llrCUDA

[Honza]

Upgraded nVidia drivers to 266.58 (from 263.09) and llrCUDA (old version 0.55 from 20/02/2011) works fine on Win 7 x64/GTX 580.

46157*2^698207+1 is prime!  Time : 327.452 sec..  Time per bit: 0.466 ms.

But large exponents make app crash.

llrcuda.exe -d -q"19249*2^13018586+1"
Starting Proth prime test of 19249*2^13018586+1
Using complex irrational base DWT, FFT length = 2097152, a = 3
...and crash

[zyzhan]

Originally Posted by Honza

Upgraded nVidia drivers to 266.58 (from 263.09) and llrCUDA (old version 0.55 from 20/02/2011) works fine on Win 7 x64/GTX 580.

46157*2^698207+1 is prime!  Time : 327.452 sec..  Time per bit: 0.466 ms.

But large exponents make app crash.

llrcuda.exe -d -q"19249*2^13018586+1"
Starting Proth prime test of 19249*2^13018586+1
Using complex irrational base DWT, FFT length = 2097152, a = 3
...and crash

Llr.c

#ifdef    WIN32
  giant initgiants = newgiant (1<<26);    // "1<<26" is big enough for ~60M  number, original:"giant initgiants = newgiant (1<<19);"

[msft]

Originally Posted by zyzhan

Llr.c

#ifdef    WIN32
  giant initgiants = newgiant (1<<26);    // "1<<26" is big enough for ~60M  number, original:"giant initgiants = newgiant (1<<19);"

Great work thank you,

[ltd]

@zyzhan: Can you post the sourcecode plus the project file you created.
I want to find out why your code is running even on my system without any problem where my own tests all failed to run.

On my GTX260 it runs with only 2% of CPU usage and 95% GPU usage. Which is brilliant. Driver Version used: 267.24

46157*2^698207+1 is prime! Time : 898.752 sec.. Time per bit: 1.329 ms.

I think this is a good time for this rather old GPU.

[Honza]

Thanks for this build - CPU usage is low, GPU is high and large numbers are not crashing :-)

[zyzhan]

@ltd:I didn't do anything to the code except add some definition to avoid build error. and change the max size of gian at the giant initialization routine like i said before.

win7 x64, vs2008 sp1, cuda 3.2, driver 266.58, GT440

@Honza:u are welcome. but the host memory usage is very high and the speed is slow, only 25ms/iter,while i do proth prime test "19249*2^13018586+1". maybe my GT440 problem.

[msft]

Sorry I can not support now.(Tokyo need save electric power.)
I guess zyzhan's change is OK.

[pschoefer]

I still see one full CPU core used (i7 980X, GTX470, Win7 Prof.). However, I'm still using driver 260.99, maybe an update will fix it.

[ltd]

Thanks for the data. I will check the differences in the evening.
For my build I also only changed the definitions. So I think I messed up something with the CUDA build options. I did not change anything in the options but maybe I did choose the wrong "project type". We will see.

Originally Posted by zyzhan

@ltd:I didn't do anything to the code except add some definition to avoid build error. and change the max size of gian at the giant initialization routine like i said before.

win7 x64, vs2008 sp1, cuda 3.2, driver 266.58, GT440

@Honza:u are welcome. but the host memory usage is very high and the speed is slow, only 25ms/iter,while i do proth prime test "19249*2^13018586+1". maybe my GT440 problem.

[ltd]

Originally Posted by pschoefer

I still see one full CPU core used (i7 980X, GTX470, Win7 Prof.). However, I'm still using driver 260.99, maybe an update will fix it.

Is that all the way through the test or only in the first seconds (or on large numbers in the first few minutes?)
When I test large numbers at first I also see a complete core used but later I see that there is no load on the CPU core anymore. If the CPU load stays high at the time when the first "x.yz% done" message is displayed it might be an option to update your driver as latest then I see the drop in CPU load.

[pschoefer]

Originally Posted by ltd

Is that all the way through the test or only in the first seconds (or on large numbers in the first few minutes?)
When I test large numbers at first I also see a complete core used but later I see that there is no load on the CPU core anymore. If the CPU load stays high at the time when the first "x.yz% done" message is displayed it might be an option to update your driver as latest then I see the drop in CPU load.

CPU load stayed that high.

After update to driver 267.24, CPU load went down to 1/8 of one core. Unfortunately, GPU load is only ~40%, if all CPU cores are under load. With one core idle, GPU load is at 98%.