The prime-crunching on dedicated hardware FAQ

[rogue]

Originally Posted by jasong

There's NO evidence that the "integer transform is slower" belief is true.

[jasonp]

Originally Posted by jasong

I'm not sure how long this "answer" has been here, but it's simply wrong, the AMD based LLR program I mentioned an hour or so ago in another thread uses this method.

There's NO evidence that the "integer transform is slower" belief is true.

Jason, I spent the better part of a year in 2004 building the fastest all-integer large convolution code available, and for Fermat-mod arithmetic it was 15% slower than mlucas for very large numbers. That FAQ has been there since the beginning, because I have code to back it up. Believe me, I wish I was wrong, because I've spent the better part of a decade trying to make all-integer large-number arithmetic go faster than the floating point version.

[starrynte]

Is there any way for PS3 to trial factor, then? Or is this impractical too? (Though too many trial factorers isn't that good either...)

[Uncwilly]

Originally Posted by starrynte

Is there any way for PS3 to trial factor, then? Or is this impractical too? (Though too many trial factorers isn't that good either...)

It could be done. How much effort does someone want to put into writing the code is something else. Also, how practical is a third item.

[petrw1]

With a little help from Gov't funding....

http://www.itworldcanada.com/Pages/D...33c&sub=490495

[lavalamp]

To be honest I'm surprised that billions haven't already been poured into this, not least by the American government.

[rogue]

Originally Posted by lavalamp

To be honest I'm surprised that billions haven't already been poured into this, not least by the American government.

Why would the American government want to do that? Having an grade school education with strong science curricula has been against the will of government for decades. The NCLBA (No Child Left Behind Act) has been more interested in political correctness than promoting academic competition, whether its between students in a school or between schools. Its been a punishment on the schools because some kids (those at the left end of the bell curve) should not even be in the public school system and those on the right end are not given the tools to excel.

[jasonp]

Depending on how powerful you think US intelligence agencies are, it's possible that the chance to implement Schor's factoring algorithm on a top secret quantum computer has already merited billions of (secret) US government dollars...

[lavalamp]

Originally Posted by jasonp

Depending on how powerful you think US intelligence agencies are, it's possible that the chance to implement Schor's factoring algorithm on a top secret quantum computer has already merited billions of (secret) US government dollars...

Well that was the very big reason I was thinking of for the US gov't to get in on it.

However, I think people give gov'ts far too much credit for keeping things secret and just generally being competent. Over here it seems that our lot seem to be able to lose an unencrypted laptop/CD/flash drive with important data on at a rate of 2 or 3 per week.

From what I've seen, businesses that actually need to make money to keep going have their wits about them, but anything state run gets buried in paper and procrastination, including schools. The NHS is probably the biggest example, billions just get poured into a black hole and nothing improves.

[ewmayer]

Originally Posted by jasonp

Jason, I spent the better part of a year in 2004 building the fastest all-integer large convolution code available, and for Fermat-mod arithmetic it was 15% slower than mlucas for very large numbers. That FAQ has been there since the beginning, because I have code to back it up. Believe me, I wish I was wrong, because I've spent the better part of a decade trying to make all-integer large-number arithmetic go faster than the floating point version.

And since then the speed of Mlucas on Intel-style hardware has doubled due to use of SSE2, which (even using SSE3/4/5) sucks at 64-bit integer math, especially multiply.

But hey, what do we know - all we`ve ever done is spent years and years studying the algorithms and the hardware ISAs and writing hundreds of thousands of lines of code. When that code ends up running at measured speed X on hardware Y, that`s not a "fact", that`s like, just the hardware`s opinion.

JasonG, I notice you mentioned an LLR code that uses pure-integer arithmetic - so why don`t you provide us with a head-to-head comparison of that running on the same hardware vs Prime95, for various convolution sizes. In other words, back up your noise with some actual simple-to-obtain "data".

Also, in early September you mentioned that your mystery friend was all better and that we`d "be hearing all about his work" and be suitably in awe in 2-3 months. 4 months later ... tick, tock. Or did I miss the resulting New York Times "Computational revolution!!! ... Lights All Askew In The Heavens/Men Of Science More Or Less Agog Over Results" front page? This is not to disparage your friend ... I have a sneaking suspicion that if he knew the extent and extremity of the bragging-on-his-behalf you`ve been doing, he`d be rather embarrassed.

[jasonp]

Ernst, you ask nothing that hasn't already been asked dozens of times. I only hope someday the inferno in Jason's head will drive him to light a candle instead of cursing the darkness.