## Other

drummers-lowrise

Message boards : Proth Prime Search : PPS Sv on GTX 745

Author Message
Bur
Volunteer tester

Joined: 25 Feb 20
Posts: 332
ID: 1241833
Credit: 22,611,276
RAC: 4,081

Message 144237 - Posted: 16 Oct 2020 | 15:53:04 UTC

I wanted to try some manual Proth prime sieving on a GTX 745 using ppsieve. The rates were very low though, only 1.4M primes/s, while an i3 manages 20M per core. I was sieving 100K<=n<=4M at P around 100T.

I'm wondering because ppsieve seems to run so fast on GPU on Primegrid. Is the 745 that slow or is something else wrong?

(It would also mean much longer runtime, except that the sieving many k at once lets you speed things up on a GPU.)
Ravi wrote this in another thread, is that the reason? I have only one k in the sieve.
____________
Primes: 1281979 & 12+8+1979 & 1+2+8+1+9+7+9 & 1^2+2^2+8^2+1^2+9^2+7^2+9^2 & 12*8+19*79 & 12^8-1979 & 1281979 + 4 (cousin prime)

Crun-chi
Volunteer tester

Joined: 25 Nov 09
Posts: 3008
ID: 50683
Credit: 59,175,698
RAC: 1,834

Message 144261 - Posted: 17 Oct 2020 | 10:16:13 UTC - in response to Message 144237.

I wanted to try some manual Proth prime sieving on a GTX 745 using ppsieve. The rates were very low though, only 1.4M primes/s, while an i3 manages 20M per core. I was sieving 100K<=n<=4M at P around 100T.

I'm wondering because ppsieve seems to run so fast on GPU on Primegrid. Is the 745 that slow or is something else wrong?

(It would also mean much longer runtime, except that the sieving many k at once lets you speed things up on a GPU.)
Ravi wrote this in another thread, is that the reason? I have only one k in the sieve.

You have only one k to sieve? Then use sr1sieve and sieve under CPU it will be faster for any other option. So dont use ppsieve or sr2sieve, use sr1sieve!

You can cheat ppsieve in one way: take 7-10 k add one k you need and then make N in wide range: and try different ranges of N: you will find one that PPSsieve will start: but I doubt it will be faster then sr1sieve. And also ppsieve on low range of K will use one full cpu core: so again: sr1sieve is faster!
____________
92*10^1439761-1 REPDIGIT PRIME :) :) :)
314187728^131072+1 GENERALIZED FERMAT
31*332^367560+1 CRUS PRIME
Proud member of team Aggie The Pew. Go Aggie!

Bur
Volunteer tester

Joined: 25 Feb 20
Posts: 332
ID: 1241833
Credit: 22,611,276
RAC: 4,081

Message 144277 - Posted: 17 Oct 2020 | 16:31:19 UTC

Ok, thanks. I wasn't aware sr1sieve is faster than s2sieve for one k! I've been using it for weeks (on and off) to get P=100T...

There's also srsieve2 as a successor to both, do you know about that? But I can just try and compare.
____________
Primes: 1281979 & 12+8+1979 & 1+2+8+1+9+7+9 & 1^2+2^2+8^2+1^2+9^2+7^2+9^2 & 12*8+19*79 & 12^8-1979 & 1281979 + 4 (cousin prime)

Crun-chi
Volunteer tester

Joined: 25 Nov 09
Posts: 3008
ID: 50683
Credit: 59,175,698
RAC: 1,834

Message 144280 - Posted: 17 Oct 2020 | 17:49:10 UTC - in response to Message 144277.

Ok, thanks. I wasn't aware sr1sieve is faster than s2sieve for one k! I've been using it for weeks (on and off) to get P=100T...

There's also srsieve2 as a successor to both, do you know about that? But I can just try and compare.

I know about that sieve and it is fast but has low P max limit for now. So sr1sieve is still fastest you have get
____________
92*10^1439761-1 REPDIGIT PRIME :) :) :)
314187728^131072+1 GENERALIZED FERMAT
31*332^367560+1 CRUS PRIME
Proud member of team Aggie The Pew. Go Aggie!

Bur
Volunteer tester

Joined: 25 Feb 20
Posts: 332
ID: 1241833
Credit: 22,611,276
RAC: 4,081

Message 144288 - Posted: 17 Oct 2020 | 20:56:18 UTC

Thanks, but I can't find it for download anywhere. Do you have a link for win binaries?
____________
Primes: 1281979 & 12+8+1979 & 1+2+8+1+9+7+9 & 1^2+2^2+8^2+1^2+9^2+7^2+9^2 & 12*8+19*79 & 12^8-1979 & 1281979 + 4 (cousin prime)

Crun-chi
Volunteer tester

Joined: 25 Nov 09
Posts: 3008
ID: 50683
Credit: 59,175,698
RAC: 1,834

Message 144292 - Posted: 17 Oct 2020 | 21:26:32 UTC - in response to Message 144288.

