The mainstream AI chip such as GPU, FPGA, ASIC is comparative

Main component is the AI chip of current main trend 3 kinds, GPU, FPGA, ASIC. Among them GPU, FPGA all is the chip framework with relatively mature early days, belong to general-purpose chip. ASIC is belonged to for the chip with AI specific custom-built setting.

GPU(image handles unit) : GPU assumes image computation job at first, can undertake running paralell computational, accordingly GPU framework itself suits deepness study algorithm quite, those who pass pair of GPU optimize, farther contented deepness studies much consideration demand. Its are main defect depends on power comsumption taller.


FPGA(FieldProgrammableGateArray, the spot but) of array of process designing door: FPGA, have cost of sufficient computational capacity, inferior trial and error and enough flexibility. The computational rate of FPGA is rapid it is to result from it is not to have the instruction, architecture that need not share memory substantially. To saving the requirement of condition, the register in FPGA and piece go up memory (BRAM) is to belong to respective control logic, need not arbitrate gentle is put needlessly, because this FPGA is in operation speed is enough fast, excel GPU. In the meantime, photograph take rough measurements produces the ASIC chip with high cost, because it is a kind of half custom-built hardware, can define among them unit configuration and link framework to have consideration through process designing, because this is on flexibility excel ASIC, have cost of inferior trial and error.

) of integrated circuit of ASIC(ApplicationSpecificIntegratedCircuit specific utility: Undertake specific design and production integrated circuit according to the demand of the product, aggrandizement can undertake on specific function, have higher processing rate and lower specific power consumption. Defect is cost tall, and because be custom-built,change, can duplicate the gender is general, because this only useful capacity is sufficient big when early days of ability enough share is devoted, reduce cost.

Add a Comment

Your email address will not be published. Required fields are marked *