Unified Memory on Jetson TK1

Autonomous Machines Jetson & Embedded Systems Jetson TK1
1

Dear All,

I read from CUDA 6.0 version onwards Unified memory is introduced to reduce development efforts and reducing efforts to transfer data between CPU-GPU…

Working with Jetson TK1 and documentation states physical memory is unified … does it mean i can use all memory as unified memory ?

I want to know size of unified memory available on Jetson TK1 ? How is the speed while using unified memory vs GPU memory ?

2

Most of that I can’t answer. However, I believe the GPU does not have its own memory on TK1, although it has its own memory address. Transferring data from a program to the GPU or back implies shuffling data around on the physical memory address space.

3

Hi,KapilMehta

Not really sure your use case for transferring data between CPU and GPU, but you could refer to below 2 links to get more understanding, and see if could get the helps.

[url]https://devtalk.nvidia.com/default/topic/912470/?comment=4802579[/url]
[url]https://devtalk.nvidia.com/default/topic/898086/?comment=4733920[/url]

Thanks