Thanks, but I can't find it for download anywhere. Do you have a link for win binaries?

Sr1sieve?
____________
92*10^1439761-1 REPDIGIT PRIME :) :) :)
314187728^131072+1 GENERALIZED FERMAT
31*332^367560+1 CRUS PRIME
Proud member of team Aggie The Pew. Go Aggie!

dannyridel
Volunteer tester

Joined: 3 Feb 19
Posts: 791
ID: 1097922
Credit: 3,783,734
RAC: 799

Message 144303 - Posted: 18 Oct 2020 | 1:47:00 UTC - in response to Message 144292.

Thanks, but I can't find it for download anywhere. Do you have a link for win binaries?

Sr1sieve?

It's on mersenneforum, in the software forum.
____________
SHSID Electronics Group
SHSIDElectronicsGroup@outlook.com

GFN-14: 50103906^16384+1
Proth "SoB": 44243*2^440969+1

Bur
Volunteer tester

Joined: 25 Feb 20
Posts: 332
ID: 1241833
Credit: 22,611,276
RAC: 4,081

Message 144307 - Posted: 18 Oct 2020 | 5:52:22 UTC

edit: btw, yes, I meant sr1sieve. I realized I only have srsieve and sr2siseve. Didn't know there was a difference between srsieve and sr1sieve.

Do you mean this thread? I was looking there but only found outdated links and the few working attachments were for sr2sieve.

I also just found this but iunfortunately it's just the sources.
____________
Primes: 1281979 & 12+8+1979 & 1+2+8+1+9+7+9 & 1^2+2^2+8^2+1^2+9^2+7^2+9^2 & 12*8+19*79 & 12^8-1979 & 1281979 + 4 (cousin prime)

KEP

Joined: 10 Aug 05
Posts: 288
ID: 110
Credit: 8,529,584
RAC: 1

Message 144312 - Posted: 18 Oct 2020 | 8:27:32 UTC - in response to Message 144307.

I hope it still works, but else send Rebirther a PM at mersenneforum, he maintains the boinc confederation software part.
____________

Bur
Volunteer tester

Joined: 25 Feb 20
Posts: 332
ID: 1241833
Credit: 22,611,276
RAC: 4,081

Message 144334 - Posted: 18 Oct 2020 | 15:44:15 UTC - in response to Message 144312.

Thanks, but that's just the site I mentioned that only contains sources...

At mersenneforum a while back, rogue wrote this:

If k is so large that you cannot build Legendre tables, then srsieve2 is probably faster. The Legendre logic fully coded and tested in srsieve2.
I tried srsieve2 with 70E12<p<71E12 and only got 35,000 p/s. That's so slow it seems to be some specific problem? With sr2sieve I get 10,000,000 p/s.

I'd still like to do a comparison with sr1sieve, so if anyone has binaries, please let me know. :D
____________
Primes: 1281979 & 12+8+1979 & 1+2+8+1+9+7+9 & 1^2+2^2+8^2+1^2+9^2+7^2+9^2 & 12*8+19*79 & 12^8-1979 & 1281979 + 4 (cousin prime)

Crun-chi
Volunteer tester

Joined: 25 Nov 09
Posts: 3008
ID: 50683
Credit: 59,175,698
RAC: 1,834

Message 144335 - Posted: 18 Oct 2020 | 16:02:03 UTC - in response to Message 144334.

But , yes in that archive is source but also exe file for windows, look better :)

____________
92*10^1439761-1 REPDIGIT PRIME :) :) :)
314187728^131072+1 GENERALIZED FERMAT
31*332^367560+1 CRUS PRIME
Proud member of team Aggie The Pew. Go Aggie!

Bur
Volunteer tester

Joined: 25 Feb 20
Posts: 332
ID: 1241833
Credit: 22,611,276
RAC: 4,081

Message 144342 - Posted: 18 Oct 2020 | 20:44:35 UTC - in response to Message 144335.

You're right, embarassingly enough I thought I carefully checked the archive.

Haha, anyway, sr1sieve is indeed faster by about 10%. 11.7E6 vs 10.6E6 p/s.

Thanks for the help!
____________
Primes: 1281979 & 12+8+1979 & 1+2+8+1+9+7+9 & 1^2+2^2+8^2+1^2+9^2+7^2+9^2 & 12*8+19*79 & 12^8-1979 & 1281979 + 4 (cousin prime)

Message boards : Proth Prime Search : PPS Sv on GTX 745