User login

FPGA for scientific computing?

Hello, I work in computational methods in a scientific/engineering field. Recently someone has tried to sell his idea of FPGA to me... basically, use FPGA to build a new computer simulation tool that runs on top of hardware directly. It seems that FPGA is like a low-cost, parallel computer that can speed up my scientific computation (lots of a linear matrix operation to solve Ax=b where A is a matrix and b is a vector).

I wonder... if FPGA is really so good (low-cost compared to parallel computer and a lot faster than a standalone desktop), why I have not seen much application in scientific computation? After some investigation, I found:
(1) FPGA is good for logic operation, but not very capable in floating-point or double-precision calculation.
(2) FPGA may have significant overhead. (But does this mean it fits to computation that consists of many independent, easy-to-parallelize steps?).

Can anyone confirm or advise? Thansk a millions!

It all depends on application

Most of modern FPGA'a, apart from having embedded quite powerful processors (Virtex-4 has PowerPC), have components that work like 18x18 multipliers. Since there are many of those in one FPGA I recon matrix multiplication can be done fast due to extensive parallelism.

Why it is not used? I don't know. Maybe FPGA are in fact widely used, but we are not aware of that?

I am not an expert on these topics, so others' opinions would be in place.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.

Upcoming FPGA Events