Re: Mersenne: Factoring on a P4 - CORRECTION
--- Forwarded message follows --- From: "Brian J. Beesley" <[EMAIL PROTECTED]> To: [EMAIL PROTECTED] Date sent: Fri, 22 Jun 2001 18:46:43 - Subject:Re: Mersenne: Factoring on a P4 Copies to: [EMAIL PROTECTED] On 22 Jun 2001, at 13:12, [EMAIL PROTECTED] wrote: > For some reason, I am at a loss to explain, a v21 P4 1.4 GHz > factors significantely slower that a P3 v20 700MHz. Is there a > reason, and solution, for this? Good question. AFAIK George has done nothing to the factoring code. You will see a big "speed loss" if you compare factoring under 2^64 with factoring over 2^64 on _any_ system - that's simply explained by triple- precision integer arithmetic being much slower than double-precision integer arithmetic. Intel's development for the P4 was very much geared towards making SSE2 work well. Unfortunately this left less room in the silicon for some of the "ordinary" integer stuff on which the factoring code (but not the LL testing code) in Prime95 depends. If memory serves me right, the 32 bit by 32 bit integer multiply instruction was badly hit by this. Consequently the factoring throughput of a P4 would be expected to be considerably less than that of a PIII running at the same clock speed. I would expect that a PIII-700 and a P4-1400 would probably run factoring at about the same speed. Earlier I wrote: For now, those who are lucky enough to have P4 systems are probably best advised to stick to LL testing (and trial factoring - which will not be affected by any inefficiency in the P4 integer instructions), leaving trial factoring to those with slower systems. Slip of the brain, I'm afraid. I meant "LL testing (and P-1 factoring..." Incidentally ECM ought to run pretty well on the P4, though there may be some more optimization to come for the very small run lengths typically used by ECM. Regards Brian Beesley _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: Factoring on a P4
Bradford J. Brown wrote: >For some reason, I am at a loss to explain, a v21 P4 1.4 GHz >factors significantely slower that a P3 v20 700MHz. Is there a >reason, and solution, for this? Hmmm... Good question... AFAIK, the only change George has or is going to make in the factoring code since v19... is to change the Athlon over to use the P2/P3 code path... instead of the 486 code path... Doing such will allow Athlons to trial-factor 2.5-3x faster... There really isn't a whole lot more speed increase that can be gained from the factoring code as a whole, AFAIK... You will also notice a BIG speed decrease with trial-factoring on ANY machine as you move from factoring up to 2^62... to factoring up to 2^64... to factoring above 2^64... This is expected with the extra instructions necessary to handle the larger trial-factor sizes... Eric _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: Proth observations
> Windows NT and Win2000 users should consider changing prime95's priority > to two. There have been reports that idle priority doesn't work as documented > in the Microsoft documentation. I'd be curious about that... I haven't heard anything, but then I haven't looked either. :) As I've said before, the only time I've ever seen an actual program run slower when Prime95/NT was running is when I'm running any sort of video capture, such as NetMeeting. NetMeeting vid conferences just run DOG slow when I have my ntprime going, but if I stop the service, then the video picks up greatly. I figured that perhaps the codec just ran at idle priority also (which would make sense to me anyway), so you have two CPU intensive things competing for resources... Aaron _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: Proth observations
Gordon Bower <[EMAIL PROTECTED]> observes > After seeing a post on this list a few weeks ago I decided to branch out > and try a few ranges from Michael Hartley's page looking for k*2^n-1 > primes. I must say there is a bit of a thrill in actually discovering a > new prime every day I run the program instead of proving two numbers a > month composite. :) > I assumed that one value of k was pretty much the same as any other as far > as execution time and the chance of finding primes. To my surprise this > turned out not to be so: On the P3-500, for "most" 650 about 5 hours for 16000 k=701 it took less than 2 and just over 6 hours, respectively. The > phenomenon is reproducible, doesn't seem to be an artifact of other > programs or reboots or suchlike. Any number theorists care to explain what > is special about k=701 that makes it easy to check for primality? > Fix k = 701. We check that If n == 1 (mod 2) then k*2^n == 1 (mod 3) If n == 0 (mod 4) then k*2^n == 1 (mod 5) If n == 6 (mod 8) then k*2^n == 1 (mod 17) If n == 0 (mod 3) then k*2^n == 1 (mod 7) Therefore k*2^n - 1 can be prime only if n == 2 or 10 (mod 24). We can eliminate more potential values of n using If n == 8 (mod 18) then k*2^n == 1 (mod 19) If n == 18 (mod 20) then k*2^n == 1 (mod 41) If n == 6 (mod 28) then k*2^n == 1 (mod 29) Some congruences are redundant; for example If n == 6 (mod 12) then k*2^n == 1 (mod 13) eliminates nothing new. k = 701 has less such redundancy than the typical k. _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: Donations
