CUDALucas (a.k.a. MaclucasFFTW/CUDA 2.3/CUFFTW) - Page 183

owftheevil · 2013-11-18, 22:46

Quote:

Originally Posted by flashjh

The new code is compiled and the windows binaries (release/debug) are posted on SourceForge.

@owftheevil: The -memtest functions, but something isn't right with the iterations. For example 56 1000 1 on my 580 says ETA 12181:18:07

I posted a working memtest.zip to sourceforge

EDIT: Please only use 2.05 Beta .exe files for testing the code. It is not ready for production use yet. Thanks!

That does seem a bit slow.

Usage:

Code:

./CUDALucas -memtest k n

where k * 25 MB of memory are tested, n * 10000 iterations are done for each of 5 data types at each of the k positions. So with k = 56, n = 1000 you are reading 75MB and writing 25 MB 2.8 billion times. Only ~39GB/s bandwidth on the reads. I'll take a look.

flashjh · 2013-11-18, 23:21

That same test before only took a few seconds.

Manpowre · 2013-11-19, 08:39

Quote:

Originally Posted by owftheevil

That does seem a bit slow.

Usage:

Code:

./CUDALucas -memtest k n

where k * 25 MB of memory are tested, n * 10000 iterations are done for each of 5 data types at each of the k positions. So with k = 56, n = 1000 you are reading 75MB and writing 25 MB 2.8 billion times. Only ~39GB/s bandwidth on the reads. I'll take a look.

hmm, it is x 10k iterations.. oo.. thats different than from memtest right ?
second parameter was not multiplied with 10k ?

owftheevil · 2013-11-19, 15:35

Looked at the ETA code for memtest last night. Didn't find anything wrong, but changed the formula to smooth out the results. Its working as expected on a 570 and 560 ti. New code up at sourceforge.

Code:

./CUDALucas -memtest 35 10

gives an ETA of just over 4 hours on the 560 ti.

Code:

./CUDALucas -memtest 28 2000

gives an ETA of just over 1200 hours on the 570 while it is simultaneously running stage 1 of CUDAPm1.

owftheevil · 2013-11-19, 15:38

Those of you having the fft too big problem while running the self check, could you please post the *fft.txt files, at least up to the line with fft 32 in it? I need to make sure I understand what the problem is.

flashjh · 2013-11-19, 16:13

Quote:

Originally Posted by owftheevil

gives an ETA of just over 1200 hours on the 570 while it is simultaneously running stage 1 of CUDAPm1.

Maybe I missed some discussion about the memtest -- is it meant to run for 50 days in conjunction with CUDALucas or CUDAPm1?

owftheevil · 2013-11-19, 16:21

Quote:

Originally Posted by flashjh

Maybe I missed some discussion about the memtest -- is it meant to run for 50 days in conjunction with CUDALucas or CUDAPm1?

No, I was just trying to guess at what might have given you such a large ETA.

Could you please try running CUDALucas with

Code:

-memtest 56 1

Add the -d 1 if tyou want it to run on device 1.

flashjh · 2013-11-19, 23:08

Recompiled from r43.

-memtest 56 1:

Code:

 
 Initializing memory test using 1400MB of memory on device 0
Beginning test.
 Position 0, Data Type 0, Iteration 10000, Errors: 0, completed 0.36%, Read 4.73G
B/s, Write 1.58GB/s, ETA 11:59:48)

-memtest 35 1:

Code:

Initializing memory test using 875MB of memory on device 0
Beginning test.
 Position 0, Data Type 0, Iteration 10000, Errors: 0, completed 0.06%, Read 117.1
4GB/s, Write 39.05GB/s, ETA 3:02:15)
Position 0, Data Type 0, Iteration 20000, Errors: 0, completed 0.11%, Read 117.0
8GB/s, Write 39.03GB/s, ETA 3:02:12)
Position 0, Data Type 0, Iteration 30000, Errors: 0, completed 0.17%, Read 117.0
9GB/s, Write 39.03GB/s, ETA 3:02:06)
Position 0, Data Type 0, Iteration 40000, Errors: 0, completed 0.23%, Read 117.0
8GB/s, Write 39.03GB/s, ETA 3:02:00)
Position 0, Data Type 0, Iteration 50000, Errors: 0, completed 0.29%, Read 117.0
7GB/s, Write 39.02GB/s, ETA 3:01:55)

