 1,000,000 credits (1,158,608)PSP LLR Amethyst: Earned 1,000,000 credits (1,076,339)SoB LLR Gold: Earned 500,000 credits (637,941)SR5 LLR Amethyst: Earned 1,000,000 credits (1,027,811)SGS LLR Amethyst: Earned 1,000,000 credits (1,108,426)TRP LLR Amethyst: Earned 1,000,000 credits (1,321,197)Woodall LLR Gold: Earned 500,000 credits (500,901)321 Sieve (suspended) Bronze: Earned 10,000 credits (44,377)Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (4,967,439)Generalized Cullen/Woodall Sieve Amethyst: Earned 1,000,000 credits (1,818,632)PPS Sieve Double Bronze: Earned 100,000,000 credits (195,271,606)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Gold: Earned 500,000 credits (627,444)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (3,956,401)AP 26/27 Jade: Earned 10,000,000 credits (12,767,381)GFN Jade: Earned 10,000,000 credits (11,624,689)PSA Emerald: Earned 50,000,000 credits (86,755,160)
Message 100052 - Posted: 21 Oct 2016 | 18:08:34 UTC
Last modified: 21 Oct 2016 | 18:08:53 UTC

There might be a speed benefit to separate instances running with different (near-optimal) values for wieferich_blocks.

I'm still something behind, let me first determine the best configuration for one run
(unfortunately I am also still the next week on the road)
____________
Sysadm@Nbg
my current lucky number: 3299*2^1441747+1
PSA-PRPNet-Stats-URL: http://u-g-f.de/PRPNet/

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100053 - Posted: 21 Oct 2016 | 18:24:19 UTC
Last modified: 21 Oct 2016 | 18:25:38 UTC

I'm wondering if any claims I made about M p/sec for the whole GPU are misleading. In the context of running multiple instances, the only reliable sum of instance speeds is the second one printed on the terminal (or captured to a file) after all the instances start at the same time, because the start times become skewed and GPU loading starts to fluctuate. I haven't been quoting speeds that way.

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100058 - Posted: 21 Oct 2016 | 19:35:32 UTC

Heeding my own advice and totaling M p/sec from the second %complete message from each instance, all started at the same time with a 1-workunit batch from PRPNet (the first %complete message is no good since network latency gives some instances a head start without competition from the others - throw away that first cup o' hooch).

Here's the aggregate throughput on 4 prpclient instances with 4 threads per instance. GPU utilization is 99 or 100% in each case.

blks Mp/sec
1211 47.7
1212 39.6
1213 57.3
1214 44.8
1215 42.4

Guess which one is prime, folks.

For the record, here are the factors.

1211: 7 173
1212: 2 2 3 101
1213: 1213
1214: 2 607
1215: 3 3 3 3 3 5

I'll stand behind these numbers. Way back. For a real benchmark, we'll want to control the actual p range and run wwwwcl on the command line.

GPU Specs
EVGA GTX 760 factory over-clocked to 1176 MHz (graphics) and 6008 MHz (memory)
1152 CUDA cores
2 GB RAM
256 bit memory bus
PCIe 3, x16 link, slotted that way too.

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100059 - Posted: 21 Oct 2016 | 19:53:46 UTC

And just for fun... let's blame the hardware for poor performance (1152 CUDA cores).

blks Mp/sec
1151 54.3
1152 44.6
1153 54.6

factor 1151 1152 1153
1151: 1151
1152: 2 2 2 2 2 2 2 3 3
1153: 1153

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100060 - Posted: 21 Oct 2016 | 20:37:11 UTC

Hey, here's an idea....

The runtime of the wwwwcl seive is algorithmically linear in the size of the range p, right?
But empirically we find that runtime of wwwwcl depends on primality of the blocking factor N, regardless of p.
Hence we have a new general test of probable primality for N in the runtime of wwwcl for a range of p.
Since runtime depends more on p than N, we can get an answer for the primality of N very quickly by using a small range for p.
Let's christen this the Wieferich Test Function (WTF).

It goes like this:
Run 3 wwwcl tests on a small range p, using N-1, N, and N+1 for the blocking factor.
We know that at least 2 of N-1, N, and N+1 are composite.
Compare the runtimes. If the test for N runs like 25% faster than the others, then WTF it's probably prime.

Sorry, I couldn't wait until April 1.

Joe O
Send message
Joined: 13 Jul 07
Posts: 51
ID: 9819
Credit: 19,694,269
RAC: 0
PSA Jade: Earned 10,000,000 credits (19,692,669)
Message 100394 - Posted: 29 Oct 2016 | 16:31:15 UTC

For three consecutive numbers, one or two are even, ie divisible by 2 ie composite.
Take a look at the odd ones. If the last digit is a 5 it is divisible by 5.
Add all the digits of the number together, again, and again till the final result is a single digit. If that digit is 3, 6, or 9, the original number was divisible by 3.
Once you have eliminated numbers divisible by 2, 3, or 5 test the remaining candidate(s).
We really should use 6 consecutive numbers, or we run the risk of eliminating all the candidates.

