settings for cudaLucas

[fairsky]

Hi guys,

I was just informed that all of my cudaLucas jobs are invalidated.

How can I have the correct settings for this type of job in cudaLucas?

My GPUs are:
1): GTX 295
2): Tesla C2050
CPU is: Xeon W5590
Mems: 24GB

my settings from CUDALucas.ini:

# You can use this file to customize CUDALucas without having to create a long
# and complex command. I got tired of having to hit the up arrow a bunch of
# times whenever I rebooted, so I created this. You can set most of the command
# line options here; however, if you do use command line options, they will
# override their corresponding value in this file.

CheckpointIterations=10000

WorkFile=worktodo.txt

Polite=0

CheckRoundoffAllIterations=1

SaveAllCheckpoints=0

SaveFolder=savedfiles

Interactive=0

Threads=512

DeviceNumber=2

PrintDeviceInfo=0

FFTLength=0

Thank you for any advices.

[chalsall]

Originally Posted by fairsky

Thank you for any advices.

Understand who you are dealing with.

[fairsky]

Originally Posted by chalsall

Understand who you are dealing with.

Sorry, but what did that mean?

[kracker]

Originally Posted by fairsky

Sorry, but what did that mean?

I don't, either. What exponents are you running and what fft?

[fairsky]

Originally Posted by kracker

I don't, either. What exponents are you running and what fft?

for example

M( 62327431 )C, 0xffffffff80000000, n = 393216, CUDALucas v2.03

from my results.txt under cudaLucas folder.

Actually, I noticed that I got this result pattern

M(XXX),0xffffffff80000000, n=different-value, v2.03

for all 6 cudaLucas jobs.
I thought there might be some problem, but after all i am a human who cannot identify what was going on to these results...

Any comment?

[kracker]

Originally Posted by fairsky

for example

M( 62327431 )C, 0xffffffff80000000, n = 393216, CUDALucas v2.03

from my results.txt under cudaLucas folder.

Actually, I noticed that I got this result pattern
for all 6 cudaLucas jobs.
I thought there might be some problem, but after all i am a human who cannot identify what was going on to these results...

Any comment?

Aha. Your FFT is way too small. Is that pre selected or automatic? it should be around 10 times bigger... lol. How are you executing CUDALucas?

[fairsky]

Originally Posted by kracker

Aha. Your FFT is way too small. Is that pre selected or automatic? it should be around 10 times bigger... lol. How are you executing CUDALucas?

[del]I think it was possibly selected automatically, because of the setting.
I don't know how worked, but it did get through. :(
[/del]

Oh, I set it I think.

For recent jobs, I set the fft length by

-f 327680

then it also worked....

On my Tesla card. Then does this matter? (thread is by default 512)

[kracker]

Originally Posted by fairsky

I think it was possibly selected automatically, because of the setting.

I don't know how it worked, but it did get through. :(

For recent jobs, I set the fft length by

-f 327680

then it also worked....

On my Tesla card. Then does this matter? (thread is by default 512)

-f 327680

THAT is what is wrong. You may want to remove that...

[fairsky]

I knew there would be some errors, and I found the error is around 0.1 not normally i think the 0.01 you expect.

But why my results are always 0xfffffff80000000????
And why 0.1 is not allowed? I can see the error should be less than .25 by default computed from cudaLucas itself according to the first few steps, right?

for example:

iteration = 26 < 1000 && err = 0.320312 >= 0.25,

[fairsky]

Originally Posted by kracker

-f 327680

THAT is what is wrong. You may want to remove that...

But in what sense it is wrong?
Since it is suggested in the setting file that we can set by -f option the fft length.

Otherwise, what size of fft length is reasonable and can guarantee the correct output?

[kracker]

Originally Posted by fairsky

But in what sense it is wrong?
Since it is suggested in the setting file that we can set by -f option the fft length.

Otherwise, what size of fft length is reasonable and can guarantee the correct output?

I guess you could tweak fft manually, but you really don't need to pass -f, just let it do it's own thing.