Pentium M - SSE2 or not?

[OmbooHankvald]

I would maybe begin sieving but I've heard that P4s are slow as hell and thus better off doing something else. My laptop (hey, I guess every little bit helps) is a "Pentium M".

Now I wonder if Pentium M can sieve or not. Intel's own website at http://www.intel.com/products/processor/index.htm?iid=HPAGE+header_products_processors& shows both a Pentium M Notebook and a Pentium 4 M one.

EVEREST tells:

CPU Type Mobile Intel Pentium M, 1600 MHz (16 x 100)
CPU Alias Banias
Instruktions Sæt x86, MMX, SSE, SSE2
Oprindelig Hastighed 1600 MHz
Engineering Sample Nej
L1 Kode Cache 32 KB
L1 Data Cache 32 KB
L2 Cache 1 MB (On-Die, ATC, Full-Speed)

CPU Fysisk Information
Type 479 Ball uFCBGA / 478 Pin uFCPGA
Transistorer 77 million
Proces Teknologi 6M, 0.13 um, CMOS, Cu, Low-K
Teknologi 83 mm2
Core Spænding 1.00 - 1.48 V
I/O Spænding 1.00 - 1.48 V
Typisk Forbrug 4.0 - 24.5 W (afhængig af clock hastighed)
Maksimalt Forbrug 4.2 - 31.2 W (afhængig af clock hastighed)

CPU Producent
Producent Intel Corporation
Produkt Information http://www.intel.com/products/browse/processor.htm

Anyone able to figure it out? Should it sieve or not?

[Joe O]

Why don't you benchmark it with both programs?

My guess is that you will do very well.

899000-899500
Start each time with sobstatus.dat containing
the following two lines:

pmax=899080000000
pmin=899000000000

copy/paste the sobstatus.dat for each program to a post in this thread

If you need any help, post here or email me factrange at yahoo dot com.

[vjs]

Actually the Banias core is not bad for sieve.

The Banias is not really a P4 it's more like an athlon64 or P3 with a shorter pipeline.

http://www.geek.com/procspec/intel/banias.htm

Also it's not that P4's are bad in sieve they just do better in prp first or secondpass (i.e. the main client). We have a couple people using P4's in sieve just like people are using athlons and p3's in the main effort, maybe not the most efficient but it's still work done and it makes them happy.

Benchmarks here...

http://www.free-dc.org/forum/showthread.php?s=&threadid=9083

It will probably be faster with the sse2 client than the cmov client, my guess is that you will get something around 500kp/s

[garo]

I would suggest using the SSE2 version. I have the same CPU and I get about 417kp/s with CMOV and 437kp/s with SSE2. OOh! Forgot to mention, that is with the 991-50M dat.

[OmbooHankvald]

Okay I'm not sure I understand what you're saying so I'm writing down what I do here so I don't screw up.

1) I've downloaded the SobDat thing from http://www.aooq73.dsl.pipex.com/sobdat/SobDat_n1M-20M.zip

2) I have proth_sieve SSE2 version

And I'm stuck! Joe O --> What SoBStatus.dat are you talking about? I see no in proth_sieve. I don't understand the last part of your post..

[Joe O]

Originally posted by OmbooHankvald

And I'm stuck! Joe O --> What SoBStatus.dat are you talking about? I see no in proth_sieve. I don't understand the last part of your post..

If you start the program without an SoStatus.dat, it will prompt you for a range to work on and then create the file. The file will look like this
pmax=100150000000000
pmin=100100000000000
pmin=100100010000061 @ 230 kp/s
pmin=100100020000061 @ 207 kp/s
pmin=100100030000081 @ 210 kp/s
pmin=100100040000091 @ 201 kp/s
pmin=100100050000141 @ 225 kp/s
pmin=100100060000141 @ 229 kp/s

If you stop the program, it will use this file to restart near the point where it left off.

