Just got the alert of the new P2 instances available in AWS and did some quick tests and was generally disappointed. Wondering what might be going on behind the scenes.
So, I have updated the NVIDIA drivers to the latest betas for the Tesla K80 and what I am seeing is this:
-
CPU utilization, which is my bottleneck on all my transcoding machines, remains a bottleneck and the P2 did not improve this
-
With 1 transcoding stream, I am seeing 20%, while my production g2.2xlarge is there with 3 times more load.
I was hoping this could solve a scaling problem for me, but it looks like there may not be enough plumbing underneath to take full advantage of the herculean GPU power of the new P2 machines.
Anyone play around with the P2 instances and have any thoughts?
To the Wowza team, are we getting to a place where that may be an alternative?
Thanks in advance,
Bob