This time it worked:
[code]
gary@herford:~/Desktop/gpu-stuff/llrcuda$ time ./llrCUDA -d -q3*2^4235414-1
Starting Lucas Lehmer Riesel prime test of 3*2^4235414-1
Using real irrational base DWT, FFT length = 262144
V1 = 5 ; Computing U0...done.
3*2^4235414-1 is prime! Time : 5448.741 sec.

real 90m54.120s
user 46m7.570s
sys 24m49.730s
[/code]
It looks like the upper limit is somewhere between n=~4.23M and ~5.08M.

[QUOTE=mdettweiler;252989]I...FFT length = 524288, a = 5
Segmentation fault[/code][/QUOTE]
Trouble with 512k FFT?

Fix Bug.:smile:


3*2^382449+1 is prime! Time : 143.208 sec.
3*2^414840-1 is prime! Time : 139.163 sec.
9999*2^458051+1 is prime! Time : 267.380 sec.
1000065*2^390927-1 is prime! Time : 202.897 sec.
3*2^5082306+1 is prime! Time : 12980.957 sec.

Support check point.

[CODE]#ifdef _WIN32
#define WIN32_LEAN_AND_MEAN
#endif[/CODE]

Windows build of the 0.55 Version again at:

[url]www.psp-project.de/test/llrcuda.rar[/url]

Be warned again. It will only run on some versions of the nvidia drivers.
On others it will at least confuse the video RAM so that a reboot is needed.
So far no total crashes of the system but who knows?
As I can not run it on my system I have no clue about the CPU load.

It will run on drivers, which support CUDA 3.2. All WHQL'd 26x.xx should do.

In principle yes but the combination I am running crashes.
Its a GTX260 with the 266.58 drivers on Windows 7 64Bit.

Win7 x64
i7 980X @ 4 GHz
GTX 470 @ 750/1500/1674
Driver 260.99

One test while CPU under full load:

>llrcuda.exe -d -q"3*2^382449+1"
Starting Proth prime test of 3*2^382449+1
Using complex irrational base DWT, FFT length = 32768, a = 5
3*2^382449+1 is prime! Time : 141.339 sec.. Time per bit: 0.334 ms.

About 75% of a CPU core used on average. GPU load jumps between ~50% and 99%.


And another test while CPU idle:

>llrcuda.exe -d -q"3*2^414840-1"
Starting Lucas Lehmer Riesel prime test of 3*2^414840-1
Using real irrational base DWT, FFT length = 32768
V1 = 3 ; Computing U0...
V1 = 3 ; Computing U0...done.
Starting Lucas-Lehmer loop...
3*2^414840-1, iteration : 10000 / 414840 [2.41%]. Time per iteration : 0.314 ms
[...]
3*2^414840-1, iteration : 410000 / 414840 [98.83%]. Time per iteration : 0.250 ms
3*2^414840-1 is prime! Time : 103.670 sec.

One full CPU core used, at least GPU load at a constant 98%. :smile:

FYI: With Ken's GeneferCUDA build, CPU load is about 20% on average.

Can somebody try if this version runs better on Windows.
I try to set some cuda flags.

[url]www.psp-project.de/test/llrcuda_flag.rar[/url]

Win 7 x64/i5-2500@3.3Ghz, GTX 580, Driver version 263.06.

[CODE]>llrcuda.exe -d -q"3*2^414840-1"
Starting Lucas Lehmer Riesel prime test of 3*2^414840-1
Using real irrational base DWT, FFT length = 32768
V1 = 3 ; Computing U0...
V1 = 3 ; Computing U0...done.
Starting Lucas-Lehmer loop...
3*2^414840-1 is prime! Time : 86.222 sec.[98.83%]. Time per iteration : 0.207 ms.[/CODE]

Wanted to try some Woodall numbers already done on CPUs to have some performance comparation and to see if large numbers needs whole CPU core as well.
[CODE]>llrcuda.exe -d -q"8759318*2^8759318-1"[/CODE]
Well, app allocates 256MB RAM and crashed within one second before any screen output.