How to Recover Lost Hardware Drivers

The NPU hardware accelerator drivers can be removed from the system. If this happens, the following scrips will recover them

If you notice a severe slowdown in the time it takes to train a NIML model, the NPU hardware drivers may have been removed from the system. If this happens, the NIML NPU will only operate as a software only NPU, even if the parameters are with the required specifications for hardware acceleration.

Check if Drivers are installed

To check if the driver are installed you can run a query in the /dev folder:

ll /dev/xdma*  # Note the first characters are lower case LL

If the result is empty then the drivers will need to be reinstalled as follows:

1. Run the Setup Scripts

source ~/aws-fpga/sdk_setup.sh

FPGA set up scrips

2. Load the FPGA Image

sudo fpga-load-local-image -S 0 -I agfi-0a30dc5b4f9b59e7d

FPGA load image

3. Verify the Correct FPGA Image ID is Installed 

 The image ID should start with 'agfi'

sudo fpga-describe-local-image -S 0 -H

FPGA agfi confirm

4. Install the XDMA Drivers

cd aws-fpga/sdk/linux_kernel_drivers/xdma

sudo make install

5. Verify Drivers are Installed

ll /dev/xdma*       # Again, note the lower case LL

FPGA XDMA drivers