Huawei’s Kirin 990 SoC Is the First Chip With an Integrated 5G Modem


This site may earn affiliate commissions from the links on this page. Terms of use.

Huawei’s year has been anything but good, but the company has pushed ahead with new technology introductions and smartphone designs. The Chinese firm has now announced its latest SoC, the Kirin 990. The new chip will ship in two flavors — the Kirin 990, and the Kirin 990 5G. These two chips are based on the same SoC design, but there are some significant differences between them.

First, the Kirin 990 5G is built on TSMC’s 7nm+ process node, which utilizes EUV. The Kirin 990, in contrast, is a standard 7nm design. It seems as though Huawei will be the first customer to ship a part that uses EUV for manufacturing. Huawei’s stated reason for using EUV for the 5G variant is that it allowed for a smaller die. Die size on the 5G part is larger than 100mm2, while the LTE chip is less than 90mm2. Transistor counts are also significantly different, with the LTE chip at 8B and the 5G chip at 10.3B.

Kirin-990-Comparison

One interesting fact that Anandtech mentions is that the Kirin 990 was originally expected to use ARM’s Cortex-A77 CPU, not the Cortex-A76. Apparently the Huawei team didn’t like how the Cortex-A77 clocked on TSMC’s 7nm process node. The A77 had higher peak performance, but overall power efficiency between the A76 and A77 was practically identical on 7nm and the A76 design was capable of hitting much higher clocks. Supposedly the A77 tops out around 2.2GHz on 7nm at the moment and the design may not be used widely until 5nm CPUs are available.

The new ARM Mali-G76 implementation is substantially wider than the 10-core implementation used on the previous generation Kirin 980. GPU power efficiency can often be improved by using a wider GPU clocked at lower frequencies, and Huawei believes the new GPU design will still be more power-efficient than the old Kirin 980, despite being substantially wider.

The NPU design is a homegrown Huawei effort. Where the company previously licensed an NPU from Cambricon, the new Kirin 990 uses Huawei’s Da Vinci architecture. Huawei intends to scale this AI processing block from servers to smartphones. It supports both INT8 and FP16 on both cores, whereas the older Cambricon design could only perform INT8 on one core. There’s also a new ‘Tiny Core’ NPU. It’s a smaller version of the Da Vinci architecture focused on power efficiency above all else, and it can be used for polling or other applications where performance isn’t particularly time critical. The 990 5G will have two “big” NPU cores and a single Tiny Core, while the Kirin 990 (LTE) has one big core and one tiny core.

Huawei’s Balong modem will support sub-6GHz 5G signals with a maximum of 2.3Gbps download and 1.25Gbps upload. Overall CPU performance improvements from the Kirin 980 to the Kirin 990 are modest — Huawei claims single-threaded gains of 9 percent and multi-threaded boosts of 10 percent. Power efficiency, however, has improved significantly. The top-end cores are supposedly 12 percent more efficient, the “middle” cores of Huawei’s Big.Little.littlest are supposedly 35 percent more efficient, and the low-end Cortex-A55 chips are 15 percent more efficient. Most workloads are supposed to run on the middle cores for maximum performance/watt.

It seems unlikely that these devices will come to the US market in any numbers, though you may be able to buy them from third-party resellers if the Trump Administration doesn’t take further action against the company. While devices are going to start carrying 5G modems from this point forward, I’ve yet to see a 5G phone I’d actually recommend. While it’s true that the first generation of LTE devices didn’t exactly cover themselves in glory, the first generation of LTE devices didn’t overheat and shutdown when summer temperatures rose above 85F / 29.4C. They didn’t require you to be literally standing underneath an LTE access point in order to see faster service, either. Verizon has already stated that outside city centers, its 5G network will closely resemble “good 4G,” which raises the question of what, exactly, consumers are paying all this money for.

The first LTE devices were the HTC Evo 4G, the Samsung Craft, and the HTC Thunderbolt. They sold for $200, $350, and $250, respectively, though this was in the era of two-year contracts. Apple’s first LTE device was the iPhone 5, which cost $649 if purchased without a contract. Assuming Apple and the other AndroidSEEAMAZON_ET_135 See Amazon ET commerce manufacturers continue to offer 5G as a luxury feature, we’ll likely only see it on devices at or above the $1000 price point for the next 12 months. I wouldn’t pay $1000 for a phone under any circumstances, but I definitely wouldn’t step up to a $1000+ device to buy a feature that I’ve got no chance of using at any point in the next few years.

