Google claims that its Tensor Processing Unit (TPU), advances machine learning capability by a factor of three generations.
Google CEO Sundar Pichai told the Google’s I/O developer conference that TPUs deliver an order of magnitude higher performance per watt than all commercially available GPUs and FPGA.
Pichai said the chips powered the AlphaGo computer that beat Lee Sedol, the world champion in the incredibly complicated game called Go. Google still is not going into details of the Tensor Processing Unit but the company did disclose a little more information in its blog.
“We’ve been running TPUs inside our data centres for more than a year, and have found them to deliver an order of magnitude better-optimised performance per watt for machine learning. This is roughly equivalent to fast-forwarding technology about seven years into the future (three generations of Moore’s Law),” the blog said. “TPU is tailored to machine learning applications, allowing the chip to be more tolerant of reduced computational precision, which means it requires fewer transistors per operation. Because of this, we can squeeze more operations per second into the silicon, use more sophisticated and powerful machine learning models, and apply these models more quickly, so users get more intelligent results more rapidly.”
The tiny TPU can fit into a hard drive slot within the data centre rack and has already been powering RankBrain and Street View, the blog said.
What Google is not saying is what a TPU actually is and if it will be a replacement for a CPU or a GPU. Word on the street is that the TPU could be a form of chip that implements the machine learning algorithms that are crafted using more power hungry GPUs and CPUs.
Chipzilla’s managers are incandescent with rage that they are being ordered to “sell” Intel’s decimation of staff to the people who are given the boot.
Intel wants to get rid of more than 11 percent of its staff as it realises that it has lost the battle for the hearts and minds of the mobile industry. It is shutting down its projects to build chips for Mobile Phones and Tablets.
Techeye has been told that a special webcast is scheduled on Monday for managers only, and it appears that it will be part of a hard core sales pitch to pressurise managers to convince employees to take voluntary redundancy.
As one manager moaned: “Why the Fuck do I have to deliver the VSP Bullshit to my affected reports instead of them doing their own dirty work?”
He said it was bad enough when Intel’s front line were told to trap staff to hit SSL targets and then force them to have the ISP blood on their hands through the respect and dignity execution.”
“Will probably be required to take waterboarding training next,” he muttered.
The comment has appeared on a staff site called “the lay-off” and already has attracted a few comments from hacked off fellow managers. Oddly one does not seem particularly upset at front line manager’s (FLMs) plight.
“OP, please, do not protect and SLMs. They are falsch/fake/trash, low level life forms (respect to 10% of them who R exceptions), protecting ONLY their fat asses.”
Another thought that the whole situation proved that Intel had “really had become a shit place to work”.
An Nvidia engineering team has built a self-driving car with one camera, one Drive-PX embedded computer and only 72 hours of training data.
Nvidia published an academic preprint of the results of the DAVE2 project entitled End to End Learning for Self-Driving Cars.
DAVE2 is named after a 10-year-old Defense Advanced Research Projects Agency (DARPA) project known as DARPA Autonomous Vehicle (DAVE). Coincidently it is also the name of the astronaut in 2001, a Space Oddessy. The phrase “I can’t do that Dave” is now the blue screen of death” of robotic history.
The Nvidia team trained a convolutional neural network (CNN) to map raw pixels from a single front-facing camera directly to steering commands. Nvidia’s breakthrough is the autonomous vehicle automatically taught itself by watching how a human drove, the internal representations of the processing steps of seeing the road ahead and steering the autonomous vehicle without explicitly training it to detect features such as roads and lanes.
Although in operation the system uses one camera and one Drive-PX embedded computer, the training system used three cameras and two computers to acquire three-dimensional video images and steering angels from the vehicle driven by a human that were used to train the system to see and drive.
Intel has become a monster which eats its own tail and firing thousands of staff who do not fit into its glorious Internet of Things and cloudy vision of the future.
Intel has announced that it is giving more that 11 percent of its employees their pink slips by next year. This decimation will come through site consolidations worldwide, a combination of voluntary and involuntary departures, and a re-evaluation of programmes.
Intel expects its restructuring to deliver $750 million in savings this year and annual run rate savings of $1.4 billion by mid-2017. The company will record a one-time charge of approximately $1.2 billion in the second quarter.
In the official announcement Intel described it as a “restructuring initiative” which will help it evolve from a PC company to one that powers the cloud and billions of smart, connected computing devices. Intel will intensify its focus in high-growth areas where it is positioned for long-term leadership, customer value and growth, while making the company more efficient and profitable.
