I've been on the hunt for some decent devkits that can give me extra boost for my nummer crunching project.
The prices of the big FPGA/DSP Soc and modules is really crazy (anywhere from E4.5K to E12K - they're f...ing nutz) that it makes one bypass that bank account breaking idea..
Looked at 'all' FPGA types, DSP types.. anything that could give me a decent cheapy.
With that in the background I found this on my travels..
https://asteroidsathome.net/boinc/cpu_list.php (https://asteroidsathome.net/boinc/cpu_list.php)
This price performance ratio is definitely in favour of ye olde PC CPU/GPU combination ;)
Forgive the comment if it does not fit what you are doing but if you are not already using a multi thread design, there is performance there if you can split up the number crunching. I don't know what precision you are using but the SSE2 FP instructions are no slouch.