Now Read:

 




10 minutes mail – Also known by names like : 10minemail, 10minutemail, 10mins email, mail 10 minutes, 10 minute e-mail, 10min mail, 10minute email or 10 minute temporary email. 10 minute email address is a disposable temporary email that self-destructed after a 10 minutes. https://tempemail.co/– is most advanced throwaway email service that helps you avoid spam and stay safe. Try tempemail and you can view content, post comments or download something

Facebook Is Building a Minecraft AI Because Games May Be Great Training Tools


This site may earn affiliate commissions from the links on this page. Terms of use.

It turns out that video games may be an excellent method of teaching skills to artificial intelligence assistants. That’s the theory of a group of researchers working for Facebook, who have focused on Minecraft as a potential teaching tool for building generalist AI — a so-called ‘virtual assistant.’ The research team isn’t trying to build an artificial intelligence that’s super-good at classifying images or other content — it wants to build a generalist AI that can perform a much larger number of tasks reasonably well.

This is, to-date, an under-studied area of research. The authors’ write:

There has been measured progress in this setting as well, with the mainstreaming of virtual personal assistants. These are able to accomplish thousands of tasks communicated via natural language, using multi-turn dialogue for clarifications or further specification. The assistants are able to interact with other applications to get data or perform actions.

Nevertheless, many difficult problems remain open. Automatic natural language understanding (NLU) is still rigid and limited to constrained scenarios. Methods for using dialogue or other natural language for rich supervision remain primitive. In addition, because they need to be able to reliably and predictably solve many simple tasks, their multi-modal inputs, and the constraints of their maintenance and deployment, assistants are modular systems, as opposed to monolithic ML models. Modular ML systems that can improve themselves from data while keeping well-defined interfaces are still not well studied.

According to the team, they picked Minecraft because it offered a regular distribution of tasks with “hand-holds for NLU research,” as well as enjoyable opportunities for human-AI interaction, with plenty of opportunities for human-in-the-loop research. Minecraft, for those of you who haven’t played or heard of it, is a block-based crafting and exploration game in which players explore a 3D voxel grid universe populated with various types of materials, neutral characters, and enemies. The team’s goal is to build an AI virtual assistant that can be given instructions in natural language by a Minecraft player, and that can reliably complete some of the primary tasks that player might engage in, including gathering materials, building structures, fighting mobs, and crafting items.

The authors of the paper target three specific achievements: Create synergy between machine-learning and non-machine-learning components, allowing them to work together; create a “grounded” natural language simulation that allows the AI to understand what players want it to do, and can communicate its success or failure back to the end-user; and create an AI that shouldn’t just be capable of doing what the player wants it to do, but that also its performance should improve based on observation of the human player.

They write:

We intend that the player will be able to specify tasks through dialogue (rather than by just issuing commands), so that the agent can ask for missing information, or the player can interrupt the agent’s actions to clarify. In addition, we hope dialogue to be useful for providing rich supervision. The player might label attributes about the environment, for example “that house is too big”, relations between objects in the environment (or other concepts the bot understands), for example “the window is in the middle of the wall”, or rules about such relations or attributes. We expect the player to be able to question the agent’s “mental state” to give appropriate feedback, and we expect the bot to ask for confirmation and use active learning strategies.

The machine learning code being used for the Facebook-Minecraft bot is available on GitHub. Better AI tools could be useful in many games, though they could also raise serious questions about what constitutes cheating in multiplayer.

Now Read:




10 minutes mail – Also known by names like : 10minemail, 10minutemail, 10mins email, mail 10 minutes, 10 minute e-mail, 10min mail, 10minute email or 10 minute temporary email. 10 minute email address is a disposable temporary email that self-destructed after a 10 minutes. https://tempemail.co/– is most advanced throwaway email service that helps you avoid spam and stay safe. Try tempemail and you can view content, post comments or download something

Amelia Earhart Expedition Ends With Potential Clues, but No Answers


This site may earn affiliate commissions from the links on this page. Terms of use.

Robert Ballard’s expedition to Nikumaroro to search for evidence of Amelia Earhart’s final resting place has ended without locating the Lockheed Electra. The search isn’t heading back entirely empty-handed — a land-based search party is still working on the island, and National Geographic pointed to a silver sheet of metal with what appear to be rivet holes as potentially being an artifact connected to the lost Earhart aircraft  — but no firm or final artifact was found to mark the spot where Earhart landed.

Ballard focused his search in the northwest corner of the island, near the lagoon. There were multiple reasons to focus on this location. The so-called Bevington Photo, taken by Cadet Officer Eric Bevington, while leaving Nikumaroro — then called Gardner Island — shows a strange object (inside the red circle) sticking up above the waves. The photo was initially discovered by TIGHAR, which has led searches for Earhart for decades.

Bevington Photo

The Bevington Photo. SS Norwich City in the background.

