CPU almost redundant?

[M0CZY]

Recently I did a small assignment for Operation Billion Digits (71→72 bits), which took around 30 hours on my Pentium 4.
I was a bit miffed to see that the next result reported was someone who used the GPU on their graphics card to do a similar assignment in 3½ minutes!
I feel quite redundant now, as it appears that a handful of GPU users will be able to do far more work between them than all the CPU users put together.

[firejuggler]

That 'someone' is me and -at the time - I wasn't aware that the trial factoring below 75 were supposed to be done by non-GPU computer.
Sorry about that. I probably shouldn't have bragged about the time it took.

[petrw1]

Originally Posted by M0CZY

Recently I did a small assignment for Operation Billion Digits (71→72 bits), which took around 30 hours on my Pentium 4.
I was a bit miffed to see that the next result reported was someone who used the GPU on their graphics card to do a similar assignment in 3½ minutes!
I feel quite redundant now, as it appears that a handful of GPU users will be able to do far more work between them than all the CPU users put together.

Happens all the time and (assuming Moore is correct) will continue to happen.

My first PC was a PII - 400 Mhz and took 15 months to do one LL test.
A year later I added a PIV 1.7Ghz which could do the same test in less than 1 month ... and like you I felt PC 1 was now redundant ... and now my Quad makes PC 2 seem redundant ... etc etc etc ... and it won't be long until firejuggler's current GPU is redundant.

I think this is why George added stats for the last 365 days to the server reports --- comparing 2011 output to 1996 output (or even 2008) output is meaningless (or even bordering on depressing).

[sonjohan]

Which is also the reason why I'm dropping in ranks, as I haven't changed PC's in almost 3 years.

[lycorn]

Originally Posted by M0CZY

I feel quite redundant now

It´s a matter of choosing the appropriate work for your PC.
P4s are good at factoring above 64 bits, which is currently the lowest level available for nearly all exponents below 1B.
Just choose 64->65 bits assignments and you´ll be OK. There are millions to choose from.
I use a P4 3200 for P-1 work, and still feel quite happy with it, although my i5-750 would perform the same task roughly 3 times faster. This P4 used to be my main LL testing machine a while ago.

[ixfd64]

That doesn't mean you shouldn't contribute to GIMPS. Every little bit helps - heck, there are people who are still crunching with Pentium II's!

[petrw1]

Originally Posted by ixfd64

That doesn't mean you shouldn't contribute to GIMPS. Every little bit helps - heck, there are people who are still crunching with Pentium II's!

And unconfirmed rumors of one of these: http://en.wikipedia.org/wiki/ENIAC

Some specs:

A multiplication of a 10-digit number by a d-digit number (for d up to 10) took d+4 cycles, so a 10- by 10-digit multiplication took 14 cycles, or 2800 microseconds—a rate of 357 per second. If one of the numbers had fewer than 10 digits, the operation was faster. Division and square roots took 13(d+1) cycles, where d is the number of digits in the result (quotient or square root). So a division or square root took up to 143 cycles, or 28,600 microseconds—a rate of 35 per second.

Anyone care to (or have the time to) extrapolate this to the current range of LL tests?

[Christenson]

I´m wondering if the energy cost of running the old computer over a year leads to a point where it is cheaper to get a new computer than to continue running the old one?
One reason to keep the old one is that it gives you a console that can be used for the mundane computing tasks in your life, like spreadsheets and such.

[xilman]

Originally Posted by Christenson

I´m wondering if the energy cost of running the old computer over a year leads to a point where it is cheaper to get a new computer than to continue running the old one?
One reason to keep the old one is that it gives you a console that can be used for the mundane computing tasks in your life, like spreadsheets and such.

I still have a PII-450 doing useful work. It's a monitor stand

Paul

[fivemack]

I used an Acorn A7000 for that until late last year when I got an iMac with integral monitor stand.

[cheesehead]

Originally Posted by petrw1

And unconfirmed rumors of one of these: http://en.wikipedia.org/wiki/ENIAC

Some specs:


Anyone care to (or have the time to) extrapolate this to the current range of LL tests?

The main LL compute loop contains no divisions, so we need not consider that execution time.

If a 10-digit (base-10, I presume) by 10-digit multiplication took 2800 microseconds, then a 20-million-digit by 20-million-digit multiplication (the basic squaring step) would -- assuming sufficient main storage capacity, sufficient variable-length operand addressing and processing capability, and vacuum-tube lifetime to perform this successfully, and execution time linearly proportional to each operand's length -- take about 2 million * 2 million * 2800 microseconds = 4 million * 2800 seconds = 11200 million seconds. At roughly 31.5 million seconds per year, that's over 350 years -- for one squaring iteration.^*

I think the vacuum-tube lifetimes would be the limiting factor.

- - -

^* Fortunately, not only is the mod step free, but no separate carry operation is needed because the operands and result are single variable-length numbers rather than arrays of limited-precision numbers.