Known Issue: CUFFT Failure on Tesla (in use) + Fermi

[Old posts from the commercial version of ArrayFire] Discussion of ArrayFire using CUDA or OpenCL.

Moderator: pavanky

Known Issue: CUFFT Failure on Tesla (in use) + Fermi

Postby vishy » Tue Nov 09, 2010 4:00 pm

On machines with a Tesla and a Fermi, gselecting a Fermi and performing FFT-based computations results in a CUFFT error.
Our test system consisted of a Linux 32 machine with a C2070 and C1060, running Jacket 1.5.1. The C1060 was gselected, and the following Jacket code was run on it. This caused an "Internal CUFFT Error".

Code: Select all
A = gdouble(rand(1000,1));
B = A.*A;
C = fft(A);

??? Error using ==> fft
Jacket Error A115: Internal CUFFT error. Please read the FAQ entry at
http://www.accelereyes.com/faq?q=a115 (toolbox/frontend/pfft.cu:245)


This is a known issue related to FFT on cards of different compute capabilities and is reproducible on Jacket 1.5 and higher. We anticipate that the CUDA 3.2 toolkit, released recently by NVIDIA, will fix this issue, and are currently working on getting Jacket built with this toolkit version.

(This issue is also documented in the release notes, section "known issues": http://wiki.accelereyes.com/wiki/index.php/Release_Notes#09.2F16.2F2010_-_v1.5)
--------------------------------------------------
Vish Venugopalakrishnan
Software Engineer (Q/A)
AccelerEyes LLC
vishy.v@accelereyes.com

--------------------------------------------------
Resources:
Getting Started, FAQ, Tips, Syntax
User avatar
vishy
 
Posts: 411
Joined: Thu Apr 16, 2009 11:46 am

Return to [archive-commercial] Programming & Development with ArrayFire

cron