An enlarged image of that object is shown below:

When we ran our deep dive on Amelia Earhart and the search for her aircraft, one of our readers — a senior mechanical engineer at a U.S. electronics manufacturer — contacted us with a bit of sleuthing they’d done on their own time. He used multiple AI programs to sharpen the enhanced version of that image, and writes:

The first program I used, Sharpen A.I., has three modes: Shapen, Focus, and Stabilize. I tried all three and found that using Stabilize worked best. The photo was oversampled given its limited resolution so there is more “blur” than “information.” However, there was still a very small amount of vertical motion in the camera when the photo shot, resulting in some vertical motion blur, which Focus A.I. can do amazingly good work with.

Then I used Gigapixel A.I. to double the pixel count. When you have genuine information to work with (sharp images limited by resolution), Gigapixel A.I. can sometimes give results that truly astonish. There’s so little real information in the Bevington Object photo that the improvement afforded by Gigapixel A.I. was pretty limited. But the improvement seemed to be there, so I kept it.

After doubling the pixel count with Gigapixel A.I., I then halved the resolution (restoring it to its original 343-pixel width) using the Preview app in MacOS. To me, frankly, the sharpened photo appears more like water breaking over a rock than anything else. The way people’s vision system recognizes objects is heavily influenceable by suggestion.

The sharpened version of the AI image is below:

stablize double half (1)

In our larger write-up on the topic, we discuss why the Bevington Photo was considered to be such potentially powerful evidence that the Lockheed Electra had landed in this spot. Despite extensively surveying the target search zone and underwater canyons and “chutes” that descend from the coast of Nikumaroro, Ballard and his team found little of note.

There is, however, some cause for hope. National Geographic writes that the land search team may have found fragments of the skull found on Nikumaroro back in 1940. Once theorized to belong to Earhart, the skull and other bones were thought lost in WW2. The skull is being reconstructed and will be DNA-tested in the next few months. Nat Geo doesn’t provide any further information on the status of the land team, but the DNA analysis of those bone fragments might just answer this question once and for all. If we can’t find the wreckage of the Electra or any objects guaranteed to be associated with it, DNA evidence of either Earhart or her navigator, Fred Noonan would also confirm the story.

Ballard, meanwhile, is heading to Howland and Baker Islands — Howland was Earhart’s original destination — to map the undersea areas around both of them. He has no plans to return to the Earhart search but knows where he’d look if he did: the southern part of the island, where there are beaches and a potential campsite that may have been associated with the castaways.

Now Read:




10 minutes mail – Also known by names like : 10minemail, 10minutemail, 10mins email, mail 10 minutes, 10 minute e-mail, 10min mail, 10minute email or 10 minute temporary email. 10 minute email address is a disposable temporary email that self-destructed after a 10 minutes. https://tempemail.co/– is most advanced throwaway email service that helps you avoid spam and stay safe. Try tempemail and you can view content, post comments or download something

Megvii, the Chinese startup unicorn known for facial recognition tech, files to go public in Hong Kong – gpgmail


Megvii Technology, the Beijing-based artificial intelligence startup known in particular for its facial recognition brand Face++, has filed for a public listing on the Hong Kong stock exchange.

Its prospectus did not disclose share pricing or when the IPO will take place, but Reuters reports that the company plans to raise between $500 million and $1 billion and list in the fourth quarter of this year. Megvii’s investors include Alibaba, Ant Financial and the Bank of China. Its last funding round was a Series D of $750 million announced in May that reportedly brought its valuation to more than $4 billion.

Founded by three Tsinghua University graduates in 2011, Megvii is among China’s leading AI startups, with its peers (and rivals) including SenseTime and Yitu. Its clients include Alibaba, Ant Financial, Lenovo, China Mobile and Chinese government entities.

The company’s decision to list in Hong Kong comes against the backdrop of an economic recession and political unrest, including pro-democracy demonstrations, factors that have contributed to a slump in the value of the benchmark Hang Seng index. Last month, Alibaba reportedly decided to postpone its Hong Kong listing until the political and economic environment becomes more favorable.

Megvii’s prospectus discloses both rapid growth in revenue and widening losses, which the company attributes to changes in the fair value of its preferred shares and investment in research and development. Its revenue grew from 67.8 million RMB in 2016 to 1.42 billion RMB in 2018, representing a compound annual growth rate of about 359%. In the first six months of 2019, it made 948.9 million RMB. Between 2016 and 2018, however, its losses increased from 342.8 million RMB to 3.35 billion RMB, and in the first half of this year, Megvii has already lost 5.2 billion RMB.

