AMD Processors
Decrease font size
Increase font size
Topic Title: AMD64 Question for FPU type projects
Topic Summary:
Created On: 01/14/2004 05:50 PM
Status: Read Only
Linear : Threading : Single : Branch
<< 1 2 Previous Last unread
Search Topic Search Topic
Topic Tools Topic Tools
View similar topics View similar topics
View topic in raw text format. Print this topic.
 01/16/2004 09:47 PM
User is offline View Users Profile Print this message

Author Icon
Brian128
Senior Member

Posts: 2644
Joined: 11/06/2003

does it matter which one we run? SSE2, Regular, cmov?

and do you have a quick link to SoB.dat to test

-------------------------
 01/16/2004 10:04 PM
User is offline View Users Profile Print this message

Author Icon
Brian128
Senior Member

Posts: 2644
Joined: 11/06/2003

I seriously dont understand this program for some reason. Is this the correct syntax: proth_sieve_sse2 -f [7000 7001] if so then mine finishes in like 3 seconds.. any syntax would be great from you if i did it wrong

-------------------------
 01/16/2004 10:10 PM
User is offline View Users Profile Print this message

Author Icon
b2riesel
Junior Member

Posts: 15
Joined: 01/12/2004

QUOTE (Logan[TeamX),Jan 15 2004, 01:03 PM] Mind if I ask a really silly question that I just don't see the answer to: why? Why crunch all this data? What is the purpose in life for all of this work? If you draw the majority of your code and purpose from the GIMPS project, what makes it different?

Thanks

Logan
Logan,

The purpose of our project is to prove the Riesel Conjecture. Our project is almost identical to the Seventeen or Bust project which is trying to prove the Sierpinski Conjecture. While this may mean little to you..and you can read up on either or both if you like...but the project is trying to prove a conjecture made in 1956 that I think is long overdue on being proven. All it takes is a lot of CPU power and time.

Unlike GIMPS, our project is finite. We have 97k's left to find a prime for. Why 97? Well in 1956 Han's Riesel stated that there exist infinitely many odd integers k such that k.2^n - 1 is composite for every n > 1. That means that there are instances where you can input an odd integer for k..and no matter what power you raise 2 by...whether it's 2, 100, 9898989898989, to infinity..it will never ever ever produce a prime number....ever.

He also said that the lowest k to do this is 509203. So he says that 509203*2^n-1 will never ever find a prime....ever...no matter what value n is. You can try to infinity and it will never be prime....ever.

Thru the decades people have taken all the odd integer k's below k=509203 and found a prime for them. Some were as easy as n=2. The largest so far has been 212893*2^730387 - 1...which is k=212893 times 2 raised to the 730387th power then minus all that by 1. Nice 219874 digit prime. Kind of amazing it took a number that large to find a prime for k=212893. The largest for our project so far is k=261221 n=689422 which we found December 22nd.

Now out of the over 250,000 odd integer k's less than 509203 we are down to just 97 remaining candidates. Just 97. As you can see by the test mentioned above..we can test one k/n pair using LLR in a little over an hour. How many numbers do we have?

Well presently we have 97 values for k. Our n values for each one are from a low of about 550,000 to 20,000,000. Then we take those numbers and run thru a sieve. Are any of you numbers divisible by 2? how about 3?...how about 5? how about 7? and so on dividing by small prime numbers. So we are asking if 417643*2^692487-1 is divisible by small primes...and in our case 8 trillion is a small prime..well not 8 trillion..but a number close to that..since 8 trillion is divisible by 2...but anyway currently the remaining k/n pairs escaping the sieve is about 7.5 million values. We eliminate thousands with the sieve every day and we LLR test the next availabe lowest values sorted by n. Sieving is the most efficient way to eliminate numbers but it will never find a prime...just eliminate those divisible by the numbers we divide by....LLR takes time..so we find a good balance and go to work. When a prime is found...up to 200,000 of those numbers are instantly removed from the database since that k...and all associated n values are then removed.

As you can see...it's a finite project..it has an end..this end may not be for another 5 years and we hit primes rivaling those of GIMPS..but it does have an end....and it will prove the Riesel Conjecture once and for all and then a conjecture made in 1956 can become fact.

If you want to know even more...come visit our irc channel and you will hear me get even more religious sounding for my love of this project.


Lee Stephens
B2
www.rieselsieve.com
 01/16/2004 10:16 PM
User is offline View Users Profile Print this message

Author Icon
b2riesel
Junior Member

Posts: 15
Joined: 01/12/2004

QUOTE (Brian128 @ Jan 16 2004, 06:47 PM) does it matter which one we run? SSE2, Regular, cmov?

and do you have a quick link to SoB.dat to test
Ok...don't use the SoB.dat...use the riesel.dat

The SoB.dat is for the Sierpinski project and only has 11 values for k left...we have 97 values..thus the speed differences are insanely different.

