fft limited to 8e6 elements?

[DWK]

I'm trying to use GPUmat to accelerate some large ffts, using a Quadro 5000 (Compute 2.0). I'm running into the following error:

K>> fft(GPUsingle(rand(2^23,1)));
Error using fft
Maximum number of elements for 1D FFT/IFFT is 8e6. (ERROR code FFT.2)
K>>
When I look at the CUDA zone information, however, it indicates a limit of 128e6 elements.

Is this an inherent limitation or an old ceiling that's never been raised?
Thanks,
Dan

[Gp-you Group]

Hi,

The value 8e6 is hardcoded. We will change it. Thanks for the information.

We can build a new package for you. Let us know which version/architecture do you need, so maybe you can try and check if it works.

[DWK]

Thank you so much for your prompt reply. win64 is what I am using: I'm putting the output of ver below, in case it is useful:

ver

---

MATLAB Version: 7.14.0.739 (R2012a)
MATLAB License Number: 282602
Operating System: Microsoft Windows 7 Version 6.1 (Build 7601: Service Pack 1)
Java Version: Java 1.6.0_17-b04 with Sun Microsystems Inc. Java HotSpot(TM) 64-Bit Server VM mixed mode

---

MATLAB Version 7.14 (R2012a)
Fixed-Point Toolbox Version 3.5 (R2012a)
Global Optimization Toolbox Version 3.2.1 (R2012a)
MATLAB Compiler Version 4.17 (R2012a)
Optimization Toolbox Version 6.2 (R2012a)
Signal Processing Toolbox Version 6.17 (R2012a)

Best regards,
Dan

[Gp-you Group]

The attached file is a new GPUmat build with the corrected max limit. Unfortunately this was not compiled on our production machine and didn't run also all the regression tests we perform in the release. Nevertheless, maybe you can give it a try and let us know if it works.

[DWK]

Gave a try with the new build and got:

GPUstart

GPUmat, Copyright (C) 2012 GP-you Group (http://gp-you.org)

By using GPUmat, you accept all the terms and conditions
specified in the license.txt file.

Please send any suggestion or bug report to gp-you@gp-you.org.

Starting GPU
- GPUmat version: 0.280
- Required CUDA version: 4.2
There is 1 device supporting CUDA
CUDA Driver Version: 4.20
CUDA Runtime Version: 4.20

Device 0: "Quadro 5000"
CUDA Capability Major revision number: 2
CUDA Capability Minor revision number: 0
Total amount of global memory: 2684026880 bytes
- CUDA compute capability 2.0
...done
- Loading module EXAMPLES_CODEOPT
- Loading module EXAMPLES_NUMERICS
-> numerics20.cubin
- Loading module NUMERICS
-> numerics20.cubin
- Loading module RAND

fft(GPUsingle(rand(2^23,1)));
Error using fft
Error in cufftExec.

Same error with:
fft(GPUsingle(rand(2^23,1)));

Thanks,
Dan

[DWK]

Hi,
Just wondering if there has been any progress on this item?

Thanks,
Dan