Investment risks listed by Megvii include high R&D costs, the U.S.-China trade war and negative publicity over facial recognition technology. Earlier this year, Human Rights Watch published a report that linked Face++ to a mobile app used by Chinese police and officials for mass surveillance of Uighurs in Xinjiang, but it later added a correction that said Megvii’s technology had not been used in the app. Megvii’s prospectus alluded to the report, saying that in spite of the correction, the report “still caused significant damages to our reputation which are difficult to completely mitigate.”

The company also said that despite internal measures to prevent misuse of Megvii’s tech, it cannot assure investors that those measures “will always be effective,” and that AI technology’s risks and challenges include “misuse by third parties for inappropriate purposes, for purposes breaching public confidence or even violate applicable laws and regulations in China and other jurisdictions, bias applications or mass surveillance, that could affect user perception, public opinions and their adoption.”

From a macroeconomic perspective, Megvii’s investment risks include the restrictions and tariffs placed on Chinese exports to the U.S. as part of the ongoing trade war. It also cited reports that Megvii is among the Chinese tech companies the U.S. government may add to trade blacklists. “Although we are not aware of, nor have we received any notification, that we have been added as a target of any such restrictions as of the date this Document, the existence of such media reports itself has already damaged our reputation and diverted our management’s attention,” the prospectus said. “Whether or not we will be included as a target for economic and trade restrictions is beyond our control.”


10 minutes mail – Also known by names like : 10minemail, 10minutemail, 10mins email, mail 10 minutes, 10 minute e-mail, 10min mail, 10minute email or 10 minute temporary email. 10 minute email address is a disposable temporary email that self-destructed after a 10 minutes. https://tempemail.co/– is most advanced throwaway email service that helps you avoid spam and stay safe. Try tempemail and you can view content, post comments or download something

DeepMind Co-Founder Mustafa Suleyman Placed on Leave


This site may earn affiliate commissions from the links on this page. Terms of use.

Artificial intelligence and machine learning are transforming the technology industry, which is why Google dropped $650 million on UK-based DeepMind in 2014. DeepMind co-founder Mustafa Suleyman has been at the heart of the company’s research and industry outreach efforts, but now he’s mysteriously been placed on leave from the company he started with current CEO Demis Hassabis. DeepMind has been vague about the situation, but Bloomberg claims the move stems from controversy over some of Suleyman’s projects. 

It’s hard to overstate how vital neural networks like the ones developed by DeepMind can be to a company like Google — it’s more than machines that can crush human StarCraft II and Go players. Google uses neural networks to generate natural-sounding speech, categorize photos, and process photos. AI is what took Google’s smartphone cameras from poor to industry-leading. 

At DeepMind, Suleyman headed up the “Applied AI” group, which attempted to find uses for the company’s technology in fields like energy and healthcare. This led Suleyman to become a key figure communicating the company’s intentions and ethical controls on the use of AI. That hasn’t always gone smoothly, though. One 2016 initiative from the applied AI group involved a collaboration with the UK’s National Health Service. The team wanted to create a mobile app called Stream that could predict acute kidney issues. The company got access to 1.6 million patient records, but a UK court later ruled that deal violated patient privacy. DeepMind and Suleyman both issued apologies. 

In 2018, Google announced it was absorbing the Streams team, creating a new division called Google Health. DeepMind Health ceased operation, and Suleyman was removed from the day-to-day running of the unit. It seems unlikely the latest development is still blowback from Streams. Suleyman is important to DeepMind, and DeepMind is important to Google. By some measures, the company burns $500 million per year, but Google is leaning hard into AI as its future. 

DeepMind spokespeople say that Suleyman is “taking time out right now after 10 hectic years.” They frame this as a mutual decision that will only last a few months, and that could be entirely true. However, this would not be the first time a nasty breakup was smoothed over by PR, something hinted at in the Bloomberg report. Bloomberg doesn’t delve into the nature of the “controversy,” leaving everyone to speculate.

Now read:




10 minutes mail – Also known by names like : 10minemail, 10minutemail, 10mins email, mail 10 minutes, 10 minute e-mail, 10min mail, 10minute email or 10 minute temporary email. 10 minute email address is a disposable temporary email that self-destructed after a 10 minutes. https://tempemail.co/– is most advanced throwaway email service that helps you avoid spam and stay safe. Try tempemail and you can view content, post comments or download something

Scientists Are Using AI to Recreate the Rules of Long Lost Board Games


This site may earn affiliate commissions from the links on this page. Terms of use.

Human beings tend to bury objects that they consider most important to their lives. It’s not surprising, therefore, that some people chose to buy games, including board games. Some of these artifacts have survived the centuries, leaving modern-day archaeologists with a puzzle — how do you understand a game when you have no idea how it was played? Ancient humans, it seems, were downright terrible at losing the manuals for their boxed titles.