Links for downloads for everything in our project' ">http://www.rieselsieve.com/dload.php

Depending on the OS..but generally just type in ./proth_sieve or in windows just double click it..it will ask start range...end range...then initialize

You want to use the CMOV client...does the 64 have SSE2? If so..try both..the other client is for when you have a CPU that simply won't run the other two...AMD's are all running CMOV at the moment.
 01/16/2004 10:21 PM
User is offline View Users Profile Print this message

Author Icon
Brian128
Senior Member

Posts: 2644
Joined: 11/06/2003

I must be doing something wrong, because it still finishes in like 10 seconds.. and my sobstatus.dat is
pmax=7001
pmin=5339352820176032

and yes the A64 has SSE2, my syntax is proth_sieve_sse2 -s riesel.dat -f [7000 7001]

-------------------------
 01/16/2004 10:28 PM
User is offline View Users Profile Print this message

Author Icon
b2riesel
Junior Member

Posts: 15
Joined: 01/12/2004

QUOTE (Brian128 @ Jan 16 2004, 07:21 PM) I must be doing something wrong, because it still finishes in like 10 seconds.. and my sobstatus.dat is
pmax=7001
pmin=5339352820176032

and yes the A64 has SSE2, my syntax is proth_sieve_sse2 -s riesel.dat -f [7000 7001]
Well if you have an SobStatus.dat then you aren't doing Riesel numbers

You have to have the Riesel.dat in the same directory. Then forget all the command line options. Just type in ./Proth_sieve then Enter...then it will ask you for the start range...type 7000 and Enter...then type 7001..and Enter....bam.. ready to go.

Lee Stephens
B2
www.rieselsieve.com
 01/16/2004 10:51 PM
User is offline View Users Profile Print this message

Author Icon
jes
Senior Member

Posts: 1134
Joined: 10/22/2003

Ok, I'm trying that too on my box, the only thing is I'm using the linux version which I assume isn't using SSE2 since it's compiled for 32bit. Do you have the source available?

-------------------------
The opinions expressed above do not represent those of Advanced Micro Devices or any of their affiliates.
http://www.shellprompt.net
Unix & Oracle Web Hosting Provider powered by AMD Opterons
 01/16/2004 10:59 PM
User is offline View Users Profile Print this message

Author Icon
b2riesel
Junior Member

Posts: 15
Joined: 01/12/2004

QUOTE (jes @ Jan 16 2004, 07:51 PM) Ok, I'm trying that too on my box, the only thing is I'm using the linux version which I assume isn't using SSE2 since it's compiled for 32bit. Do you have the source available?
No..you can contact Mikael Lasson whose email is listed on the page you downloaded for. He's offered to give it before but I've found no need for it so far. It's constantly being tweaked and is currently the fastest multi k sieve I know of. His newer version to be released sometime this weekend is 9% faster reportedly..reported by him to me. The CMOV version should give a great guage of speed though.
 01/16/2004 11:04 PM
User is offline View Users Profile Print this message

Author Icon
jes
Senior Member

Posts: 1134
Joined: 10/22/2003

Well, given the large scale math you're dealing with I would expect to see quite a substantial performance improvement with it running in 64bit versus 32bit. So it'd be interesting to compile from source.

-------------------------
The opinions expressed above do not represent those of Advanced Micro Devices or any of their affiliates.
http://www.shellprompt.net
Unix & Oracle Web Hosting Provider powered by AMD Opterons
 01/16/2004 11:11 PM
User is offline View Users Profile Print this message

Author Icon
b2riesel
Junior Member

Posts: 15
Joined: 01/12/2004

QUOTE (jes @ Jan 16 2004, 08:04 PM) Well, given the large scale math you're dealing with I would expect to see quite a substantial performance improvement with it running in 64bit versus 32bit. So it'd be interesting to compile from source.
Most of all that is generalized down to a point where k*2^n-1 Mod P is done very very quickly....insanely quick. Again..it uses some asm nuggets that are 32bit and another area where we need to look at the AMD64 optimized FFTs and such.
 01/16/2004 11:17 PM
User is offline View Users Profile Print this message

Author Icon
jes
Senior Member

Posts: 1134
Joined: 10/22/2003

Yep....you have to remember that running fast on AMD64 isn't necessarily the same as being optimized for AMD64. There are some great docs on the AMD site on how to get the best out of the 64bit architecture from the programmers perspective. I'd be *very* surprised if an assembly routine optimised for 32bit couldn't be improved on for 64bit (especially as I say when dealing with the large numbers).