Maybe I'm asking the wrong question -- On the original memtest you wrote, -memtest 56 1 only took a few seconds. Did you re-wrote the code to take 12 hours on purpose? Is the test 'updated' to run the way you think it needs to be written for a proper test?

owftheevil · 2013-11-20, 14:47

Quote:

Originally Posted by flashjh

Recompiled from r43.

-memtest 56 1:

Code:

 
 Initializing memory test using 1400MB of memory on device 0
Beginning test.
 Position 0, Data Type 0, Iteration 10000, Errors: 0, completed 0.36%, Read 4.73G
B/s, Write 1.58GB/s, ETA 11:59:48)

-memtest 35 1:

Code:

Initializing memory test using 875MB of memory on device 0
Beginning test.
 Position 0, Data Type 0, Iteration 10000, Errors: 0, completed 0.06%, Read 117.1
4GB/s, Write 39.05GB/s, ETA 3:02:15)
Position 0, Data Type 0, Iteration 20000, Errors: 0, completed 0.11%, Read 117.0
8GB/s, Write 39.03GB/s, ETA 3:02:12)
Position 0, Data Type 0, Iteration 30000, Errors: 0, completed 0.17%, Read 117.0
9GB/s, Write 39.03GB/s, ETA 3:02:06)
Position 0, Data Type 0, Iteration 40000, Errors: 0, completed 0.23%, Read 117.0
8GB/s, Write 39.03GB/s, ETA 3:02:00)
Position 0, Data Type 0, Iteration 50000, Errors: 0, completed 0.29%, Read 117.0
7GB/s, Write 39.02GB/s, ETA 3:01:55)

Maybe I'm asking the wrong question -- On the original memtest you wrote, -memtest 56 1 only took a few seconds. Did you re-wrote the code to take 12 hours on purpose? Is the test 'updated' to run the way you think it needs to be written for a proper test?

Yes, kind of. Too few iterations, like 1000, will miss errors in marginal cases, so I made sure enough iterations are done on each part of the memory chunk. However, its not supposed to last as long on those settings as it is. Also, somethings wrong with your output. With 1 as the parameter for iterations, it should not be repeating Data Type 0. And for some reason, its not reading or writing very fast with 56 for the size of the memory chunk its testing.

Thanks for posting this. Now I have something to look at.

Edit: How much real time is it taking between screen updates in those two cases?

Edit 2: I have a new version up with a diagnostic line. Could you please try the same thing with the new version when you get a chance?

flashjh · 2013-11-20, 22:49

r46: -memtest 56 1

Code:

C:\CUDA\CuLu\test>CUDALucas_205Betar46 -memtest 56 1
 ------- DEVICE 0 -------
name                GeForce GTX 580
 
Initializing memory test using 1400MB of memory on device 0...
Input: size = 56, iterations = 1
Beginning test.
 Position 0, Data Type 0, Iteration 10000, Errors: 0, completed 0.36%, Read 4.63G
B/s, Write 1.54GB/s, ETA 12:14:50)
Position 0, Data Type 1, Iteration 20000, Errors: 0, completed 0.71%, Read 4.64G
B/s, Write 1.55GB/s, ETA 12:12:05)
Position 0, Data Type 2, Iteration 30000, Errors: 0, completed 1.07%, Read 4.61G
B/s, Write 1.54GB/s, ETA 12:10:50)
Position 0, Data Type 3, Iteration 40000, Errors: 0, completed 1.43%, Read 4.62G
B/s, Write 1.54GB/s, ETA 12:08:32)
Position 0, Data Type 4, Iteration 50000, Errors: 0, completed 1.79%, Read 4.63G
B/s, Write 1.54GB/s, ETA 12:05:44)

Observations:

Before the GPU would stay at 100% usage, now every few seconds it drops down to between 20% to 80% and then goes back to 100%

