wilmott.com

Hi,Does anyone have experience with FPGA's in finance, and could advise on a sensible starting point in terms of kit, software, and relevant reading to try implementing option pricing algorithms and related problems?I've seen there are high level programming abstractions such as OpenCL which might make FPGA's more accessible to programmers rather than hardware specialists, any thoughts on how practical this is today, or if this is still too early to realistically work well?Thanks!

Dop you mean GPU or FPGA ?

FPGA

There's been a talk at the recent / 2014 LLVM Developers' Meeting which you may find interesting:Custom Hardware State-Machines and Datapaths: Using LLVM to Generate FPGA Accelerators http://llvm.org/devmtg/2014-10/#talk19

This looks quite promising http://www.altera.com/support/examples/ ... holes.html with technical document http://www.altera.co.uk/literature/wp/w ... opencl.pdf I wonder how well this scales - i.e. can the OpenCL FPGA compiler really take more complex problems and successfully configure an FPGA, or do you end up having to resort to traditional FPGA design methodologies (e.g. compiler produces inefficient designs etc).

I expect very similar trade-offs to the ones we've observed historically (e.g., for regular CPUs): Yes, optimizing compilers will miss some opportunities compared to, say, handcrafted assembly (think SSE/AVX vectorization, taking CPU-specific cache effects into account) -- but the flip side is increased productivity / decreased time-to-market.BTW, looks like the applications have been spreading into data, too:http://zhehaomao.com/papers/memcached-f ... =6560058// Slides: http://www.hotchips.org/wp-content/uplo ... UTexas.pdf