best bang/buck hardware for developing/running computer vision (opencv) applications

[lobsteroh]

hi all, i have been developing open source computer vision applications in java and some of the central frameworks were handled by the old qpple quicktime library. because of this i have been mainly stuck with using java 1.6 in 32 bit mode, and there was little incentive to go for modern and speedy hardware to replace my fairly old mac powerbook. in my libraries i have finally managed to shed all reliance on that deprecated and outdated framework and i have replaced it with opencv with java bindings. this combination really rocks and now i am itching to see how far i can push performance. i use elphel 353 open source cameras and am hoping to get close to 60fps in hd mode with synchronous access. my current limits are imposed by the speed of the processing/graphics card and i am hoping to get suggestions on what type of hardware i can get the most performance from. i am happy to continue my transition towards unix/linux if that is where the performance is, i am not likely to go in the direction of windows. main use is with opencv, CUDA for putting bitmapped operations to the graphics card, and java 1.8/eclipse. thanks for any suggestions,