-------------------------
The opinions expressed above do not represent those of Advanced Micro Devices or any of their affiliates.
http://www.shellprompt.net
Unix & Oracle Web Hosting Provider powered by AMD Opterons
 01/16/2004 11:25 PM
User is offline View Users Profile Print this message

Author Icon
jes
Senior Member

Posts: 1134
Joined: 10/22/2003

Ok, I've ran it for a bit (dual 244's)....how does this look?

QUOTE
jes@cerberus[4:16am]~/tools/riesel> more RieselStatus.dat
pmax=7001000000000
pmin=7000000000000
pmin=7000010000021 @ 86 kp/s
pmin=7000020000031 @ 86 kp/s
pmin=7000030000063 @ 86 kp/s
pmin=7000040000159 @ 86 kp/s
pmin=7000050000193 @ 86 kp/s
pmin=7000060000237 @ 86 kp/s
pmin=7000070000237 @ 86 kp/s
pmin=7000080000239 @ 86 kp/s
pmin=7000090000249 @ 86 kp/s
pmin=7000100000273 @ 86 kp/s
pmin=7000110000293 @ 86 kp/s
pmin=7000120000333 @ 86 kp/s
pmin=7000130000333 @ 85 kp/s
pmin=7000140000377 @ 86 kp/s
pmin=7000150000379 @ 86 kp/s
pmin=7000160000387 @ 86 kp/s
pmin=7000170000413 @ 86 kp/s

-------------------------
The opinions expressed above do not represent those of Advanced Micro Devices or any of their affiliates.
http://www.shellprompt.net
Unix & Oracle Web Hosting Provider powered by AMD Opterons
 01/16/2004 11:30 PM
User is offline View Users Profile Print this message

Author Icon
b2riesel
Junior Member

Posts: 15
Joined: 01/12/2004

QUOTE (jes @ Jan 16 2004, 08:25 PM) Ok, I've ran it for a bit (dual 244's)....how does this look?

QUOTE
jes@cerberus[4:16am]~/tools/riesel> more RieselStatus.dat
pmax=7001000000000
pmin=7000000000000
pmin=7000010000021 @ 86 kp/s
pmin=7000020000031 @ 86 kp/s
pmin=7000030000063 @ 86 kp/s
pmin=7000040000159 @ 86 kp/s
pmin=7000050000193 @ 86 kp/s
pmin=7000060000237 @ 86 kp/s
pmin=7000070000237 @ 86 kp/s
pmin=7000080000239 @ 86 kp/s
pmin=7000090000249 @ 86 kp/s
pmin=7000100000273 @ 86 kp/s
pmin=7000110000293 @ 86 kp/s
pmin=7000120000333 @ 86 kp/s
pmin=7000130000333 @ 85 kp/s
pmin=7000140000377 @ 86 kp/s
pmin=7000150000379 @ 86 kp/s
pmin=7000160000387 @ 86 kp/s
pmin=7000170000413 @ 86 kp/s
Ouch...I hope that is with two clients going together on the dualie..or something else going on in the background...my 2400+ does about 93-96kp/sec.

Is that with the CMOV client?...o wait..is that linux?..Linux is slower due to gcc not being as optimized as vcc or whatever windows uses.


Hmm...going to need to look at this.
 01/16/2004 11:37 PM
User is offline View Users Profile Print this message

Author Icon
jes
Senior Member

Posts: 1134
Joined: 10/22/2003

Yep that's with the CMOV version on Linux, remember that hasn't got SSE2 compiled into it to.

Hmmmm, well gcc has pretty good support for the AMD64 now so it shouldn't be *that* much slower than VisualC

-------------------------
The opinions expressed above do not represent those of Advanced Micro Devices or any of their affiliates.
http://www.shellprompt.net
Unix & Oracle Web Hosting Provider powered by AMD Opterons
 01/17/2004 12:09 AM
User is offline View Users Profile Print this message

Author Icon
Brian128
Senior Member

Posts: 2644
Joined: 11/06/2003

Statistics:
pmin : 7000000000000
pmax : 7001000000000
Total time: 5766773 ms.
173kp/s.

SSE2

-------------------------
 01/17/2004 12:16 AM
User is offline View Users Profile Print this message

Author Icon
b2riesel
Junior Member

Posts: 15
Joined: 01/12/2004

QUOTE (Brian128 @ Jan 16 2004, 09:09 PM) Statistics:
pmin : 7000000000000
pmax : 7001000000000
Total time: 5766773 ms.
173kp/s.

SSE2
Now that is sweet looking...and on an unoptimized client for the 64 also...
 01/17/2004 12:17 AM
User is offline View Users Profile Print this message

Author Icon
Brian128
Senior Member

Posts: 2644
Joined: 11/06/2003

if you ever get a 64bit version of the code that will work in the 64bit version of XP, let me know and I'll run it again.

