Back from job
 

CUDALucas 1.54 compiled for Win64 4.0/2.0 and 4.1/2.1 and new Makefile for Win64.
All untested.

Hi ,LaurV
Residues is same to MLucas.
Thank you,

Perfect
 

CUDALucas vs Residues: 12:0, another good DC with 1.49/1.50:
[CODE]Processing result: M( 50806499 )C, 0x1d11ecca3b052704, n = 3145728, CUDALucas v1.50 LL test successfully completes double-check of M50806499[/CODE]Switching to DCs with 1.54.

Gone?
 

Hi msft,

at first sight it looks the the checkpoint file does not exist per exponent any more, only cudalucas.ckpt. I'd prefer the former way.
Additionally, could we also have a second backup file, maybe like it was before: c<exponent>, t<exponent>? I needed it once.
I like CL being more compact with 1 single file.
Thanks for your work.

By the way, to the others, -c and -t switches are gone. I assume we get checkpoints hard coded every 10000 iterations.

Hi ,
Ver 1.55
change check point file name.
change print format.
[code]
Iteration 10000 M( 216091 )C, 0x30247786758b8792, n = 524288, CUDALucas v1.55 (2.6 msec/Iter ETA 8:55)
[/code]

I tried the latest binary Brain provided - no matter what, I cannot assign work to second GPU.
I have 2 GTX 480, and I successfully run a lot of apps on 2nd GPU while playing(both those, which support multi-gpu and dont).
I've tried -d 1(since in CUDA, 1st GPU is always 0), -d 2 and -d 1.
No matter what, CUDALucas always uses first GPU.
Is this a bug ?

[QUOTE=Karl M Johnson;290539]I tried the latest binary Brain provided - no matter what, I cannot assign work to second GPU.
I have 2 GTX 480, and I successfully run a lot of apps on 2nd GPU while playing(both those, which support multi-gpu and dont).
I've tried -d 1(since in CUDA, 1st GPU is always 0), -d 2 and -d 1.
No matter what, CUDALucas always uses first GPU.
Is this a bug ?[/QUOTE]
Contrary to mfaktc, in CudaLucas should be no space between the D and the 0 (or 1, 2, etc) and you must use uppercase (-d in lowercase is suppose to do I don't know what trick with the output as explained in the help, thing which never worked or was useful).

So, you have to try something like
[CODE]>CudaLucas.xx.bla.bla.bla -c30000 -t -D0 240007
[/CODE]
For me it worked.

Hi ,Kari M Johnson
[QUOTE=Karl M Johnson;290539]I tried the latest binary Brain provided - no matter what, I cannot assign work to second GPU.
I have 2 GTX 480, and I successfully run a lot of apps on 2nd GPU while playing(both those, which support multi-gpu and dont).
I've tried -d 1(since in CUDA, 1st GPU is always 0), -d 2 and -d 1.
No matter what, CUDALucas always uses first GPU.
Is this a bug ?[/QUOTE]
Can you try?
[code]
$ ./CUDALucas -d 1 216091
[/code]

1.55 Win64 CUDA 4.0/4.1 SM 2.0/2.1 compile, untested.
 

[QUOTE=msft;290524]Hi ,
Ver 1.55
change check point file name.
change print format.
[/QUOTE]
1.55 Win64 CUDA 4.0 / SM 2.0 & CUDA 4.1 SM 2.1 compiles, both untested.

[QUOTE=msft;290549]Hi ,Kari M Johnson

Can you try?
[code]
$ ./CUDALucas -d 1 216091
[/code][/QUOTE]

Haha!:smile:
It works, thanks!
The most important detail I forgot to mention - the device flag doesnt work along with "-r" flag, which is the residue test.
But for a regular exponent, everything's fine.

[QUOTE=msft;290524]
Ver 1.55
[/QUOTE]

I've modified this version as follows:

1) changed time calculation logic to use more precise time functions - now time per iteration calculation precision is better then 0.1 msec

2) SDK samples dependencies are now in included file - no more need to pull SDK samples to compile.

3) Merged in Etan's transpose optimization - it seems it gives some marginal improvement but ms/iter of previous version is not good enough to make conclusions.


BTW Ctrl-C handler lost in 1.52 needs to be brought back :-(