Parallela FPGA-64 performance compared to GPUs and expensive FPGAs?

Question

This is Parallela:

http://anycpu.org/forum/viewtopic.php?f=13&t=66

It has 64 cores, 1 GB of RAM, runs Linux, Ethernet - everyone screams about it ....

My question is: in terms of performance / features, how does Parallela compare to more expensive FPGAs? Do they just have wider buses / more memory / faster processor clocks / more processors on a chip?

I understand that GPUs are designed for massive parallel simple operations, and processors are better suited for more complex single-threaded computing, so where are the expensive FPGAs and Parallela suitable for this curve?

Parallela launches Linux, but have I always been impressed that FPGAs have their logic superimposed on them by writing verilog or VHDL?

Answer 1

Partial answer: FPGAs, as a rule, do not have ANY processor on the chip (there are exceptions), but if you think about processing by choosing instructions and executing them one by one, you really haven't caught FPGA. If you see how to complete one complete iteration of your inner loop in one clock cycle, you will get there.

There will be tasks where it will be easy, and the FPGA can wipe the floor with any other solution. There will be tasks when this is not possible, and Parallela will be a rival. I do not see a single high-performance solution as a common winner; there are impressive things that are done with GPUs (low power is not one of them!), and multi-core XMOS or Parallela solutions also have their place.

Answer 2

Now only Parallelas are available - 16 cores. They have a Xilinx Zynq 7010 or 7020, which is a dual-core FM 800mhz / 1ghz and 80k FPGA console, which is used to communicate with the Parallela chip. I don't know how many of the FPGAs are available for the game, though.

Answer 3

If Parallelas has 16 cores and suppose each core has a hardware multiplier that runs at 1 GHz, the overall computing power of Parallelas is comparable to FPGA for $ 200, and certainly worse than FPGA for $ 1,000. However, in most applications, mathematical calculations are not the main tasks of the processor; they are processed by the ASIC (either the IP core or the DSP coprocessor inside the main processor), for example, the H.264 codec or WiFi data modules. For applications supported by ASICs, a high-performance processor and associated ASIC are always the best solution. Only if you want to be unique in some part, for example, improve image processing algorithms, you probably want to implement your own signal processing algorithm, and this is where multi-core DSPs, GPUs and high-end FPGAs compete.