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100409 - Posted: 30 Oct 2016 | 4:58:26 UTC
Last modified: 30 Oct 2016 | 5:00:22 UTC

My comment was sort of tongue in cheek.

But the point of the post is about using a timing side channel of another computation to guess at primality of a number, without obviously operating on the number in question. Comparing the speeds of too many consecutive numbers defeats the high speed of this "probable primality test". Mind you, it's not an efficient side-channel for numbers at the size we've looked at, which can be checked millions of times faster by trial division. At what size of prime number does this side channel break down? Memory limitation becomes a problem for larger numbers.

And why is the WFS test doing that at all? I haven't seen the source code. There is an analagous timing side channel with Wall-Sun-Sun sieve, but that timing test is not identifying prime numbers, nor anything else I can identify.

Joe O
Send message
Joined: 13 Jul 07
Posts: 51
ID: 9819
Credit: 19,694,269
RAC: 0
PSA Jade: Earned 10,000,000 credits (19,692,669)
Message 100423 - Posted: 31 Oct 2016 | 11:06:11 UTC

MSI GTX 750TI AMD A6-3650 2.60 GHz 8GB memory

block Mp/sec Cores
5443 16.09 1.87
5393 16.02 1.86
5279 16.20 1.87
5179 16.11 1.86
5119 16.17 1.87
5107 16.41 1.88
5099 16.59 1.93 <<<
5087 16.37 1.87
5081 16.42 1.87
5077 16.45 1.89
5039 16.30 1.90
5003 16.11 1.87
4591 16.10 1.81
3833 14.80 1.69
2621 15.24 1.72
2551 15.16 1.71
2339 14.76 1.68
2011 14.26 1.63
1877 13.98 1.58
1156 11.84 1.35
1153 11.92 1.37
1151 11.87 1.33

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100464 - Posted: 31 Oct 2016 | 23:41:03 UTC - in response to Message 100423.

MSI GTX 750TI AMD A6-3650 2.60 GHz 8GB memory


You have a nice list of prime numbers there, Joe, except for 1156, the only composite in the list. How did you select those numbers?

Joe O
Send message
Joined: 13 Jul 07
Posts: 51
ID: 9819
Credit: 19,694,269
RAC: 0
PSA Jade: Earned 10,000,000 credits (19,692,669)
Message 100467 - Posted: 1 Nov 2016 | 0:54:56 UTC

I started with the best results from your numbers and then added selected primes from https://primes.utm.edu/lists/small/10000.txt
I meant to use 1152 but mistyped it to get 1156.
1152 would have given 11.73 1.35

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100493 - Posted: 2 Nov 2016 | 8:13:01 UTC - in response to Message 100467.

I expect every card/computer combination to have different optimal settings. Can you run the first shell script in this thread? That does a more thorough test than hand-picking. You might want to reduce the maximum and minimum blocks to make it end sooner.

The number of threads is going to affect the maximum speed as well. I mostly use 3 or 4 threads.

Oops, 1152 is not prime.
It has a lot of small factors.
1152: 2 2 2 2 2 2 2 3 3
Try 1153.

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100600 - Posted: 5 Nov 2016 | 18:52:58 UTC - in response to Message 100423.

MSI GTX 750TI AMD A6-3650 2.60 GHz 8GB memory

block Mp/sec Cores
5099 16.59 1.93 <<<

Is that a single instance or the total rate for all simultaneous instances? I was peaking around 20 Mp/sec for 1 instance running alone.

Joe O
Send message
Joined: 13 Jul 07
Posts: 51
ID: 9819
Credit: 19,694,269
RAC: 0
PSA Jade: Earned 10,000,000 credits (19,692,669)
Message 100602 - Posted: 5 Nov 2016 | 20:06:46 UTC

All of my results are for single instances with cores=2
I guess I could run two instances, but I tend not to do that to prevent overheating. I do have a factorial instance running on that machine now as well.

Joe O
Send message
Joined: 13 Jul 07
Posts: 51
ID: 9819
Credit: 19,694,269
RAC: 0
PSA Jade: Earned 10,000,000 credits (19,692,669)
Message 100609 - Posted: 5 Nov 2016 | 23:33:40 UTC - in response to Message 100059.

And just for fun... let's blame the hardware for poor performance (1152 CUDA cores).