Tracing the evolution of games and gaming could tell us a great deal about cultural exchange and evolution between two societies or within one society over time. In a few cases, archaeologists have gotten lucky and found the actual rules to the game. More often, they’re stuck trying to intuit how a game might have worked by comparing it with other games that we do understand, or analyzing how a title is portrayed in art. The only clues for playing the ancient Egyptian game Senet, for example, are in the tomb of Queen Nefertari.

Image is in the public domain

Now, historians are using new tools to recover the rules for ancient games. Cameron Browne leads the Digital Ludeme Project. It’s a research project aimed at using computational techniques, including AI, to recreate the rules of board games. To do this, Browne and his team first break a game down into its constituent parts and elements, codifying them as units called “ludemes.” A ludeme is any game piece or rule that the researchers are aware of. Cultural information is also incorporated to help evaluate the plausibility of various evaluated rulesets. The system they’ve built for actually playing the games is called Ludii.

Modern AI techniques are used to create potential rulesets for games, which can then be computationally evaluated. AI agents are assigned to play games using proposed rules and to build movement lists. The data the AI agents collect in multiple playthroughs can then be evaluated to determine whether the end result is a playable game.

“With our system, we can put in the equipment, we can look for rules from the date [of the board’s creation], rules from the area, rules from the cultural context,” Browne told Vice. “Then we can piece together likely rule sets that might have occurred.”

The team has already had some success. This article describes a game known as latrunculi or Ludus latrunculorum, which scholars have had trouble deciphering. Other scholars have proposed their own rules, but Browne believes Ludii has defined a more likely set of rules for the game — and you can actually try it out, if you use the beta version of the application and select the appropriate game.

Training the AI data set on the rules of the thousands of known games and teaching the AI how game rules evolved and were transmitted over time in the situations we can trace already could help teach the AI how to map the most likely routes for cultural transmission in the past. Some games, like latrunculi, were played for centuries across the entire Roman Empire.

Our difficulty in understanding the games of the past sheds some interesting light on the battle to preserve gamesSEEAMAZON_ET_135 See Amazon ET commerce in the present. Efforts to preserve game data in machine-readable form and to make certain that the unique experiences of gaming are not lost to the ravages of time are not simply driven by piracy or an insane dedication to nerd-dom. The individuals who preserved humanity’s games from centuries ago were saving a valuable relic of their own time. Knowing how a society chooses to spend its leisure time and the sorts of games it finds amusing tells us something about the society, no matter what the type of entertainment is.

Feature image courtesy of Wikipedia

Now Read:




10 minutes mail – Also known by names like : 10minemail, 10minutemail, 10mins email, mail 10 minutes, 10 minute e-mail, 10min mail, 10minute email or 10 minute temporary email. 10 minute email address is a disposable temporary email that self-destructed after a 10 minutes. https://tempemail.co/– is most advanced throwaway email service that helps you avoid spam and stay safe. Try tempemail and you can view content, post comments or download something

Intel Details Its Nervana Inference and Training AI Cards


This site may earn affiliate commissions from the links on this page. Terms of use.

Hot Chips 31 is underway this week, with presentations from a number of companies. Intel has decided to use the highly technical conference to discuss a variety of products, including major sessions focused on the company’s AI division. AI and machine learning are viewed as critical areas for the future of computing, and while Intel has tackled these fields with features like DL Boost on Xeon, it’s also building dedicated accelerators for the market.

The NNP-I 1000 (Spring Hill) and the NNP-T (Spring Crest) are intended for two different markets, inference and training. “Training” is the work of creating and teaching a neural network how to process data in the first place. Inference refers to the task of actually running the now-trained neural network model. It requires far more computational horsepower to train a neural network than it does to apply the results of that training to real-world categorization or classification tasks.

Intel’s Spring Crest NNP-T is designed to scale out to an unprecedented degree, with a balance between tensor processing capability, on-package HBM, networking capability, and on-die SRAMs to boost processing performance. The underlying chip is built by TSMC — yes, TSMC — on 16nm, with a 680mm2 die size and a 1200mm2 interposer. The entire assembly is 27 billion transistors with 4x8GB stacks of HBM2-2400 memory, 24 Tensor Processing Clusters (TPCs) with a core frequency of up to 1.1GHz. Sixty-four lanes of SerDes HSIO provides 3.58Tbps of aggregate bandwidth and the card supports an x16 PCIe 4.0 connection. Power consumption is expected to be between 150-250W. The chip was built using TSMC’s advanced CoWoS packaging (Chip-on-Wafer-on-Substrate), and carries 60MB of cache distributed across its various cores. CoWoS competes with Intel’s EMIB, but Intel has decided to build this hardware at TSMC rather than using its own foundries. Performance is estimated at up to 119 TOPS.

