can not bind to texture

[dni]

I managed to build magma rc5 on Ubuntu 11.04 with mkl. But the test I run always produce "can not bind to texture". Is this some kind of indication hardware problem. Do I need to run it on a better graphics card?

testing$ ./testing_sgetrf_gpu
device 0: GeForce 9800 GT, 1375.0 MHz clock, 1023.7 MB memory

Usage:
testing_sgetrf_gpu -M 1024 -N 1024

M N CPU GFlop/s GPU GFlop/s ||PA-LU||/(||A||*N)
============================================================
can not bind to texture
can not bind to texture
can not bind to texture
can not bind to texture
can not bind to texture
can not bind to texture
can not bind to texture
can not bind to texture
960 960 2.16 11.70 1.767797e-01
can not bind to texture
can not bind to texture
can not bind to texture
...

[Stan Tomov]

I think this card is not Fermi and does not support the texture featured used. You can try to set GPU_TARGET to 0 in the make.inc file. Don't forget to do make cleanall and make.

[dni]

Thank you. Changing GPU_TARGET to Tesla did get rid of the "can not bind to texture" errors. What is the difference between Tesla and Fermi? My card "9800GT" is not Fermi but it is not a Tesla either. I would like to run some performance comparisons between CULA and Magma. Do you think this card is acceptable or I need to get a Fermi card to do the testing?

[Stan Tomov]

Fermi is the newest GPU architecture from NVIDIA. Tesla is NVIDIA brand of GPUs targeting high-performance computing (dedicated General Purpose GPUs). A Tesla can use Fermi GPUs for example. In our notations though (in the make.inc) we refer to Tesla as to GPGPUs before Fermi. I realize this is confusing and maybe we have to change this terminology.
Any card that supports CUDA is acceptable for benchmarking and performance comparison. MAGMA supports Fermi and the GPGPUs before that. The difference is that a Fermi like the C2050 is more powerful than the 9800GT and you would get higher performance on the C2050.