You can view the page at How to configure NVIDIA NVENC accelerated encoding on Amazon EC2 (G2, NVENC)
I had the opportunity to test NVENC acceleration out yesterday. I was using the g2.2xlarge and the Amazon Linux AMI 2014.09.1 (HVM) ami. The results were impressive:
Without enabling NVENC in transrate.xml, transrating to five resolutions (from 1080p to 284p) used 77 % of CPU. With NVENC enabled, CPU-load decreased to 20 %:
With regards to this article, I have a few comments:
- Check the driver version using the nvidia-smi command. If it’s version 340 or later, then nothing more needs to be done.
You cannot run nvidia-smi on a vanilla Amazon Linux AMI 2014.09.1 (HVM) ami. It seems nVidia drivers are not present by default.
Install the driver using sh ./NVIDIA-Linux-x86_64-340.24.run. Adjust the command to match the file name. During the installation, answer the following prompts as follows:
The installation will fail if you do not first install gcc:
yum install gcc kernel-devel
After installing gcc you can install the driver and run nvidia-smi.
After logging back in, check the driver version again using the nvidia-smi command. If it’s version 340 or later, then nothing more needs to be done. If the version is earlier than 340, download the latest driver from the NVIDIA website. You must accept the agreement so it can’t be downloaded directly to the EC2 instance. Upload the driver to the EC2 instance.
This is newest driver: