Before you start...

[xilman]

Originally Posted by bsquared

But there is a greater chance that they will find something better on their own and our contribution will go unused...

True, but there's also the chance that they will not. Given a big enough start and a large enough computational effort the odds could actually be quite good. Indeed, it's entirely possible that the big boys won't even try because so many cpu-millenia of high-quality computation have already been thrown at the task.


Paul

[bsquared]

Originally Posted by xilman

True, but there's also the chance that they will not. Given a big enough start and a large enough computational effort the odds could actually be quite good. Indeed, it's entirely possible that the big boys won't even try because so many cpu-millenia of high-quality computation have already been thrown at the task.


Paul

So there's sort of a catch-22; there is a chance that we could be relevant if we commit to doing cpu-millenia of computation now, but the impetus to commit to doing that amount of work may not be high enough if there is only a chance that we'll be relevant.

Also, it assumes that the other groups will be using the same tools. I wouldn't see much point in running the same software that other people have been running for thousands of cpu-years, but if I've got uber new whizz-bang software I may want to run it anyway, efforts of others be darned.

Man, I am such a drag.

[KyleAskine]

I wonder if Greg has thoughts on this... I'd love to hear his input.

[ixfd64]

Custom hardware might make an attack on RSA-1024 more feasible: http://en.wikipedia.org/wiki/TWIRL

[bsquared]

Originally Posted by ixfd64

Custom hardware might make an attack on RSA-1024 more feasible: http://en.wikipedia.org/wiki/TWIRL

You have a "few dozen million US dollars" laying around?

[chalsall]

Originally Posted by ixfd64

Custom hardware might make an attack on RSA-1024 more feasible: http://en.wikipedia.org/wiki/TWIRL

It is folly to underestimate your enemy.

GPUs have changed the playing field.

Does anyone really think that everyone buying GPUs are only playing games?

[bsquared]

Originally Posted by chalsall

It is folly to underestimate your enemy.

GPUs have changed the playing field.

GPUs are not good at everything. Solving a 1G square matrix is one of those things. Sieving a 1M square area with a list of auxillary info that may take 16GB just to store is another. Building custom hardware that *is* good at doing these things takes megabucks. The tools that let commodity cpu's do a passable job of it don't exist yet.

[Dubslow]

Who, here or anywhere, would be capable (and willing) of producing the sieving software necessary (while poly select happens)? In turn, who would be willing and capable of producing the LA software (while sieving)? jasonp, for the latter, perhaps?

Or, would it be helpful in some way to start with an "easy" number, a C256, such as the Euclid-Mullin number?

jasonp and xilman seem to think they'd be dedicated to getting this done.

[jasonp]

Ben has wanted to build a lattice sieve but has been extremely busy (plus I suspect a kilobit-ready lattice sieve is not the best beginner project). I've been working on GMP-ECM and have also been extremely busy. You can't manage code like this in 30 minutes a day.

Adding a polyselect binary to NFS@Home would be a neat idea, and I'd be happy to make modest structural changes to the Msieve source in order to make the task easier (tuning up the GPU code for mass appeal is not a modest structural change, maybe jrk could volunteer some time). Polyselect would make an ideal distributed application, it's low-memory and highly compute intensive.

I have ideas for sprucing up the filtering, and the initial stages of the LA. Finishing the matrix in less than a year will definitely need national-level computing resources, and that's the gating factor on the LA side; but if the sieving is done then I suppose the time can be donated in exchange for a whiff of notoriety.

[Dubslow]

Give me three of four years and I might be able to start then on a lattice siever (I'm taking this course this upcoming semester)

First though, I need to figure out how to write fast code.

Edit: I've been meaning to ask, what about CADO-NFS? Since it's still in development, theoretically it would be easier to extend to kilobit stuff than the GGNFS siever (and it'd certainly be easier than starting from scratch).

Aside: Why bother starting a new project at all instead of contributing to existing tools? Is it because GGNFS and Msieve were initially separate projects (and still sort of are)? It seems to me that most sensical thing would be to have everybody work on only one piece of software.

[bsquared]

Originally Posted by jasonp

Ben has wanted to build a lattice sieve but has been extremely busy (plus I suspect a kilobit-ready lattice sieve is not the best beginner project). ... You can't manage code like this in 30 minutes a day.

I literally just logged in to write something similar and saw this .
But jason's right, I don't have the time to make the attempt serious . The folks that wrote the current world class siever are professionals. They are supported by grants, universities, etc., and do/did things like that as part of their career. Still, I hope to someday try.

Originally Posted by jasonp

Adding a polyselect binary to NFS@Home would be a neat idea, and I'd be happy to make modest structural changes to the Msieve source in order to make the task easier (tuning up the GPU code for mass appeal is not a modest structural change, maybe jrk could volunteer some time). Polyselect would make an ideal distributed application, it's low-memory and highly compute intensive.

Great idea, sounds cool!