I timed the last group. CUDALucas says 2:48 elapsed, real time was 2:37.9

-memtest 35 1

Code:

C:\CUDA\CuLu\test>CUDALucas_205Betar46 -memtest 35 1
 ------- DEVICE 0 -------
name                GeForce GTX 580
 
Initializing memory test using 875MB of memory on device 0...
Input: size = 35, iterations = 1
Beginning test.
 Position 0, Data Type 0, Iteration 10000, Errors: 0, completed 0.57%, Read 125.0
3GB/s, Write 41.68GB/s, ETA 16:59)
Position 0, Data Type 1, Iteration 20000, Errors: 0, completed 1.14%, Read 124.9
4GB/s, Write 41.65GB/s, ETA 16:53)
Position 0, Data Type 2, Iteration 30000, Errors: 0, completed 1.71%, Read 124.6
9GB/s, Write 41.56GB/s, ETA 16:48)
Position 0, Data Type 3, Iteration 40000, Errors: 0, completed 2.29%, Read 125.0
6GB/s, Write 41.69GB/s, ETA 16:42)
Position 0, Data Type 4, Iteration 50000, Errors: 0, completed 2.86%, Read 124.7
3GB/s, Write 41.58GB/s, ETA 16:36)
Position 1, Data Type 0, Iteration 60000, Errors: 0, completed 3.43%, Read 124.8
9GB/s, Write 41.63GB/s, ETA 16:31)

Observations:

Usage stays at 100%

CUDALucas Time: 5 sec, Timed: 5.8 sec

owftheevil · 2013-11-21, 14:48

I was able to get into windows last night to run some tests. I'm seeing the same thing you are. On a 570 with 1250MB of memory,

-memtest 41 1

runs normally, from 42 up to 46 its very slow like what you see with 56, at 47 it can't allocate all the memory and throws a cuda error.

On Linux, everything is as expected. Up to 47, it runs full speed with no problems, at 48 it can't allocate the memory.

2013-11-19, 15:35	#2006
owftheevil "Carl Darby" Oct 2012 Spring Mountains, Nevada 100111011₂ Posts	Looked at the ETA code for memtest last night. Didn't find anything wrong, but changed the formula to smooth out the results. Its working as expected on a 570 and 560 ti. New code up at sourceforge. Code: ./CUDALucas -memtest 35 10 gives an ETA of just over 4 hours on the 560 ti. Code: ./CUDALucas -memtest 28 2000 gives an ETA of just over 1200 hours on the 570 while it is simultaneously running stage 1 of CUDAPm1.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Don't DC/LL them with CudaLucas	LaurV	Data	131	2017-05-02 18:41
CUDALucas / cuFFT Performance on CUDA 7 / 7.5 / 8	Brain	GPU Computing	13	2016-02-19 15:53
CUDALucas: which binary to use?	Karl M Johnson	GPU Computing	15	2015-10-13 04:44
settings for cudaLucas	fairsky	GPU Computing	11	2013-11-03 02:08
Trying to run CUDALucas on Windows 8 CP	Rodrigo	GPU Computing	12	2012-03-07 23:20

2013-11-18, 23:21	#2004
flashjh "Jerry" Nov 2011 Vancouver, WA 1,123 Posts	That same test before only took a few seconds.

2013-11-19, 15:38	#2007
owftheevil "Carl Darby" Oct 2012 Spring Mountains, Nevada 13B₁₆ Posts	Those of you having the fft too big problem while running the self check, could you please post the *fft.txt files, at least up to the line with fft 32 in it? I need to make sure I understand what the problem is.

2013-11-21, 14:48	#2013
owftheevil "Carl Darby" Oct 2012 Spring Mountains, Nevada 473₈ Posts	I was able to get into windows last night to run some tests. I'm seeing the same thing you are. On a 570 with 1250MB of memory, -memtest 41 1 runs normally, from 42 up to 46 its very slow like what you see with 56, at 47 it can't allocate all the memory and throws a cuda error. On Linux, everything is as expected. Up to 47, it runs full speed with no problems, at 48 it can't allocate the memory.