Bitcoin networks owes 70% of its processing prowess to Bitmain which had a considerable role to build this technical architecture. It is involved in making specialized chips which hold the potential to perform critical hash functions which are very much required during Bitcoin trading and Bitcoin mining activities as well as packaging of these chips into Antminer S9, the topmost Bitcoin mining rig.
Jihan Wu, the CEO of Bitmain has perceived a future which extends further beyond the world of cryptocurrency and blockchain. He stated out to Morgen E. Peck, the contributing editor of IEEE Spectrum in July that, “It’s quite personal that I wanted Bitcoin to be successful. But as a company we are not allowed to solely rely on the success of Bitcoin. That’s a thing we cannot afford.”
Bitmain is applying it’s Bitcoin playbook to Artificial Intelligence. Micree Zhan, the CEO of Bitmain finalized the date of 8th November to discuss about the Sophon BM1680, their latest AI chip at AIWORLD in Beijing. Following such discussion the company will start off its sale of systems built atop the AI skeleton.
Bitmain has claimed that the AI chip can handle both execution and training of deep learning algorithms. The execution task depends on the chip’s ability to solve inference. David Schneider exclaimed earlier this year at Spectrum’s that performing inference and training an AI requires a completely different skillset from the processor. Training for example requires an ability to solve high-precision maths while low precision is perfect for getting started with inference.
8-bit math is used by Google’s Tensor Processing Unit for inferencing. Its specialisation states that the BM1680 makes use of 32-bit floating point math and can perform 2 teraflops, that is 2 trillion floating point operations per second. For doing the same it consumes 25 Watts but can hike up to 41 Watts while running flat out. Bitmain has announced that its architecture is much similar to Google’s TPU. The generic architecture class hailed as systolic allows data to flow through the processing cores in waves.
Janet Zhao, a part of Bitmain’s high-performance computing group stated out in an email that, “Google TPU is a cool role model among all AI ASIC makers. When we started to develop our own AI chip…. in the end of 2015, we adopted the enhanced systolic technology which is similar to Google TPU.”
Bitmain shall start selling a specialized server soon which is powered by the Sophon BM1680. Meant to accelerate deep learning applications, the Sophon SC1 card is going to be housed inside the SS1 server especially for image and video analysis.
The company has been communicating with business and technical teams belonging to potential customers such as Alibaba, Baidu and Tencent since the last one and half years of development. Zhao added that these companies are worried about the supply stability, cost and power consumption of GPU-based AI accelerators which has added inertia in their search for a new vendor who shall guarantee all of the ones mentioned above.
KryptoMoney.com publishes latest news and updates about Bitcoin, Blockchain Technology, Cryuptocurrencies and upcoming ICO’s. Subscribe to our newsletters to stay updated.