OpenCL/Cuda suitable for crypto acceleration?
I'm thinking of putting together a box to run pfSense based on a Zotac ITX board with an Atom-330 CPU and 4GB RAM, and some spare CF cards with a corresponding CF-to-SATA converter.
The board I'm looking at IONITX-F-E has both a mini PCI-E slot with a WiFi adapter and a 16x PCI-E slot in addition to an on-board gigabit ethernet NIC. That means if I add a dual gigabit ethernet PCI card, I have plenty of connectivity: a guest WiFi network, a DMZ, a LAN and a WAN port. The dual core CPU should be plenty fast for routing and running a small PBX on top of it.
The only thing that might be a limiting factor in the long run is VPN encryption. On the other hand, there's a rather powerful ION GPU on board as well, which is going to be largely unused in a pfSense setup, unless either the PBX voice codecs or the crypto libraries used for IPSec, SSL, etc. take advantage of it using OpenCL/CUDA, provided that's suitable.
I haven't found much in regards to this subject except for a reference to this: https://www.cs.tcd.ie/~harrisoo/publications/GPU_OCF.pdf found here: http://forums.nvidia.com/index.php?showtopic=85120
If something like this could be made part of the BSD OS and find its way into pfSense, we'd have a way to have truly power efficient high performance routers and VPN gateways. Does anyone know if something like this is being worked on?