The move is remarkably similar to that which has gotten IBM into so much trouble. Intel sees the data center and Internet of Things (IoT) businesses are Intel’s primary growth engines. However these particular technologies are not making up for the cash which is been lost to the hardware business.
Intel said that memory and field programmable gate arrays (FPGAs) accelerating these opportunities and “fueling a virtuous cycle of growth for the company.”
The restructuring initiative was outlined in an e-mail from Intel CEO Brian Krzanich to Intel employees.
“Our results over the last year demonstrate a strategy that is working and a solid foundation for growth. The opportunity now is to accelerate this momentum and build on our strengths. “These actions drive long-term change to further establish Intel as the leader for the smart, connected world. I am confident that we’ll emerge as a more productive company with broader reach and sharper execution.”
Chipzilla plans to increase investments in the products and technologies that that will fuel revenue growth, and drive more profitable mobile and PC businesses. Through this comprehensive initiative, the company plans to increase investments in its data center, IoT, memory and connectivity businesses, as well as growing client segments such as 2-in-1s, gaming and home gateways.
Beancounters working for Barrons have added up the numbers and divided them by their shoe size and decided that Intel will grow like topsy next year.
Barron’s claims that Chipzilla’s shift to higher-growth businesses such as server chips and embedded chips for cars could drive a 25 increase in its shares in a year.
While there is a risk Intel could cut its financial guidance for the year when the chipmaker reports earnings on Tuesday, it is likely to return to sustainable growth by year’s end for the first time in seven years, the publication said.
Those who do not own shares in Chipzilla should wait until after the earnings call to buy shares, it added.
Intel has had a pants few years as demand for personal computer chips has dried up, Barron’s said, but growth in the company’s data centre group, which includes server chips, could eventually bring in more revenues.
The gap between the two businesses has closed over the past five years.
Last year, the data centre business’s operating profit was $7.8 billion, slightly below the $8.2 billion earned by Intel’s client computing division, which includes chips for desktop and notebook computers. In 2010, the data center division brought in just $4.4 billion, compared to the personal computer business’s $13 billion.
Meanwhile, the company’s Internet of Things division, which includes chips for cars, medical devices and factories, composed just four percent of revenue last year but is growing.
US chipmaker Integrated Device Technology (IDTI) appeared to be about to be bought out by a shareholder investor group on Monday.
A regulatory filing by a group identifying itself as Integrated Device shareholders said it made an offer to buy the company for $32 per share in cash. The price represented a 64 percent premium to Monday’s close and valued the company at $4.3 billion.
But the filing was the first and only information that Integrated Device has received from this group.
Company’s chief executive, Greg Waters, said the company has not communicated with any of these parties.
Waters said the company is unaware of any other information that would support a determination that the group’s proposal represents a credible bona fide offer. He said the company will evaluate further information received to determine the credibility of the offer.
The filing made with the Securities and Exchange Commission said that the investor group is comprised of six Chinese and one Pakistani citizens, and is led by Libin Sun, who owns a 4.4 percent stake in the company.
The filing showed that Nauman Aly, who held a 0.1 percent stake, had acquired a block of call options for IDTI shares.
However Aly appeared to have sold all his IDTI calls for a profit of about $427,700, roughly matching the activity in those two minutes of trading.
Chipzilla has officially launched its new Xeon Processor E5 v4 today.
Xeon E5 v4, which will replace last year’s E5 v3 series will target a wide array of market segments, from high-performance professional workstations to multi-socket servers for big data. In fact the v4 is pretty similar and is even socket compatible.
The Broadwell-EP based Xeon E5 v4 uses Intel’s more advanced 14nm process node and the biggest of the chips can feature up to 22 processor cores (44 threads). The E5 v4 series still supports up to quad-channel DDR4 memory, but the maximum supported speed now tops out at 2400MT/s, up from 2133MT/s.
While this is significant, the rest of the platform remains mostly unchanged. Thanks to its additional cores, the E5-2600 v4 series now features up to 55MB of last-level cache. Support for 3D die stacked LRDIMMs has been added, along with DDR4 write CRC, and of course the higher speeds. Though, with three 3DS LRDIMMs per channel, the max supported frequency drops down to 1600MHz.