When it finishes, it will place
Done as the last line. If you restart with this file, it will again prompt you for a range to work on.

It's all done behind the scenes for you, but you can create the file yourself (using a text editor) with the appropriate pmax and pmin lines so that you are not prompted for a range to work on.
Note that the p values in this file are the full value of the number, whereas the reservation (and prompt I think) is the number divided by 1000000000 (with a G suffix)

100100-100150 Joe_O

[hhh]

The easiest way to start sieving is (IMHO) to go to vjs's site

http://www.teamprimerib.com/vjs/

There you get the proth_sieve_042_all.zip and follow the instructions step by step (!).
Especially important is to put a new range in sobistrator before finishing the last one. (It makes things easier).

Some explanations to the precedent posts: there are two different dat files (which contain the work, somehow), a big one and a small one. The big one is slower, but you get factors for tests with higher n, in the same time. It is more effective, and you are recommended to use this one, please.
Some of the factors you find with this large dat file (991-50M) you will need to submit by email. You can either find information in the forum or ask again.

Please note in your reservation, if you use the large dat file.

Have fun! If you have still questions, don't hesitate.
H.

EDIT: OK, confusion is at its maximum. JoeO posted in the same time; follow his instructions, if you understand them (I don't, but I am a beginner), or restart from the beginning with what I said. H.

[vjs]

One more reply to add to the confusion...

I've also created a sieve setup and instructions.

Http://www.teamprimerib.com/vjs

If you download this one it's all setup for you all you have to do is open sobistraotr enter a next range. (i.e. the one you reserve)

Then run sse2

There is also a txt file which should help. Ask questions people are always willing to answer.

[IronBits]

If you already started work with cmov, can you switch to sse2 in mid-stream without any problems?

[vjs]

Yes absolutely,

Just stop the client then restart using sse2. It will pick-up where it left off.

[OmbooHankvald]

Okay I've made a sobstatus.dat file containing the two lines you mentioned Joe O. Now I'm running with SSE2. Shall I post the contents of the sobstatus.dat file here when it is finished?

[Joe O]

Originally posted by hhh
Some explanations to the precedent posts: there are two different Sob.dat files (which contain the work, somehow), a big one and a small one. The big one is slower, but you get factors for tests with higher n, in the same time. It is more effective, and you are recommended to use this one, please.
Some of the factors you find with this large Sob.dat file (991-50M) you will need to submit by email. You can either find information in the forum or ask again.

Please note in your reservation, if you use the large sob.dat file.

Have fun! If you have still questions, don't hesitate.
H.

EDIT: OK, confusion is at its maximum. JoeO posted in the same time; follow his instructions, if you understand them (I don't, but I am a beginner), or restart from the beginning with what I said. H.

I was explaining the use of the SobStatus.dat file, especially if you want to run benchmarks. This file is the file used by prothsieve to record your progress on a range.
Please do not confuse it with the SoB.dat file which contains the k n pairs that we have not yet found a factor for. Yes, there are two different sob.dat files, 1M<n<20M and 991<n<50M. VJS and I are trying to get everyone to use the larger one. It does 250% more work for only 10% more time. If you use the larger one, after you submit the fact.txt file in the normal way, you compress(zip) the fact.txt and factrange.txt files (optionally the factexc.txt file as well) and send them to factrange at yahoo dot com. If you want the larger sob.dat file, then send an email to that address or look at KeRoberts (http://groups.yahoo.com/group/kerobertsdatfiles?yguid=214241620) or the SierpinskiSieve Group. (http://groups.yahoo.com/group/SierpinskiSieve?yguid=214241620)

[Joe O]

Originally posted by OmbooHankvald
Okay I've made a sobstatus.dat file containing the two lines you mentioned Joe O. Now I'm running with SSE2. Shall I post the contents of the sobstatus.dat file here when it is finished?
Yes, or attach it to the post. This will show how fast your machine is, and how well it sieves, which was your original question. I'm no longer sure which machine you are talking about. If it is an SSE2 capable machine, then you can compare the SSE2 program results with the CMOV program results.