“We don’t want to waste die area on things we don’t need,” Intel VP of Hardware Carey Kloss told Next Platform. “Our instruction set is simple; matrix multiply, linear algebra, convolutions. We don’t have registers per se, everything is a tensor (2D, 3D, or 4D).” There is a lot that is defined in software, including the ability to program the same when breaking a model to run on or off die. “Think of it as a hierarchy,” Kloss said in the interview. “You can use the same set of instructions to move data between two clusters in one group next to one HBM or between groups or even die in a network. We want to make it simple for software to manage the communication.”

The slideshow below steps through the NNP-T architecture. All data is courtesy of Intel, and the performance figures shared in the company’s microbenchmarks have obviously not been validated by ExtremeTech.

The NNP-T is designed to scale outwards effectively without requiring a chassis. Multiple NNP-T accelerators can be connected together in the same chassis, and the cards support chassis-to-chassis and even rack-to-rack glueless connection without needing a switch. There are four QFSP (Quad Small Form Factor Pluggable) network ports on the back of each mezzanine card.

We don’t have performance data yet, but this is the high-end training card Intel will come to market with to compete against the likes of Nvidia. It’s not yet clear how eventual solutions like Xe, which won’t ship for data centers until 2021, will fit into the company’s future product portfolio once it has both tensor processing cores and GPUs in the data center market.

Spring Hill / NNP-I: Icelake On-Board

Spring Hill, Intel’s new inference accelerator, is an entirely different beast. Where the NNP-T is designed for 150-250W power envelopes, the NNP-I is a 10-50W part intended to plug into an M.2 slot. It features two Icelake CPU cores paired with 12 Inference Compute Engines (ICE).

The 12 ICE engines and dual CPU cores are backed up by 24MB of coherent L3 and support both AVX-512 and VNNI instructions. There’s two on-die LPDDR4X memory controllers connected to an on-die pool of LPDDR4 memory (no word on capacity yet). DRAM bandwidth is up to 68GB/s, but total amount of on-card DRAM is unknown. Spring Hill can be added to any modern server that supports M.2 slots — according to Intel, the device communicates over the M.2 riser like a PCIe product rather than via NVMe.

The goal, with NNP-I, is to run operations on the AI processor with less overhead required from the primary CPU in the system. The device connects via PCIe (both PCIe 3.0 and 4.0 are supported) and handles the AI workload, using the on-die Icelake cores for any necessary processing. The on-die SRAMs and DRAM provide local memory bandwidth.

The Inference Compute Engine supports various instruction formats, ranging from FP16 to INT1, with a programmable vector processor and a 4MB SRAM for each individual ICE.

There’s also a tensor engine, dubbed the Deep Learning Compute Grid, and a Tensilica Vision P6 DSP (used for processing workloads that aren’t tuned for running in the fixed-function DL Compute Grid).

The overall memory subsystem of the NNP-I is also optimized, with the L3 cache broken into eight 3MB slices, shared between the ICE and CPU cores. The goal is to keep data as near to the processing elements that need it as possible. Intel claims the NNP-I can deliver ResNet50 performance of 3,600 inferences per second when running at a 10W TDP. That works out to 4.8 TOPS/watt, which meets Intel’s overall efficiency goals (the company claims that NNP-I is most efficient at lower wattages).

Intel doesn’t expect the NNP-I to come to the retail market, but inference solutions are doing a brisk business compared with the high-end data center-centric training solutions. The NNP-I could ship to a wide range of customers in the not-too-distant future, depending on overall uptake.

Both of these solutions are intended to challenge Nvidia in the data center. While they’re both quite different from Xeon Phi, you could argue that they collectively target some of the spaces Intel wanted to sell Xeon Phi into, albeit in very different ways. That’s not necessarily a bad thing, however — when the original Larrabee was built, the idea of using GPUs for AI and data center work was a distant concept. Revisiting the topic with a new specialized architecture for both inference and training is a smart move for Intel, if the company can grab volume away from Nvidia.

Now Read:




10 minutes mail – Also known by names like : 10minemail, 10minutemail, 10mins email, mail 10 minutes, 10 minute e-mail, 10min mail, 10minute email or 10 minute temporary email. 10 minute email address is a disposable temporary email that self-destructed after a 10 minutes. https://tempemail.co/– is most advanced throwaway email service that helps you avoid spam and stay safe. Try tempemail and you can view content, post comments or download something

Cerebras Systems Unveils 1.2 Trillion Transistor Wafer-Scale Processor for AI


This site may earn affiliate commissions from the links on this page. Terms of use.

Credit: Getty Images

Modern CPU transistor counts are enormous — AMD announced earlier this month that a full implementation of its 7nm Epyc “Rome” CPU weighs in at 32 billion transistors. To this, Cerebras Technology says: “Hold my beer.” The AI-focused company has designed what it calls a Wafer Scale Engine. The WSE is a square, approximately eight inches by nine inches, and contains roughly 1.2 trillion transistors.

