[SOLVED] Can an FPGA outperform a GPU in Graphics computing?

joshua123033

Honorable
Dec 12, 2018
88
1
10,545
As I have learned from my last post (https://forums.tomshardware.com/thr...g-what-would-the-performance-be-like.3476679/), that FPGAs will not outperform any CPU in serial computing, then how will they perform against GPUs in graphical computing.
I know that FPGAs are made for parallel computing, along with GPUs, so I'm curious to see how it will perform against GPUs. I know that GPUs are made for graphic computing so I just want to see how big the gap between the 2 will be.

And no, this isn't for homework.
 
Solution
Well, both Intel and nvidia both use fpga's in the design process for their processors before it goes to prototype. So I'd say an fpga can indeed be used as a gpu. However, fpga's are also very limited, you can only design a circuit in them of specific sizes, and thats were it falls flat. Intel and nvidia are only building a partial processor, used for specific, finite areas and not as a complete solution. So while it might survive basic serial or parallel computing, having the need to handle both, simultaneously, along with other abilities is beyond its scope.

An fpga is like a 1/4 scale model aircraft in a wind tunnel. Great for testing wind patterns, lift differentials, drag co-efficients etc, but it'll never fly and can't be...
Well, both Intel and nvidia both use fpga's in the design process for their processors before it goes to prototype. So I'd say an fpga can indeed be used as a gpu. However, fpga's are also very limited, you can only design a circuit in them of specific sizes, and thats were it falls flat. Intel and nvidia are only building a partial processor, used for specific, finite areas and not as a complete solution. So while it might survive basic serial or parallel computing, having the need to handle both, simultaneously, along with other abilities is beyond its scope.

An fpga is like a 1/4 scale model aircraft in a wind tunnel. Great for testing wind patterns, lift differentials, drag co-efficients etc, but it'll never fly and can't be piloted or carry passengers or cargo like a full sized, fully powered aircraft can.
 
Solution
Well, both Intel and nvidia both use fpga's in the design process for their processors before it goes to prototype. So I'd say an fpga can indeed be used as a gpu. However, fpga's are also very limited, you can only design a circuit in them of specific sizes, and thats were it falls flat. Intel and nvidia are only building a partial processor, used for specific, finite areas and not as a complete solution. So while it might survive basic serial or parallel computing, having the need to handle both, simultaneously, along with other abilities is beyond its scope.

An fpga is like a 1/4 scale model aircraft in a wind tunnel. Great for testing wind patterns, lift differentials, drag co-efficients etc, but it'll never fly and can't be piloted or carry passengers or cargo like a full sized, fully powered aircraft can.
That was a really through explanation! Also, really great analogy!