1153 was in the table
and now for another combo
MSI GTX 750TI AMD A6-3650 2.60 GHz 8GB memory

block Mp/sec Cores
1151 10.54 .18
1152 10.54 .23 235
1153 10.44 .19
1156 10.62 .24
1877 12.81 .67 193 sec
2011 13.71 .67 181
2339 14.30 1.18 178
2551 13.08 1.03 190
2621 13.38 1.05 185
3833 17.43 1.38 143
4591 19.33 1.66 129 <<< local max
5003 17.08 1.70 146
5039 17.21 1.71 145
5077 17.32 1.71 144
5081 17.33 1.70 144
5087 17.35 1.69 143
5099 17.42 1.69 143
5939 19.90 1.71 126
9973 20.37 1.89 122 << :)

Joe O
Send message
Joined: 13 Jul 07
Posts: 51
ID: 9819
Credit: 19,694,269
RAC: 0
PSA Jade: Earned 10,000,000 credits (19,692,669)
Message 100610 - Posted: 5 Nov 2016 | 23:35:40 UTC

and for my original combo
MSI GTX 750TI AMD A6-3650 8GB memory

block Mp/sec CPU
9973 17.16 1.99
5939 16.42 1.92

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 100768 - Posted: 10 Nov 2016 | 14:29:46 UTC - in response to Message 100609.

I'm sorry, I don't see where you say how your combos are different. Is it by processor speed? By what else is running?

Joe O
Send message
Joined: 13 Jul 07
Posts: 51
ID: 9819
Credit: 19,694,269
RAC: 0
PSA Jade: Earned 10,000,000 credits (19,692,669)
Message 100876 - Posted: 14 Nov 2016 | 2:28:25 UTC - in response to Message 100609.


and now for another combo
MSI GTX 750TI Intel Core I5 760 2.60 GHz 8GB memory
is what I should have posted for teecond combo

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 101000 - Posted: 17 Nov 2016 | 12:02:16 UTC

Depending on the speed of your CPU and GPU (and how many threads are set up in wwww.ini), WFS needs more than
one CPU core to pump each instance of a GPU task, and you will run multiple instances to keep the GPU fully occupied.

For example, at the end of a workunit, the output of WFS on my rig shows the line

CPU/GPU utilization: 1.30 (cores), 0.39 (devices)
This says I need to run 1 / 0.39 =~ 3 (rounded up to next integer) instances of WFS
to use the full processing capacity of the GPU, and I need to use 1.3 cores per instance.

An efficient way of using 1.3 cores is to team up the pair of logical processors on a physical core with hyperthreading turned on.
This provides about 1.5 cores of CPU power with resource contention. This an excellent match with the need for 1.3 cores,
it better exploits the caches attached to that core, and confining the instance to a single core reduces interference with other cores.

In Linux, the lstopo command displays a simple diagram showing the relationship between logical processors, physical cores,
and different levels of memory cache. I have a 6-core machine with 12 hyperthreads, and I see that logical processors 0 to 5
are on distinct cores and they are matched with logical processors 6 to 11.

This differs from Windows, where consecutive even/odd logical processor pairs occupy the physical cores.

Here's a souped-up version of startx.sh that sets the processor affinity in Linux so that all the threads of an instance of WFS
running in the directory prpclient-x (with x >= 1) are confined to the pair of logical processors sharing the same core.

#!/bin/bash
instance=$(pwd | sed -re 's/.*-//')
CPUs=$(grep processor /proc/cpuinfo | wc -l)
cores=$(dc -e "$CPUs 2 / p")
CPUa=$(dc -e "$instance $cores % p")
CPUb=$(dc -e "$CPUa $cores + p")
echo running on logical CPUs $CPUa and $CPUb
rm -f nohup.out && nohup taskset -c $CPUa,$CPUb ./prpclient -v >nohup.out 2>&1 &

Some notes about this script.

I realize the same thing could be accomplished with cgroups in Linux, but I haven't figured that out yet.

Serious challengers are probably less concerned about efficiency and will overcommit physical cores to minimize WFS runtime.

The modulo operator % ensures that each instance is assigned to a real core, no matter how large the instance number becomes
while bunkering for a challenge (there's a workunit download limit per instance).

The script runs an instance of WFS in the background (detached from the terminal).
All terminal output is redirected into the file nohup.out in the instance's directory.

You can stop all WFS instances with the command
killall -2 wwwwcl
You must use it twice to stop WFS. Once barely gets its attention.

The progress of all instances can be monitored in one terminal window with the command
tail -f prpclient-*/nohup.out

By confining each WFS instance to a single core with this scheme, the unused cores are available for running non-GPU tasks at
full speed. I haven't benchmarked this script for LLR, whether it's as fast as with hyperthreading disabled, or whether it's as fast as
confining LLR to one logical processor (with HT enabled). I always use one logical processor with LLR.