I’m genuinely surprised to see a company bringing a wafer-scale product to market this quickly. The idea of wafer-scale processing has attracted some attention recently as a potential solution to performance scaling difficulties. In the study we discussed earlier this year, researchers evaluated the idea of building an enormous GPU across most or all of a 100mm wafer. They found that the technique could product viable, high-performance processors and that it could also scale effectively to larger node sizes. The Cerebras WSE definitely qualifies as lorge large — its total surface area is much larger than the hypothetical designs we considered earlier this year. It’s not a full-sized 300mm wafer, but it’s got a higher surface area than a 200mm does.

The largest GPU,SEEAMAZON_ET_135 See Amazon ET commerce just for comparison, measures 815 square millimeters and packs 21.1B transistors. So the Cerebras WSE is just a bit bigger, as these things go. Some companies send out pictures of their chips held up next to a diminutive common object, like a quarter. Cerebras sent out a photo of their die next to a keyboard.

cerebras-1-100808712-large

Not Pictured: PCIe x1600 slot.

As you can see, it compares fairly well.

The Cerebras WSE contains 400,000 sparse linear algebra cores, 18GB of total on-die memory, 9PB/sec worth of memory bandwidth across the chip, and separate fabric bandwidth of up to 100Pbit/sec. The entire chip is built on TSMC’s 16nm FinFET process. Because the chip is built from (most) of a single wafer, the company has implemented methods of routing around bad cores on-die and can keep its arrays connected even if it has bad cores in a section of the wafer. The company says it has redundant cores implemented on-die, though it hasn’t discussed specifics yet. Details on the design are being presented at Hot Chips this week.

The WSE — “CPU” simply doesn’t seem sufficient — is cooled using a massive cold plate sitting above the silicon, with vertically mounted water pipes used for direct cooling. Because there’s no traditional package large enough to fit the chip, Cerebras has designed its own. PCWorld describes it as “combining a PCB, the wafer, a custom connector linking the two, and the cold plate.” Details on the chip, like its raw performance and power consumption, are not yet available.

A fully functional wafer-scale processor, commercialized at scale, would be an exciting demonstration of whether this technological approach has any relevance to the wider market. While we’re never going to see consumer components sold this way, there’s been interest in using wafer-scale processing to improve performance and power consumption in a range of markets. If consumers continue to move workloads to the cloud, especially high-performance workloads like gaming, it’s not crazy to think we might one day see GPU manufacturers taking advantage of this idea — and building arrays of parts that no individual could ever afford to power cloud gaming systems in the future.

Now Read:




10 minutes mail – Also known by names like : 10minemail, 10minutemail, 10mins email, mail 10 minutes, 10 minute e-mail, 10min mail, 10minute email or 10 minute temporary email. 10 minute email address is a disposable temporary email that self-destructed after a 10 minutes. https://tempemail.co/– is most advanced throwaway email service that helps you avoid spam and stay safe. Try tempemail and you can view content, post comments or download something

Google discloses its acquisition of mobile learning app Socratic as it relaunches on iOS – gpgmail


Google publicly disclosed its acquisition of homework helper app Socratic in an announcement this week, detailing the added support for the company’s A.I. technology and its relaunch on iOS. The acquisition apparently flew under the radar — Google says it bought the app last year.

According to one founder’s LinkedIn update, that was in March 2018. Google hasn’t responded to requests for comment for more details about the deal, but we’ll update if that changes.

Socratic was founded in 2013 Chris Pedregal and Shreyans Bhansali with the goal of creating a community that made learning accessible to all students.

Initially, the app offered a Quora-like Q&A platform where students could ask questions which were answered by experts. By the time Socratic raised $6 million in Series A funding back in 2015, its community had grown to around 500,000 students. The company later evolved to focus less on connecting users and more on utility.

It included a feature to take a photo of a homework question in order to get instant explanations through the mobile app launched in 2015. This is similar to many other apps in the space, like Photomath, Mathway, DoYourMath, and others.

However, Socratic isn’t just a math helper — it can also tackle subjects like science, literature, social studies, and more.

In February 2018, Socratic announced it would remove the app’s social features. That June, the company said it was closing its Q&A website to user contributions. This decision was met with some backlash of disappointed users.

Socratic explained the app and website were different products, and it was strategically choosing to focus on the former.

“We, as anyone, are bound by the constraints of reality—you just can’t do everything—which means making decisions and tradeoffs where necessary. This one is particularly painful,” wrote Community Lead Becca McArthur at the time.

