[jeffreyrogers]

I'm an FPGA engineer and I doubt they will go mainstream. They work great for prototyping, low-volume production, or products that need flexibility in features, but they are hard to use (unlikely to get better in my opinion) and it's hard to see where they would fit into a compute pipeline given that you need to transfer the data to the FPGA, perform your computation/processing, and then transfer the data back.

That said, they are very cool! And learning to create FPGA designs teaches you a lot about how processors and other low level stuff works.

[cinquemb]

>it's hard to see where they would fit into a compute pipeline given that you need to transfer the data to the FPGA, perform your computation/processing, and then transfer the data back.

I see them going mainstream when brain computer interfaces go mainstream (prob a long way away) since a lot of it (in my experience working in a couple of labs and some related hardware) depends on processing a lot of the data from the sensors, of which most is thrown away due to the sheer volume, and transferring it back and being able to update the filtration matrices easily tailored to sampled data.

[ironman1478]

Fpgas are too expensive, power hungry, and large. We use them for many tasks at my workplace and we are spinning up an ASIC team because using fpgas just doesn't meet our power and size requirements. Also, building asics can be cheaper in the long run if the future of what needs to be done is relatively stable.

[cinquemb]

> Also, building asics can be cheaper in the long run if the future of what needs to be done is relatively stable.

I don't doubt it, yet I found hard to describe the human brain over time, especially across people, as that; at least from a DSP and beamforimg of impedance measurements from the scalp to gauge the relative output of power at variable regions in the brain perspective.