Bitmain reveals Sophon BM1680 neural processor for the machine learning market

A couple of months ago, we wrote about the AI efforts of Bitmain. This is one of the dominant forces in the Bitcoin market, this company's mining pool accounts for nearly 30 percent of the processing power on the Bitcoin network and they design their own ASICs.

One of the new markets Bitmain wants to get into is artificial intelligence, it aims to use the expertise it gained in designing Bitcoin ASICs to take on the tech industry big-wigs. Over at the AIWORLD 2017 Artificial Intelligence Conference in Beijing, Bitmain founder and CEO Micree Zhan revealed the company's new Sophon BM1680 neural processor as well as accelerator cards and a full server based on this chip.

The Sophon BM1680 was manufactured on TSMC's 28nm HPC+ process and the chip a TDP of 41W. It offers 64 NPUs and promises 2 teraflops of single-precision computing power. The Sophon SC1 board features 16GB DDR4 and has a total board power of up to 85W. WikiChip provides some architectural details over here.

At the heart of the chip is the NPU subsystem which consists of 64 NPUs, the hub, and an NPU Schedule Engine. The scheduling engine is in charge of controlling the data flow to the individual NPUs. Bitmain has not disclosed the intimate details of the NPU cores but we do know it has 512 KiB of program-visible SRAM and supports 64 single-precision operations. With a total of 64 NPUs, the chip has a total of 32 MiB of cache and a peak performance of 2 TFLOPS (single-precision). The company stated the chip is capable of 80 billion algorithmic operations per second, but since we don’t know the exact meaning of those operations, it’s impossible to compare that number to existing chips.

The Sophon SC1 is available immediate for $589, that card features a single BM1680 plus a FPGA. The company will also make a Sophon SC1+, that version will feature two BM1680 chips and promises 4 teraflops of single-precision computing power.



A second-generation Sophon chip is expected in the second half of 2018, that chip will be made using TSMC's 12nm FinFET process.