What to do with a K6-2?
 

I have been given a 550MHz AMD K6-2 with 64MB memory, but it seems very slow for the projects I am currently working on: my 400MHz P2 is much faster for everything except PSP sieving, where it is about equal.

Does anyone know of a project that will make use of this machine, maybe something with a client that uses the 3DNow! instructions?

You could join [URL=http://www.mersenneforum.org/forumdisplay.php?f=50]Operation Billion Digit[/URL] over in the LMH section of the forum. There are some Pentium Pro's and such working onthe project.

[URL=http://home.earthlink.net/~elevensmooth/Billion.html]English home page for OBD[/URL]

Call me biased :smile: , but when PSP sieving works quite optimally on the K6-2, it seems to be a good candidate.

I think Tony Forbes' MFAC runs quite well on K6-2. He coordinates a search for factors of MM61, see [url]http://www.ltkz.demon.co.uk/ar2/mm61.htm[/url]

Alex

[QUOTE]Call me biased :smile: , but when PSP sieving works quite optimally on the K6-2, it seems to be a good candidate.[/QUOTE]
I will do PSP sieving unless I find something that is better suited. The proth_sieve program has a CMOV optimisation which the P2 can use but the K6-2 can't, I guess that accounts for the difference.

[QUOTE]You could join Operation Billion Digit over in the LMH section of the forum. There are some Pentium Pro's and such working onthe project.[/QUOTE]
That project doesn't really appeal to me, too long term :-)

[QUOTE]I think Tony Forbes' MFAC runs quite well on K6-2. He coordinates a search for factors of MM61, see [url]http://www.ltkz.demon.co.uk/ar2/mm61.htm[/url][/QUOTE]
I would like to try that project, but the link to the linux client is down. Can you point me to another one?

[QUOTE=geoff]I will do PSP sieving unless I find something that is better suited. The proth_sieve program has a CMOV optimisation which the P2 can use but the K6-2 can't, I guess that accounts for the difference.[/QUOTE]

AFAIK, cmov optimizations give you ~5% extra speed. I'd suspect the faster L2 cache of the P2 to have a major impact.

I timed some typical work from a few different projects using a 550MHz K6-2 and a 400MHz P-II. Both machines have 64Mb PC-100 SDRAM and (unused) onboard video, the P-II has 512k L2 cache but I don't know how much the K6-2 has (every program I have tried reports it as 'unknown')

'Ratio' is the P-II time divided by the K6-2 time for the same work.

[CODE]
Project Software Ratio
------------------------------------------------------
LMH (TF below 2^64) mprime 23.9 0.40
ECMNET (ECM on 2+ table) mprime 23.9 0.48
ECMNET (ECM on 2LM table) gmp-ecm 5.0.3 0.99
PSP (sieving) proth_sieve 0.42 0.95
Fermat trial factoring fermat 1.6 1.04
MM31 trial factoring MFAC 2.29 1.35
[/CODE]
gmp-ecm and fermat both use libgmp, I compiled a processor specific version for each machine. The MFAC time is based on the sample MM31 range in the instructions, I'm not sure yet how typical that is.

[QUOTE=geoff]...but I don't know how much the K6-2 has (every program I have tried reports it as 'unknown')...[/QUOTE]
[url]http://www.sandpile.org/impl/k6.htm[/url]

The K6-2's L2 is on the motherboard... I have no idea how to query it...

[QUOTE=Xyzzy]The K6-2's L2 is on the motherboard... I have no idea how to query it...[/QUOTE]
a) Look at the hardware spec table, shown just before the OS starts (at least on my Pentium I system there is shown the cache size in that table); or

b) If you know the motherboard model, you could find its specification on the internet; or

c) Open the case and count the cache chips - there should be written their capacity on them.

[QUOTE=geoff]That project doesn't really appeal to me, too long term :-)[/QUOTE]

Perhaps we should expand to more numbers. I can find out some more and fctor them to 60.

Or perhaps you could do the 64s.

I don't find trial factoring very interesting in itself, the results are too predictable, I need to have some interest in the numbers being factored. The only thing I can see that is interesting about an arbitrary billion digit Mersenne number is the possibility that it might be prime, but there is no way to test for that yet, and probably won't be for a long time. With a number like MM61 it would also interesting to discover that it is composite, so it doesn't matter that a primality test is not yet possible.