Emergency xavier cuda

I mean ,i have a pcb board with two parts. One parts is x86cpu, one part is xavier module.

The two parts connected with pcie.

Now,i want to use xavier as gpu. I want to execut cuda applications on x86 cpu,can i ???

Hi luoxiafeng1990,

Please refer to the following Jetson AGX Xavier PCIe Endpoint Software doc: https://developer.nvidia.com/embedded/dlc/jetson-agx-xavier-series-pcie-endpoint-software

That is to say, i can execute cuda application on host point such as one x86 cpu ?

Now i am trying install cuda on x86cpu .

X86-cpu is like xavier ,has its own ddr,flash,emmc and so on.

The two soc(x86-cpu,xaver) is connected by PCIE .Now i want to execute cuda application on X86-cpu, i want to know, if it possible ?

Please give me a help,ths.

Hi luoxiafeng1990,

I don’t think there will be issue to install and run CUDA application on x86 host as root port.
You can use JetPack SDK to install CUDA on x86 host system.

Really ???

xavier has two parts: one is arm core, one is gpu. I think pcie is the interface of arm core.

If i run cuda on x86 cpu, then cuda instructions go from pcie to xavier’s arm’s core. Am i right???

How does gpu core receive the instructions from x86 cpu???

Maybe i didn’t express myself clearly.

Can you explain how does cuda instructions work on xavier’s arm core ???

Arm core receives one cuda instruction then send to gpu by AXI bus ???

But now my instruction is from x86,then sends to xavier’s arm core by pcie. Will pcie transfer the cuda command to arm-core ??? then arm core send it to gpu ???

Please give me a help…thx

Hi luoxiafeng1990, only the Xavier’s ARM CPU can control it’s GPU. It’s an onboard integrated GPU, unlike the PCIe discrete GPU’s that you typically attach to an x86 CPU. The x86 CPU will not detect the Xavier’s CUDA GPU. You can still attach a separate discrete GPU to the x86 PC.

To run CUDA kernels on the Xavier GPU, you would want your x86 application to send some user-defined messages (e.g. over PCIe or IP network) to a client/server application on the Xavier which would then in turn launch the CUDA operations.