Next Generation Internet | Follow on Twitter @ARKwebx0 | @ARKblockchain
Google’s New Tensor Processing Unit Could Be Highly Disruptive
This week Google announced its second generation Tensor Processing Unit (TPU), a processor designed specific for deep learning. Unlike its first generation, Google TPU 2 can drive both training and inference, the two most important workloads for deep learning. Its first chip was useful only with inference workloads.
Traditionally, training has been an intense computational process, demanding power hungry GPUs costing thousands of dollars. Google’s new TPU does the same work at a fraction of the power and cost.
Source: Google
Google’s new TPU puts the company in direct competition with NVIDIA. Just last week NVIDIA announced Volta, a 120 teraflop GPU designed for deep learning. When configured as a 4-chip board, Google’s TPU 2 does 180 teraflops, a remarkable achievement for a company with little semiconductor expertise.
For now, the TPU 2 works only in Google’s cloud with TensorFlow software. NVIDIA’s GPUs are more widely available and support more software frameworks, perhaps offsetting the TPU 2’s speed advantage in the short term.
With its new TPU, Google’s primary objective could be to become the number one cloud provider. Its Cloud Platform currently is in third place, behind Amazon Web Services and Microsoft Azure. Because deep learning is one of the most important trends in tech, Google should be able to attract developers to its cloud with the TPU 2, making it the preferred place to deploy artificial intelligence and deep learning applications.
Another TPU application could be in Alphabet Waymo’s self-driving cars. At the Hot Chips conference in 2016, Alphabet engineers claimed that they would need a chip capable of 50 teraflops to bring their autonomous vehicles to market. At 45 teraflops, Google’s TPU 2 comes quite close, so don’t be surprised to see it in Waymo’s cars in the not to distant future.
Another Chinese Lending Company Lists on a US Exchange
China Rapid Finance (XRF), a Chinese consumer lending company, listed on the U.S. stock exchange in April 2017, following the IPO of another Chinese lender, Yirendai (YRD), in December 2015. Importantly, China Rapid Finanace is partnered with both Tencent and Baidu.
China Rapid Finance is focused on a demographic known as the Emerging Middle Class Mobile Active (EMMA) in China. EMMA individuals do not have any credit history, but do have an employment record. As stated in China Rapid Finance’s filings, “according to the People’s Bank of China, as of the end of 2015, there were approximately 500 million individuals with quality employment records but no credit history.” In other words, the size of this demographic is ~53% more than the entire population of the United States (~325 mm). It also has the U.S. equivalent of a 660–720 FICO score.[1] China Rapid Finance’s loans are small, with terms ranging from a few weeks to three years and principal amounts ranging from ~$73 to ~$14,500.[2]
Given a dearth of traditional credit history, China Rapid Finance uses data science to determine which individuals should receive loans. Its information sources include “historical behavioral data, transactional data, social data, [and] search and employment information.”[3] We are looking forward to learning more about the investment opportunities in the nascent Chinese lending market, particularly how mobile technology could change lending practices, risks, and opportunities.
[1] http://www.lendacademy.com/china-rapid-finance-ipo/
[2] http://marketintelligence.spglobal.com/our-thinking/ideas/investors-might-not-rush-into-china-rapid-finance-s-ipo
[3] http://www.nasdaq.com/markets/ipos/filing.ashx?filingid=11444526
Ethereum Distributed Networks Could Disintermediate Hyper-Scale Cloud Providers
This Friday we attended Ethereal, a conference focused on Ethereum and its broader community. As usual, we were impressed by Ethereum’s evolving intellectual network effect. The conference featured many cryptoassets, with SingularDTV enjoying a particularly strong showing in its quest to build a decentralized entertainment industry atop Ethereum.
Another intriguing project built atop Ethereum is Golem, best described as an attempt to build a worldwide supercomputer. By crowdsourcing compute power from underutilized CPUs and GPUs, Golem potentially could provide compute resources on a scale rivaling the mega-cloud vendors. Given the continued push of digital into the physical world, and the proliferation of compute power in urban areas, Golem’s timing seems right.
Tesla’s electric vehicles, for example, are powered by GPUs in Nvidia’s DrivePX, described as an “on-board ‘supercomputer’ that can provide full self-driving capability.” When parked, the cars could connect to Golem’s network, enabling a mini-GPU datacenter with operations orchestrated by Golem, as well as a source of income for Tesla owners with no wear and tear on their cars. Such a distributed network could become very powerful, potentially rivaling the hyper-scale cloud vendors.
REMINDER:
Along with William Mougayar, CoinDesk, Coin Center, and others, ARK is helping to organize the Token Summit. This is a first-ever event exploring the economics, regulation and best practices around the issuance and operations of new cryptoassets. You can sign up for this two-day event occurring on May 24 and 25 in NYC here.