-------------------------
 01/17/2004 12:18 AM
User is offline View Users Profile Print this message

Author Icon
jes
Senior Member

Posts: 1134
Joined: 10/22/2003

Brian, you certainly beat my Opty....I think it must have a broken leg or something......

-------------------------
The opinions expressed above do not represent those of Advanced Micro Devices or any of their affiliates.
http://www.shellprompt.net
Unix & Oracle Web Hosting Provider powered by AMD Opterons
 01/17/2004 12:22 AM
User is offline View Users Profile Print this message

Author Icon
Brian128
Senior Member

Posts: 2644
Joined: 11/06/2003

I used the SSE2 version.. I wonder what is wrong. Maybe memory latency really makes a huge difference? or the different in mhz from the 244 (maybe the software doesnt work in SMP) to the 3200+ @ 2.15Ghz is enough for that big of spread.. or the linux client just sucks heh

-------------------------
 01/17/2004 12:12 PM
User is offline View Users Profile Print this message

Author Icon
Logan[TeamX]
Senior Member

Posts: 3185
Joined: 12/07/2003

QUOTE (b2riesel @ Jan 16 2004, 10:10 PM) Logan,

The purpose of our project is to prove the Riesel Conjecture. Our project is almost identical to the Seventeen or Bust project which is trying to prove the Sierpinski Conjecture. While this may mean little to you..and you can read up on either or both if you like...but the project is trying to prove a conjecture made in 1956 that I think is long overdue on being proven. All it takes is a lot of CPU power and time.

Unlike GIMPS, our project is finite. We have 97k's left to find a prime for. Why 97? Well in 1956 Han's Riesel stated that there exist infinitely many odd integers k such that k.2^n - 1 is composite for every n > 1. That means that there are instances where you can input an odd integer for k..and no matter what power you raise 2 by...whether it's 2, 100, 9898989898989, to infinity..it will never ever ever produce a prime number....ever.

He also said that the lowest k to do this is 509203. So he says that 509203*2^n-1 will never ever find a prime....ever...no matter what value n is. You can try to infinity and it will never be prime....ever.

Thru the decades people have taken all the odd integer k's below k=509203 and found a prime for them. Some were as easy as n=2. The largest so far has been 212893*2^730387 - 1...which is k=212893 times 2 raised to the 730387th power then minus all that by 1. Nice 219874 digit prime. Kind of amazing it took a number that large to find a prime for k=212893. The largest for our project so far is k=261221 n=689422 which we found December 22nd.

Now out of the over 250,000 odd integer k's less than 509203 we are down to just 97 remaining candidates. Just 97. As you can see by the test mentioned above..we can test one k/n pair using LLR in a little over an hour. How many numbers do we have?

Well presently we have 97 values for k. Our n values for each one are from a low of about 550,000 to 20,000,000. Then we take those numbers and run thru a sieve. Are any of you numbers divisible by 2? how about 3?...how about 5? how about 7? and so on dividing by small prime numbers. So we are asking if 417643*2^692487-1 is divisible by small primes...and in our case 8 trillion is a small prime..well not 8 trillion..but a number close to that..since 8 trillion is divisible by 2...but anyway currently the remaining k/n pairs escaping the sieve is about 7.5 million values. We eliminate thousands with the sieve every day and we LLR test the next availabe lowest values sorted by n. Sieving is the most efficient way to eliminate numbers but it will never find a prime...just eliminate those divisible by the numbers we divide by....LLR takes time..so we find a good balance and go to work. When a prime is found...up to 200,000 of those numbers are instantly removed from the database since that k...and all associated n values are then removed.

As you can see...it's a finite project..it has an end..this end may not be for another 5 years and we hit primes rivaling those of GIMPS..but it does have an end....and it will prove the Riesel Conjecture once and for all and then a conjecture made in 1956 can become fact.

If you want to know even more...come visit our irc channel and you will hear me get even more religious sounding for my love of this project.


Lee Stephens
B2
www.rieselsieve.com
I thank you for taking the time to read my (albeit short and clipped) question and provide me with a well thought-out answer.

I've visited your site, and things are a tad clearer to me now. I've never been a huge math buff, so perhaps that's why I had a rough time understanding the project before.

Again, a sincere thanks.

Logan
Statistics
112018 users are registered to the AMD Processors forum.
There are currently 0 users logged in.

FuseTalk Hosting Executive Plan v3.2 - © 1999-2014 FuseTalk Inc. All rights reserved.



Contact AMD Terms and Conditions ©2007 Advanced Micro Devices, Inc. Privacy Trademark information