mapped memory with C1060

Accelerated Computing CUDA CUDA Programming and Performance
1

I am trying to use mapped memory on a C1060. The host machine has 16 gb of ram, and is running RHEL with the 2.6.18 kernel. When I try to allocate page locked memory with malloc and mlock, I have no issue, however when I use cudaHostAlloc() with the cudaHostAllocMapped flag, cudaHostAlloc returns an error. I have run the same code on my workstation which is running ubuntu 10.04 with a quadro 4000, and the code runs without issue.
I have set cudaDeviceMapHost flag. Has anyone else had this or a similar issue? What is the solution?

2

I am trying to use mapped memory on a C1060. The host machine has 16 gb of ram, and is running RHEL with the 2.6.18 kernel. When I try to allocate page locked memory with malloc and mlock, I have no issue, however when I use cudaHostAlloc() with the cudaHostAllocMapped flag, cudaHostAlloc returns an error. I have run the same code on my workstation which is running ubuntu 10.04 with a quadro 4000, and the code runs without issue.
I have set cudaDeviceMapHost flag. Has anyone else had this or a similar issue? What is the solution?

3

Problem fixed. Seems that with a fermi series card, you can change device flags after calling cublasInit() and on the older cards, you have to set the flags first.

4

Problem fixed. Seems that with a fermi series card, you can change device flags after calling cublasInit() and on the older cards, you have to set the flags first.