Joe O
Send message
Joined: 13 Jul 07
Posts: 51
ID: 9819
Credit: 19,694,269
RAC: 0
PSA Jade: Earned 10,000,000 credits (19,692,669)
Message 101003 - Posted: 17 Nov 2016 | 12:25:11 UTC

CPU/GPU utilization: 1.96 (cores), 0.78 (devices)
Looks like I could try 2 instances or see what happens if I tell this instance to use 3 cores instead of 2.

Joe O
Send message
Joined: 13 Jul 07
Posts: 51
ID: 9819
Credit: 19,694,269
RAC: 0
PSA Jade: Earned 10,000,000 credits (19,692,669)
Message 101006 - Posted: 17 Nov 2016 | 12:45:40 UTC - in response to Message 101003.

CPU/GPU utilization: 1.96 (cores), 0.78 (devices)
Looks like I could try 2 instances or see what happens if I tell this instance to use 3 cores instead of 2.

CPU/GPU utilization: 2.33 (cores), 1.00 (devices) is what happens if I use 3 cores
21.07M p/sec

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 101009 - Posted: 17 Nov 2016 | 13:27:03 UTC
Last modified: 17 Nov 2016 | 13:40:56 UTC

It figures, always a bug... I forgot to subtract 1 from the instance number.
taskset -c indexes cores starting at 0.
It's not really a problem, just not what I expected.
Here's the corrected version.

#!/bin/bash
instance=$(pwd | sed -re 's/.*-//')
CPUs=$(grep processor /proc/cpuinfo | wc -l)
cores=$(dc -e "$CPUs 2 / p")
CPUa=$(dc -e "$instance 1 - $cores % p")
CPUb=$(dc -e "$CPUa $cores + p")
echo running on logical CPUs $(dc -e "$CPUa 1 + p") and $(dc -e "$CPUb 1 + p")
rm -f nohup.out && nohup taskset -c $CPUa,$CPUb ./prpclient -v >nohup.out 2>&1 &


Edit: And here is manually starting the bunch of them (for 6 instances) from the directory containing all the instances.

for (( x = 1; x <= 6; x++ )); do pushd prpclient-$x; ./startx.sh $x; popd; done

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 101011 - Posted: 17 Nov 2016 | 13:39:00 UTC - in response to Message 101006.
Last modified: 17 Nov 2016 | 14:29:36 UTC

CPU/GPU utilization: 1.96 (cores), 0.78 (devices)
Looks like I could try 2 instances or see what happens if I tell this instance to use 3 cores instead of 2.

CPU/GPU utilization: 2.33 (cores), 1.00 (devices) is what happens if I use 3 cores
21.07M p/sec

By saying "use 3 cores" do you mean that's what happens when you set
wieferich_threads=3
in wwww.ini ?

The script doesn't do much unless wieferich_threads is at least 2.

Joe O
Send message
Joined: 13 Jul 07
Posts: 51
ID: 9819
Credit: 19,694,269
RAC: 0
PSA Jade: Earned 10,000,000 credits (19,692,669)
Message 101015 - Posted: 17 Nov 2016 | 15:14:31 UTC - in response to Message 101011.


By saying "use 3 cores" do you mean that's what happens when you set
wieferich_threads=3
in wwww.ini ?

Yes, I always had 2 before.

Profile composite
Send message
Joined: 16 Feb 10
Posts: 354
ID: 55391
Credit: 357,044,844
RAC: 305,030
321 LLR Ruby: Earned 2,000,000 credits (3,454,678)Cullen LLR Gold: Earned 500,000 credits (676,922)ESP LLR Amethyst: Earned 1,000,000 credits (1,021,827)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,719,804)PPS LLR Ruby: Earned 2,000,000 credits (3,584,121)PSP LLR Ruby: Earned 2,000,000 credits (3,730,893)SoB LLR Ruby: Earned 2,000,000 credits (3,580,843)SR5 LLR Turquoise: Earned 5,000,000 credits (5,071,131)SGS LLR Ruby: Earned 2,000,000 credits (2,860,720)TRP LLR Ruby: Earned 2,000,000 credits (4,212,414)Woodall LLR Amethyst: Earned 1,000,000 credits (1,291,833)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,228,005)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve Ruby: Earned 2,000,000 credits (2,206,963)PPS Sieve Double Bronze: Earned 100,000,000 credits (155,123,928)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Ruby: Earned 2,000,000 credits (2,439,709)GFN Sapphire: Earned 20,000,000 credits (26,274,348)PSA Double Bronze: Earned 100,000,000 credits (102,758,261)
Message 101041 - Posted: 18 Nov 2016 | 5:43:36 UTC - in response to Message 101015.

Sure, that's threads. You have to control core usage by other methods. When I confine an instance to one core, all the threads are running on that one core.

Post to thread

Message boards : Project Staging Area : optimal WFS settings

[Return to PrimeGrid main page]
Copyright © 2005 - 2017 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 6.57, 5.69, 5.39
Generated 19 Nov 2017 | 17:56:31 UTC