Which FPGA with OpenCL choose for a start? Main goal is computing power.

[Marcin]

Hi all,

I want to start with FPGA. I'm interested also in OpenCL so I would like to have something that will support it. Currently I'm learning AI but I want to find something not too expensive but still giving maximum learning possibilities (not AI only). However priority for now is FPGA computing power.

Here is first problem: How much performance I can expect from Cyclone V FPGA? Will my OpenCL code will be faster on it then on my RTX 2080 TI? If I won't find anything close to that performance in Cyclone V FPGA, then I want something that will be good for general FPGA learning. For example DE10-Nano for low price and easy integration with Arduino.

If performance of FPGA is as good as I heard then what should I consider to get maximum computing power? CPU, RAM and number of FPGA Logic Elements?

How FPGA communicates with CPU? I see that DE10 uses USB interface which seems to be quite slow - I'm not sure how it will impact speed due to low data transfer between CPU and FPGA. Or there is something that I miss here?

[MEIYAN_L_Intel]

Hi, 

Firstly, you may need to check whether the BSP is provided if you would like to used OpenCL in you application.

Question: "How much performance I can expect from Cyclone V FPGA? Will my OpenCL code will be faster on it then on my RTX 2080 TI?"

Answer: You will need to port over your Open CL code directly to Cyclone V to check on performance testing. The reason is performance are design dependant, some design can get better performance while some design will be the other way. You will have to look into the optimziation tecnic as well for the open CL for standard edition of Quartus Prime.

Question: If performance of FPGA is as good as I heard then what should I consider to get maximum computing power? CPU, RAM and number of FPGA Logic Elements?

Answer: As mention in the above, you will have to port over your code to test on this.

According to https://software.intel.com/en-us/node/712683, you may connect to DE10-Nano with two methods which are Ethernet over USB and Serial Connection. For the data transfer speed, you may need to contact Terasic for information on how they use the IP for the USB. 

Thanks

[Marcin]

Great thx for reply! And how number of logic elements impact my work with fpga? For example DE10-Nano has 110k while OpenVINO Starter Kit vs OpenVINO Starter Kit has 301K logic elements. Does it mean that OpenVINO is faster then DE10-Nano or it only allows more complex logic/alghorithms to be implemented?

[MEIYAN_L_Intel]

Hi,

Question: Does it mean that OpenVINO is faster then DE10-Nano or it only allows more complex logic/alghorithms to be implemented?

Answer:

It can allows more complex design for implementation.

The data transfer speed will depend on timing closure/data width. It is depending on the implementation in the design.

Thanks