Back in March I wrote a blog article about Bitcoin and the parallels it has with HPC – and the emergence of FPGA. I think maybe this is now coming true.
With the recent announcement of Intel’s own FPGA-tuned chips and, particularly with respect to OpenPower [both Xilinx and Altera are designing FPGA accelerators that will be attached to OpenPOWER designs], it does look like FPGAs will see more of the HPC spotlight. After all, FPGAs are now covered by the two biggest players, that is X86 and Power (or rather OpenPower)…
A significant reason for this up surge I believe is that the memory access will be that much quicker, in both Intel and OpenPower [OP] systems, unlike traditional GPUs and even (currently) Phi. Both systems will be able to access system main memory, potentially, which is the advantage over PCIe communications.
OP seems a little more integrated with FPGA, but this is based on the limited information that we have in the public domain.
I read an article on HPCWire recently where Nicole Hemsoth suggests 2015 might be the year of the FPGA. I agree. I think FPGAs have a huge potential for specific workloads, and I think that life sciences projects are ideal for this. For organisations like Genomics England, which is attempting to study 100,000 genomes across a 4-year period, they might well be worth considering.