Hi all, At 09:15 PM 6/22/2001 +, Russel Brooks wrote: > > I therefore suggest that a pay pal account (or similar) is set up. People > > who wished could easily donate a few dollars to make sure that George has > > available to him the latest architecture. I mean, come on, it wouldn't take > >Sounds good to me. I prefer to keep GIMPs a truly free project. There is a GIMPS member in the Orlando area with an Athlon. IF I get up energy to try some Athlon specific improvements, he has volunteered his machine. Having gained 20+% and a gut feeling that there isn't much more to be gained easily, I don't think I'll be doing that at this time. Thanks for the offers of help! Have fun, George _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Mersenne: OT: P4 latencies
Hi, At 11:45 PM 6/22/2001 +0100, Michael Bell wrote: >But they clearly don't care about it on the P4: > >Command Ticks on P2/P3Ticks on P4 >MOV 1 1 >ADD/SUB 1 1 >ADC/SBB 2 8 >MUL 4 14-18 >SHR/SHL 1 4 Evaluating an architecture based on just latencies can be very misleading. For example the P4 has longish latencies on the SSE2 instructions that prime95 uses (6 for a load, 6 for a mul, 4 for an add). However, there is enough parallelism available that these latencies are almost completely hidden. Other architecture features that may well be more important: Cache structure & memory bandwidth Number of registers (to help programmer expose parallelism) Branch prediction and miss penalties. And many, many more. That said, the Athlon is a better performer for most applications today. The P4 was designed for higher clock speeds and memory bandwidth. Time will tell if Intel can ramp up the CPU speed to offset the Athlon's advantages. Regards, George _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: Factoring on a P4
> BTW there was an "unreasonable" acceleration of trial factoring > between the P5 architecture (Pentium Classic/MMX) and the P6 > architecture (Pentium Pro / PII / PIII / Celeron / Xeon), so you > can't simply assume that Intel doesn't care about integer > performance! But they clearly don't care about it on the P4: Command Ticks on P2/P3Ticks on P4 MOV 1 1 ADD/SUB 1 1 ADC/SBB 2 8 MUL 4 14-18 SHR/SHL 1 4 Michael. _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: Proth observations
Hey Gordon, At 01:42 PM 6/22/2001 -0800, Gordon Bower wrote: >After seeing a post on this list a few weeks ago I decided to branch out >and try a few ranges from Michael Hartley's page looking for k*2^n-1 >primes. >Anyway, a few curious observations I made, which surprised me: >I have 2 computers, a P2-350 and P3-500. The program executes nearly 2 1/2 >times as fast on the latter as on the former with nothing else >running. Apparently the Proth code takes advantage of a lot of P3 >features? You should look into newpgen and prp.exe. These two programs can be used to speed up your search for k*2^n+/-1 primes. >With the same version of prime95 and the same version of proth on each >computer, if you run them both at the same time, under Win2000 proth gets >a higher priority and all the processing power, while under NT4, it's the >other way round, and prime95 has to be stopped or have its >priority reduced in the ini file to not smother proth. Curious. Windows NT and Win2000 users should consider changing prime95's priority to two. There have been reports that idle priority doesn't work as documented in the Microsoft documentation. Regards, George _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Mersenne: Proth observations
After seeing a post on this list a few weeks ago I decided to branch out and try a few ranges from Michael Hartley's page looking for k*2^n-1 primes. I must say there is a bit of a thrill in actually discovering a new prime every day I run the program instead of proving two numbers a month composite. :) Anyway, a few curious observations I made, which surprised me: I have 2 computers, a P2-350 and P3-500. The program executes nearly 2 1/2 times as fast on the latter as on the former with nothing else running. Apparently the Proth code takes advantage of a lot of P3 features? With the same version of prime95 and the same version of proth on each computer, if you run them both at the same time, under Win2000 proth gets a higher priority and all the processing power, while under NT4, it's the other way round, and prime95 has to be stopped or have its priority reduced in the ini file to not smother proth. Curious. (Why run them both at once, you ask? If the computer is going to be on all night anyway, it'd be idle when proth finished a range unless prime95 was ready to take over as soon as proth was done.) I assumed that one value of k was pretty much the same as any other as far as execution time and the chance of finding primes. To my surprise this turned out not to be so: On the P3-500, for "most" 650http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Mersenne: Donations
