The suggestion was to do an extra level or two of TF [i]before[/i] P-1 is done. You'd look for unreserved exponents in the just-about-to-P-1 range (~53M right now) and take each one up a TF level from where they currently are. Maybe 2 levels if you're ambitious. Then P-1 on that exponent will choose smaller bounds and spend less time on P-1 (because more factoring time has already been spent on the TF portion) and therefore we get more "good" P-1's done.

[QUOTE=James Heinrich;253269]The suggestion was to do an extra level or two of TF [i]before[/i] P-1 is done. You'd look for unreserved exponents in the just-about-to-P-1 range (~53M right now) and take each one up a TF level from where they currently are. Maybe 2 levels if you're ambitious. Then P-1 on that exponent will choose smaller bounds and spend less time on P-1 (because more factoring time has already been spent on the TF portion) and therefore we get more "good" P-1's done.[/QUOTE]

Thank you James! :smile:

Luigi

Hello,

find attached a executable for Windows 64bit.
Built on Windows 7 Professional x86-64 using the CUDA toolkit 3.2.16 and the C-compiler from the Windows SDK 7.0. Just in case it matters: I'm using Nvidia Geforce driver 266.58.

Don't be to rough to me if it doesn't work on your system, this is my first attempt for Windows executables. :wink:

You'll need the CUDA runtime DLL. If it is not on your system you can install the CUDA toolkit or just grab the DLL from somewhere else. I think amphoria included it in some of his executables. For future releases I'll check if I'm allowed to redistribute the DLL.

Oliver

and an executable for Windows 32bit. Same system used for the built.

[QUOTE=TheJudger;253351]You'll need the CUDA runtime DLL. If it is not on your system you can install the CUDA toolkit or just grab the DLL from somewhere else. I think amphoria included it in some of his executables.[/QUOTE]To elaborate, what you need is [i]cudart64_32_16.dll[/i], which is pretty tough to find on its own. You need to install [i]cudatoolkit_3.2.16_win_64.msi[/i] from [url]www.nvidia.com[/url] (106MB download) after which you'll find the DLL (and others) floating in your system somewhere. [i]amphoria[/i]'s previous released included a previous DLL ([i]cudart64_31_9.dll[/i]) which doesn't help here.

But if you don't want to go through the hassle of installing the toolkit, PM/email me your email address and I'll send you the DLL. If you grab it from somewhere else, make sure it's:[code]f696a8cddd96fdbacf899365e7d823be cudart64_32_16.dll 549,480[/code]

if you rename the file "[I]cudart32_31_9.dll[/I]" to "[I]cudart32_32_16.dll[/I]", it work too

[QUOTE=firejuggler;253383]if you rename the file "[I]cudart32_31_9.dll[/I]" to "[I]cudart32_32_16.dll[/I]", it work too[/QUOTE]
Didnt work for me.
But I'm happy with previous mfaktc.

[QUOTE=firejuggler;253383]if you rename the file "[I]cudart32_31_9.dll[/I]" to "[I]cudart32_32_16.dll[/I]", it work too[/QUOTE]I've confirmed this does (apparently) work, although of course it would be "[I]cudart[b]64[/b]_31_9.dll[/I]" to "[I]cudart[b]64[/b]_32_16.dll[/I]" for 64-bit Windows.
You do, however, get a warning in mfatkc:[quote]WARNING: current CUDA runtime version is lower than the CUDA toolkit version used during compile[/quote]The CUDA runtime version is 3.20 with the real cudart64_32_16.dll, and 3.10 if you fake it with the v3.10 DLL (not surprisingly). I'm not sure if this is detrimental to the operation of mfaktc or not?

Cudart shenanigans only work with 32 bit applications. For 64 bit apps, you'll need to use native cudart.

[QUOTE=TheJudger;253351]For future releases I'll check if I'm allowed to redistribute the DLL.[/QUOTE]From the CUDA Toolkit EULA.txt:[QUOTE]2.1.2. Redistribution Rights. Licensee may, transfer, redistribute and sublicense certain files of the Redistributable SOFTWARE, as defined in Attachment A of this Agreement; provided, however Licensee shall only install such files into a private (non-shared) directory location that is used only by Licensee’s product(s).[/QUOTE](Attachment A lists cudart, cufft, cublas, cusparse, curand)

Distribute at will, just don't install it in a shared location on the end-user's system?

How difficult would it be to put in some code (at least for the Windows version) to detect keyboard/mouse activity? What about running processes?
Dreaming for a second, what I'd like to see (configurable in mfkatc.ini):[list][*]if keyboard/mouse active within last <x> seconds, throttle mfaktc to <y>% (0-100)[*]if <program> active, throttle mfaktc to <y>% (0-100)[/list]This would let me run mfaktc a lot more, since I just don't run it most of the time (my system is nigh unusable when running), and I often forget to start it when I leave the computer for an extended amount of time. If it could just run in the background slowly (or paused, if I'm running something specific), and then kick into high gear when I'm gone, that would be fantastic. :smile: