We came up with a theory that CUDA and GPU parallel computing could be used for future generations of computer network equippment and we were right about it.

Since running a web server and serving hundreds of users at the same time is a highly parallel task, this is what GPUs can do really well. Scanning web trafic, running web servers and many other web server and network-related stuff is what could potentially run really well on GPUs.

This was confirmed by General Manager of CUDA Sanford Russell at our last meeting and how Cisco isn't the only one in the network pack who is looking into it. We will be following the development in this area but overall it would be interesting to see Fermi with 512 cores running on a web server and serving thousands of people with small time slices should not be a problem for GPU. Until that happens we are stuck with quad cores and six cores to do the job.