Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Generalized Fermat Prime Search :
Long term GPU influence on GFNs
| Author |
Message |
|---|
|
|
|
Given the fact that GFNs with N>24 (or so) will likely not yield any primes due to the low b value limitations, will it be possible to parallelize a Genefer program that can handle large b values, like Genefer80 currently does?
Obviously, it's going to be awhile before PrimeGrid exhausts the GeneferCUDA search range, but maybe in five years (or so), GeneferCUDA isn't going to be very useful. We'll need a program that can handle much larger b values.
Granted, this is just speculation, but is a 128-bit program likely? How will GPUs be put to use in five years? Just sieving work?
Also, would it be possible to parallelize a CPU Genefer program, to utilize 6+ cores simultaneously in order to crunch just one GFN, six times as fast?
I'm mainly just curious as to how parallelization will be taken advantage of by prime number crunchers in the future. Parallelization seems to be the wave of the future... | |
|
 Michael Goetz Volunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14037
ID: 53948
Credit: 478,615,803
RAC: 374,347
                               
|
Given the fact that GFNs with N>24 (or so) will likely not yield any primes due to the low b value limitations, will it be possible to parallelize a Genefer program that can handle large b values, like Genefer80 currently does?
I don't remember if I actually posted this or not, but OpenCL doesn't just run on GPUs. It runs on CPUs, too. On multiple cores.
I'll leave it to the reader to follow that to its logical conclusion. :)
As for 80 bits, yes, a multicore program could do 80 bit math on each of the cores in parallel, but the problem with 80 bit math is it's significantly slower than 64 bit math because it can't make use of the SIMD instructions such as SSE and AVX. The SIMD registers are only 64 bits long.
However, this may be irrelevant:
Obviously, it's going to be awhile before PrimeGrid exhausts the GeneferCUDA search range, but maybe in five years (or so), GeneferCUDA isn't going to be very useful. We'll need a program that can handle much larger b values.
A top of the head calculation yielded "awhile" == 25 years. Even if your 5 year number is the right one, a lot will change in those years. It's very premature to start thinking of what the next step would be.
Granted, this is just speculation, but is a 128-bit program likely? How will GPUs be put to use in five years? Just sieving work?
We're running a 40 MILLION bit program right now. The problem is the hardware is only 64 bits. :)
If you're asking whether 128-bit HARDWARE might become available, I don't know. There are not many applications that need it. Games don't need it, and games are a big driving force in computer technology. That being said, historically, betting against some technological breakthrough happening is a very bad bet, e.g. "There's a worldwide market for maybe five computers, max" (IBM, circa ~1950), or "Who would ever need more than 64K of memory?" (erroneouslly attributed to Bill Gates). So who knows, maybe 128-bit hardware is already on the drawing boards.
Your question about what will happen once we exhaust the B space is a good one, but it's too early to even guess at the answer.
Also, would it be possible to parallelize a CPU Genefer program, to utilize 6+ cores simultaneously in order to crunch just one GFN, six times as fast?
See answer #1.
I'm mainly just curious as to how parallelization will be taken advantage of by prime number crunchers in the future. Parallelization seems to be the wave of the future...
That's probably a safe bet. While overall throughput might be slightly higher with 4 different LLRs or Genefers running, one on each core, having all the cores working on a single problem cuts down the individual run time. With huge numbers taking REALLY long times to crunch on a CPU, there's definitely an advantage to being able to use a multi-core program.
____________
My lucky number is 75898524288+1 | |
|
|
|
|
Given how fast the technology keeps advancing, I suppose my concerns are a bit premature!
Still, I could see it being the case that 75898^524288+1 is the largest GFP we discover for quite some time... The b value limitations on higher N are pretty low for such large numbers. | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14037
ID: 53948
Credit: 478,615,803
RAC: 374,347
|
Given how fast the technology keeps advancing, I suppose my concerns are a bit premature!
Still, I could see it being the case that 75898^524288+1 is the largest GFP we discover for quite some time... The b value limitations on higher N are pretty low for such large numbers.
As nice as it is to hold the record, I certainly hope that's not the case!
There's only one way to find out. :)
____________
My lucky number is 75898524288+1 | |
|
|
|
|
I'm working on my gold GFN badge...
Unfortunately, I don't have a GPU to work with. | |
|
Dave
Send message
Joined: 13 Feb 12
Posts: 3254
ID: 130544
Credit: 2,441,096,983
RAC: 4,209,384
                      
|
|
Bung in a GT240 or GT430. | |
|
Message boards :
Generalized Fermat Prime Search :
Long term GPU influence on GFNs |