The changes to the Xeon E5 V4 family’s memory configuration bring in reduced latency and increased bandwidth. Intel’s numbers show up to a 15 per cent increase in bandwidth with latency reductions across the board.
In addition to these high-level updates, there are also new virtualization and security related features, along with more performance and efficiency enhancements as too.
TSMC has signed an agreement with the Nanjing City Government to invest $3 billion building an advanced wafer manufacturing facility in China.
TSMC said in December it planned to set up its first wholly owned advanced fabrication plant in China with a $3 billion investment, highlighting the growing importance of the Chinese market for semiconductor giants.
Now TSMC Chairman Morris Chang has given out some details on the project saying that a 12-inch fab and our design service centre will be established in Nanjing.
“We aim to provide closer support to customers as well as expand our business opportunities in China in step with the rapid growth of the Chinese semiconductor market over the last several years,” said
The new plant will make 20,000 12-inch wafers per month. Production will begin in the second half of 2018.
It was not that easy to get the deal past the Taiwan government. Taiwan has restricted manufacturing activities of its prized semiconductor sector in China, amid political tension between the neighbours. However, competition from China’s fast-growing, though fledgling chip industry has put pressure on Taiwanese companies to widen their mainland footprint.
TSMC had urged authorities to allow 12-inch facilities, which use more advanced technology processes than 8-inch plants, to be wholly owned out of concern for intellectual property protection. TSMC already has a wholly owned 8-inch chip making plant near Shanghai.
Chipzilla has decided that the world has had enough of its Tick-Tock strategy and is having its Tock removed and replaced with something with a longer rhythm called PAO.
Intel has produced chips on a yearly tick-tock cycle for the last decade. Thanks to the shrinking die sizes, that process may permanently become a three-step.
For those who came in late Chipzilla had significant issues going from 22- to 14-nanometers, and it extended the latter to a third generation with “Kaby Lake” CPUS. This was the first time that Intel had a break from tick-tock. Now it looks like the longer rhytum has a name which Intel calls “process, architecture, optimisation (PAO).” This will continue for its upcoming 10-nanometer chips.
During “tick” years, the chip giant upgraded its manufacturing technology to make circuits smaller — for its latest chips, for instance, the tick cycle reduced traces from 22- to 14-nanometers. During “tock” years, it uses the same circuit size and manufacturing technique, but changes the microcode, often drastically, to make chips faster and more energy efficient.
While Intel said that the latest 14-nanometer chips were on a “2.5 year cycle,” it plans to introduce three different 10-nanometer chips yearly.
With the three-step PAO, that slows the pace of innovation by effectively a third, meaning consumers will have to wait an extra year before they see significant speed improvements. The third year of a chip’s life cycle will likely see smaller performance gains, giving power users and gamers — who have become critical customers — less reason to upgrade.
Intel said that the new process is a direct result of the difficulty in building chips with traces that are just 20 silicon atoms wide.
“We expect to lengthen the amount of time we will utilize our 14nm and our next generation 10nm process technologies, further optimizing our products and process technologies while meeting the yearly market cadence for product introductions,” according to the document.
Intel has maintained that it will introduce 10-nanometer chips before its rivals. Furthermore, it says that “this competitive advantage will be extended in the future as the costs to build leading-edge fabrication facilities increase.”
The courtroom showdown between Department of Justice and Apple that was scheduled has been postponed and rumours are that the FBI has worked out a way to crack the iPhone and does not need Jobs’ Mob’s help any more.
The DOJ made the request after saying it may have found an alternative way into the iPhone used by Syed Farook, the gunman in the San Bernardino shooting and might not need Apple’s assistance.
After Apple had been telling the world that the phone was unhackable and that the FBI was demanding that it rewrite its operating system to do the job, a hacker apparently demonstrated to the FBI how he or she would do it.
Of course it is a lot easier than Apple claimed, and does not require any of the hoops that Apple claimed. The hack probably just depends on exploiting one of Apple’s legendary security flaws. The good part of this particular hacker’s work means that the FBI does not have to ask Apple again and is not obliged to tell Jobs’ mob where the hole is.
Had Apple done what it was told, it could have controlled how the FBI got the data and known exactly how to close any breach. It seems by carrying out the highly public war on the FBI, Apple has painted itself into a corner and forced the Feds to find a back door – the very thing that Apple claimed it did not want.
The senior law enforcement official said the DOJ is “cautiously optimistic” the method will work, and it will notify the results to the judge by April 5.