Lawrence Cairns-Smith wrote: > I therefore suggest that a pay pal account (or similar) is set up. People > who wished could easily donate a few dollars to make sure that George has > available to him the latest architecture. I mean, come on, it wouldn't take Sounds good to me. While we're discussing donations let me suggest something similar. I just received the latest newsletter from the Electronic Frontier Foundation. http://www.eff.org In it they were asking for pc donations. If we could donate a pc we could also request that it (and maybe the rest of their pcs) run Gimps. Since EFF also sponsors a prime related prize this seem like a logical group to support. A mention in their newsletter would also be good advertising for Gimps. Cheers... Russ _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Mersenne: Fw: [PrimeNumbers] AMD vs. Intel Floating Point
- Original Message - From: "Milton Brown" <[EMAIL PROTECTED]> To: <[EMAIL PROTECTED]> Cc: "Milton Brown" <[EMAIL PROTECTED]> Sent: Friday, June 22, 2001 7:36 AM Subject: [PrimeNumbers] AMD vs. Intel Floating Point > The prime number group, might be interested in > these timings. > > Milton L. Brown > - Original Message - > From: "Jens-Peer Kuska" <[EMAIL PROTECTED]> > To: <[EMAIL PROTECTED]> > Sent: Thursday, June 21, 2001 11:20 PM > Subject: [mg29486] Re: AMD vs. Intel Floating Point > > > > Hi, > > > > a) the Mathematica speed comparsion from > > > >http://fampm201.tu-graz.ac.at/karl/timings40.html > > > >is posted regular in this news group > > b) on the www-site of *this* news group > > > >http://smc.vnet.net/mathgroup.html > > > >the second head line is a link to various speed > >comparsions found at > > > >http://smc.vnet.net/mathbench.html > > > > and it is quite natural to assume, that a poster to a > > news-group has visited the newsgroup hompage and > > is able to read and understand the headings on a page > > that begins with: > > > > --- > > >Designed by S. Christensen. > > > > > >MathGroup > > > > > >The Email Group for Mathematica Users > > > > > >Comparison of Mathematica on Various Computers > > > ~ > > > > > >MathGroup is now linked to the moderated newsgroup > > > > > >comp.soft-sys.math.mathematica > > > > > > on the Internet. Contact your local system administrator > > > to find out how to read this new group. > > --- > > > > *and* it is quite natural to assume that a poster to the news group > > has read the group rules (on the same page) one of it say: > > > > >PLEASE SEARCH THE ARCHIVES BEFORE YOU ASK WHAT MIGHT BE A COMMON > QUESTION. > > >See the links above for this. > > > > It must be also sayed, that the Mathematica speed depends in the > > most (symbolic) applications not on the floating point power > > of the CPU. The most actions performed by Mathematica are pointer > > operations with it's internal data structures. I would assume that > > 80-90 % of Mathematica's CPU load are pure interger operations. > > High precision calculations, symbolic operations, operations with > > integers, rationals ... all that don't use the floating point hardware. > > > > It depends shaply on the application how much floating point operations > > are used. But when a Mathematica function has such a huge floating > > point > > load it is always better to write a MathLink program. > > > > Regards > > Jens > > > > > > > > Morfeas79a wrote: > > > > > > Kofi > > > as it is well known the AMD processors up untill the model of K6-3D > > > have serious problems in their floating point operations - this can be > > > observed by running programs with great CPU load like SETI@home, the > > > time for a AMD computer to finish one work unit is about twice as big > > > as this in an Intel computer running on the same MHz. The problem has > > > been solved in later models. Of course all this is not known to > > > Mr.Kuska who thinks that 90% of the questions sent in this newsgroup > > > are of trivial nature or in anycase foolish. > > > > > > Regards > > > Jim > > > > > Unsubscribe by an email to: [EMAIL PROTECTED] > The Prime Pages : http://www.primepages.org > > > > Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/ > > _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: Factoring on a P4