I do hope you used
pmax=899080000000
pmin=899000000000
for your test. this will be a very short run, but long enough to get an idea of the speed of the two programs on your machine.

[OmbooHankvald]

Wow, that was fast! Here are the sobstatus's:

First the CMOV version:
pmax=899080000000
pmin=899000000000pmin=899010000001 @ 350 kp/s
pmin=899020000003 @ 324 kp/s
pmin=899030000023 @ 348 kp/s
pmin=899040000043 @ 349 kp/s
pmin=899050000067 @ 318 kp/s
pmin=899060000081 @ 334 kp/s
pmin=899070000097 @ 346 kp/s
Done

Then the SSE2 version:
pmax=899080000000
pmin=899000000000pmin=899010000001 @ 344 kp/s
pmin=899020000003 @ 334 kp/s
pmin=899030000023 @ 374 kp/s
pmin=899040000043 @ 344 kp/s
pmin=899050000067 @ 375 kp/s
pmin=899060000081 @ 361 kp/s
pmin=899070000097 @ 306 kp/s
Done

Are my laptop suited for sieving? Are these good results?

[Joe O]

CMOV 338 kp/s
SSE2 348 kp/s

Yes, your laptop will do well on sieving. The next thing to do is to reserve a range in this thread. (http://www.free-dc.org/forum/showthread.php?s=&threadid=3501)
Currently the next range starts at 820400 so you would post:

820400-820500 OmbooHankvald


This would be about 3 days(?) of work. You can always come back for more, or reserve a larger block now.
820400-820600 OmbooHankvald
Should take a week.
When you are done you post
820400-820600 OmbooHankvald [complete]
in that thread and submit your fact.txt file
on this page. (http://www.seventeenorbust.com/sieve/)

Note: Your sobstatus.dat file should contain

pmax=820600000000000
pmin=820400000000000


Remember these numbers assume that you make a reservation before anyone else does. Adjust them accordingly if all or part of this range is reserved before you post your reservation.

[Joe O]

Forgot to say that after you submit your fact.txt file, it would be nice, but not necessary to email it and your factrange.txt file to factrange at yahoo dot com.
But, if you are using the 991<n<50M sob.dat file, then this step is necessary and important, for us to collect the factors for n>20M. Sorry about the extra work, but we are working on a smoother solution.

[OmbooHankvald]

Thanks for the answers.
Just in case: Where can I get the large .dat-file? And is it also updated daily?

[vjs]

http://www.teamprimerib.com/vjs

It's the SoB.dat_991-50M_2005..> file

It's not updated daily since we are testing from fixed n values not advancing doublecheck. But the dat is accurate and good to go.

[Joe O]

Daily updating is of value only to the factor folks. It helps them to factor only the k n pairs that do not already have a factor, since it does take a lot of resources to factor a k n pair. It used to be done by using a static sob.dat file and a daily results.txt file that the program would use to eliminate k n pairs factored since the creation of the sob.dat. This is no longer done, because the results.txt file is too large. It takes less time, space, and bandwith to create a new sob.dat file daily. And MikeH graciously offered to do it.

Sieving speed, on the other hand, is not that dependent on the size of the sob.dat file. If there happen to be a few k n pairs in the sob.dat file that we already know a factor for, then so be it. There is a small chance that we would find a second factor for that k n pair, and report it. These are the "duplicates" in Mikeh's stats. When we started the sieve effort(s), it was more important because the sob.dat file(s) was(were) much larger. We were also finding thousands of factors a day, instead of less than a hundred that the primary effort does today. The 991-50M effort does find more per day, drumroll for VJS stats. But it would be possible to use the current 991-50M dat until the next prime is found. Then it pays to eliminate all the k n pairs for the k which yielded a prime.

Here is a graph of sob.dat size changing by date earlier this year.