I need to test a large number of consecutive numbers for finding primes which fit on several hundreds of bits.
Most actual implementations are for CUDA, but I don't have a recent nVidia card.

I tried to dial for this implementation. The guy told me he will send it, but I didn't had any answer.

I don't mind if basic computations are done serially, because I'm expecting to perform miller-rabin tests in parallel. I need more speed because, it take 25min for GMP to find a 32 Ko long prime on a last gen i7-core.

I don't mind about the language. I've been looking for C, as well for writing the kernel in Python or PHP.

  • Given that NVidia cards are cheap compared to your labor, why don't you just get one? – Ira Baxter May 1 '14 at 9:55
  • @IraBaxter : My CPU is already an high-end one, so the card need to have a large amount of power to be worthfull (I can use CUDA with my 120nm Geforce). You can use both the CPU and the GPUs for the same task in OpenCL. You only use the GPU with CUDA. Do you say +200€ is cheap?. – user2284570 May 1 '14 at 12:16
  • How much did pay for your high-end machine? What is your time worth/what are you paid? I'm pretty sure compared to those 200€ is cheap. And you are not counting the cost of the time you are spending to find an alternative solution to your problem. An answer to this question looks like it would be of use mostly only to you, which makes answering it for a general audience sort of pointless. (SO calls this "too localized") – Ira Baxter May 1 '14 at 13:42
  • @IraBaxter : I'm not counting the time I spend : I'm on my own. I just find stupid to pay for something and can't use it a for an another purpose. If I was asking for cuda : It would be too localized (OpenCL can still be worthy without any GPU). Actually, the real answer for computing large numbers is only GMP. C‌​UMP is only for peoples who own a vendor specific card. OpenCL is seen as the future/common denominator for CUDA; ATI Stream and Intel Xeon phi by many peoples. All CUDA implementations may run slower than their GMP equivalent : they don't use the CPU. – user2284570 Jun 7 '14 at 20:35
  • @IraBaxter Also because now that I’ve a Nvidia ɢᴘᴜ, I see no ᴄᴜᴍᴘ bignum implementations support division. Something critical to find if a number is a probable prime… – user2284570 Sep 30 '17 at 15:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.