On 22 Jun 2001, at 13:12, [EMAIL PROTECTED] wrote: > For some reason, I am at a loss to explain, a v21 P4 1.4 GHz factors > significantely slower that a P3 v20 700MHz. Is there a reason, and > solution, for this? Good question. AFAIK George has done nothing to the factoring code. You will see a big "speed loss" if you compare factoring under 2^64 with factoring over 2^64 on _any_ system - that's simply explained by triple- precision integer arithmetic being much slower than double-precision integer arithmetic. Intel's development for the P4 was very much geared towards making SSE2 work well. Unfortunately this left less room in the silicon for some of the "ordinary" integer stuff on which the factoring code (but not the LL testing code) in Prime95 depends. If memory serves me right, the 32 bit by 32 bit integer multiply instruction was badly hit by this. Consequently the factoring throughput of a P4 would be expected to be considerably less than that of a PIII running at the same clock speed. I would expect that a PIII-700 and a P4-1400 would probably run factoring at about the same speed. For now, those who are lucky enough to have P4 systems are probably best advised to stick to LL testing (and trial factoring - which will not be affected by any inefficiency in the P4 integer instructions), leaving trial factoring to those with slower systems. Conversely, if you need a system with optimal integer performance, you'd be much better advised to buy a system based on a fast Athlon processor than a system based on a Pentium 4. Athlons running integer code even run much cooler; the FPU turns itself off when it isn't needed, leading to a big drop in current consumption. BTW there was an "unreasonable" acceleration of trial factoring between the P5 architecture (Pentium Classic/MMX) and the P6 architecture (Pentium Pro / PII / PIII / Celeron / Xeon), so you can't simply assume that Intel doesn't care about integer performance! Regards Brian Beesley _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Mersenne: Factoring on a P4
For some reason, I am at a loss to explain, a v21 P4 1.4 GHz factors significantely slower that a P3 v20 700MHz. Is there a reason, and solution, for this? Bradford J. Brown - This message was sent using GSWeb Mail Services. http://www.gasou.edu/gsumail _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: Good news for Pentium 3 and Celeron 2 owners
On Fri, 22 Jun 2001, John R Pierce wrote: >> I feel it is ridiculous that George has to beg/borrow the latest >> architecture in order to optimise Prime95. I also know from being a member >> of the prime search community for the last three years the amount of hard >> work George puts into the project. > >agreed. how come AMD isn't burying George in processors so he can better >optimize for their architecture? Sounds like a good idea to me. I've long been impressed by the amount of work that George puts into this project, and would be very willing to contribute a few bucks so that we could purchase the needed hardware to further the project. The members of Team Anandtech have often contributed money and/or hardware to various things. I believe they supplied the $$$ and hardware for the team's personal proxy server for their distributed.net efforts, and have contributed to at least one "game box" for the team members. Why can't we do something similar? Of course, George would have to find the time to do all of this wonderful new coding...;-) Kel _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: Good news for Pentium 3 and Celeron 2 owners
> I feel it is ridiculous that George has to beg/borrow the latest > architecture in order to optimise Prime95. I also know from being a member > of the prime search community for the last three years the amount of hard > work George puts into the project. agreed. how come AMD isn't burying George in processors so he can better optimize for their architecture? _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Mersenne Digest V1 #863
Mersenne Digest Friday, June 22 2001 Volume 01 : Number 863 -- Date: Wed, 20 Jun 2001 08:06:28 +0200 From: "george de fockert" <[EMAIL PROTECTED]> Subject: Re: Mersenne: Prime95 - V21.1.1 aka v21a - - Original Message - From: "Brian J. Beesley" <[EMAIL PROTECTED]> > It would also be interesting to find out if the PIII prefetch code > benefits the AMD K6 processor. The K6-2 and K6-3 do have an Athlon- > compatible prefetch instruction, though it's using 32-byte cache > lines and there is no prefetch queue i.e. only one prefetch can be > active at any time. Just tried it on my k6-2, needs CpuSupportsPrefetch=1 in local.ini, and results in an illegal instruction exception. So the program probably does not use the AMD 3dnow! prefetch instruction, PREFETCH(W), but one of the intel MMX2 or SIMD prefetch instructions not on the K6-2, PREFETCHNTA orPREFETCHT[123] > BTW what happens to the prefetch instructions for those processors > (like PII) that don't support prefetch? See above for the K6-2 George de Fockert _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers -- Date: Wed, 20 Jun 2001 08:51:11 -0700 From: Eric Hahn <[EMAIL PROTECTED]> Subject: Re: Mersenne: Prime95 - V21.1.1 aka v21a George Woltman wrote: >Some Athlons are seeing a speed increase others are not. The >two that I know are not enjoying a speed increase are running >under Win2K. Maybe there is a bug in the way v21.1 determines >if prefetch is supported. > >For those Athlon owners that are not seeing a speed boost, try setting > CpuSupportsPrefetch=1 >in local.ini This is a possibility. The other possibility is that there might be something in Windows that is interfering. The reason I say this, is that I have a P3 running at 733Mhz, and after a reboot, Prime95 was back to running at the same iteration time before v21 starting running. After stopping Prime95 and adding the line to local.ini, it was back to the increased speed. Interesting... to say the least... Eric P.S. BTW, this P3 is running Win98SE... _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers -- Date: Wed, 20 Jun 2001 16:55:04 -0400 From: George Woltman <[EMAIL PROTECTED]> Subject: Mersenne: More v21 Hi all, Regarding some of the comments on v21: The strange data output when doing an Advanced/Time on 10,000,000 is due to some instrumentation I accidentally left in. I have removed it and maybe it will speed up the program by a very, very, very tiny amount. You can download it at ftp://mersenne.org/gimps/p95v21a.zip I've built a Windows NT Service version and a Linux version. You can download them at ftp://mersenne.org/gimps/p95v21n.zip and ftp://mersenne.org/gimps/p95v21x.zip respectively. They have been through the "robust" QA process of proving M11213 prime. Use at your own risk! Brian Beesley, a big QA volunteer, will undoubtedly download the Linux version and report bugs or a higher level of confidence in mprime's capabilities. Since the prefetch code seems to work on the Athlon, I won't be needing the services of Athlon programmers or Orlando residents just now. Thanks to those that offered. Although, I'm sure we could get a few more percent by using PREFETCHW or by a modified memory layout for the 64-byte cache lines, I'll wait until v21 is finalized before looking at any more changes. Finally, the PPro, PII, and Celeron 1 do not crash when a prefetch instruction is encountered. This indicates to me that Intel planned on putting this instruction in prior to the PIII but ran into technical difficulties or time constraints. These CPUs are affected only by the prefetch instructions using up some cache space and decoder bandwidth. Thus, I'm reluctant to increase prime95's size by a third to support 4 different FFT algorithms: Pentium classic, Pentium Pro, Pentium Pro with prefetching, and SSE2. I haven't made a final decision yet (although I sure wish I'd put the FFT code in a DLL!) Regards, George _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers -- Date: Wed, 20 Jun 2001 15:32:24 -0700 (PDT) From: Francois Gouget <[EMAIL PROTECTED]> Subject: Re: Mersenne: Prime95 - V21.1.1 aka v21a On Tue, 19 Jun 2001, Brian J. Beesley wrote: [...] > Unfortunately I can't contribute timings to this argument as none of > my three Athlon systems can run Prime9
Re: Mersenne: Good news for Pentium 3 and Celeron 2 owners
Can I put a suggestion to the group.. I feel it is ridiculous that George has to beg/borrow the latest architecture in order to optimise Prime95. I also know from being a member of the prime search community for the last three years the amount of hard work George puts into the project. The PIII and Athlon processors have been around now for many months and indeed, taking a quick look at the stats, over 60% of members are using either a PIII, AMD or (unspecified type of) Celeron. Imagine the amount of extra work that could have been done if George would have been able to make a 20% speed increase for all these people say 6 months ago. I therefore suggest that a pay pal account (or similar) is set up. People who wished could easily donate a few dollars to make sure that George has available to him the latest architecture. I mean, come on, it wouldn't take too many 'couple of dollars' to buy the latest system board and processor with each new release of architecture and the benefits to the entire group would be immense. I hope George is not offended by this post and ask for discussion. Regards, Lawrence.. >Are there any Athlon owners that would like to take a crack at >optimizing the v21 source so that Athlon owners can enjoy a speed boost too? >Alternatively, are there any Athlon owners in the Orlando area that might >be willing to let me use their machine sometime in the next month? _ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