That strategy, apparently, was to make Socratic a Google A.I.-powered product. According to Google’s blog post penned by Bhansali — now the Engineering Manager at Socratic — the updated iOS app uses A.I. technology to help users.

The new version of the iOS app still allows you to snap a photo to get answers, or you can speak your question.

For example, if a student takes a photo from a classroom handout or asks a question like “what’s the difference between distance and displacement?,” Socratic will return a top match, followed by explainers, a Q&A section, and even related YouTube videos and web links. It’s almost like a custom search engine just for your homework questions.

Google also says it has built and trained algorithms that can analyze the student’s question then identify the underlying concepts in order to point users to these resources. For students who need even more help, the app can break down the concepts into smaller, easy-to-understand lessons.

googleai v2

In addition, the app includes subject guides on over 1,000 higher education and high school topics, developed with help from educators. The study guides can help students prepare for tests or just better learn a particular concept.

explorer 56kHA30

“In building educational resources for teachers and students, we’ve spent a lot of time talking to them about challenges they face and how we can help,” writes Bhansali. “We’ve heard that students often get ‘stuck’ while studying. When they have questions in the classroom, a teacher can quickly clarify—but it’s frustrating for students who spend hours trying to find answers while studying on their own,” he says.

This is where Socratic will help.

That said, the acquisition could help Google in other ways, too. In addition to its primary focus as a homework helper, the acquisition could aid Google Assistant technology across platforms, as the virtual assistant could learn to answer more complex questions that Google’s Knowledge Graph didn’t already include.

The relaunched, A.I.-powered version of Socratic by Google arrived on Thursday on iOS, where it also discloses through the app update text the app is now owned by Google.

The Android version of the app will launch this fall.

 


10 minutes mail – Also known by names like : 10minemail, 10minutemail, 10mins email, mail 10 minutes, 10 minute e-mail, 10min mail, 10minute email or 10 minute temporary email. 10 minute email address is a disposable temporary email that self-destructed after a 10 minutes. https://tempemail.co/– is most advanced throwaway email service that helps you avoid spam and stay safe. Try tempemail and you can view content, post comments or download something

Intel Announces Cooper Lake Will Be Socketed, Compatible With Future Ice Lake CPUs


This site may earn affiliate commissions from the links on this page. Terms of use.

Intel may have launched Cascade Lake relatively recently, but there’s another 14nm server refresh already on the horizon. Intel lifted the lid on Cooper Lake today, giving some new details on how the CPU fits into its product lineup with Ice Lake 10nm server chips already supposedly queuing up for 2020 deployment.

Cooper Lake’s features include support for the Google-developed bfloat16 format. It will also support up to 56 CPU cores in a socketed format, unlike Cascade Lake-AP, which scales up to 56 cores but only in a soldered, BGA configuration. The new socket will reportedly be known as LGA4189. There are reports that these chips could offer up to 16 memory channels (because Cascade Lake-AP and Cooper Lake both use multiple dies on the same chip, the implication is that Intel may launch up to 16 memory channels per socket with the dual-die version).

bfloat16-vs-float16

The bfloat16 support is a major addition to Intel’s AI efforts. While 16-bit half-precision floating point numbers have been defined in the IEEE 754 standard for over 30 years, bfloat16 changes the balance between how much of the format is used for significant digits and how much is devoted to exponents. The original IEEE 754 standard is designed to prioritize precision, with just five exponent bits. The new format allows for a much greater range of values but at lower precision. This is particularly valuable for AI and deep learning calculations, and is a major step on Intel’s path to improving the performance of AI and deep learning calculations on CPUs. Intel has published a whitepaper on bfloat16 if you’re looking for more information on the topic. Google claims that using bfloat16 instead of conventional half-precision floating point can yield significant performance advantages. The company writes: “Some operations are memory-bandwidth-bound, which means the memory bandwidth determines the time spent in such operations. Storing inputs and outputs of memory-bandwidth-bound operations in the bfloat16 format reduces the amount of data that must be transferred, thus improving the speed of the operations.”

The other advantage of Cooper Lake is that the CPU will reportedly share a socket with Ice Lake servers coming in 2020. One major theorized distinction between the two families is that Ice Lake servers on 10nm may not support bfloat16, while 14nm Cooper Lake servers will. This could be the result of increased differentiation in Intel’s product lines, though it’s also possible that it reflects 10nm’s troubled development.

Bringing 56 cores to market in a socketed form factor indicates Intel expects Cooper Lake to expand to more customers than Cascade Lake / Cascade Lake-AP targeted. It also raises questions about what kind of Ice Lake servers Intel will bring to market, and whether we’ll see 56-core versions of these chips as well. To-date, all of Intel’s messaging around 10nm Ice Lake has focused on servers or mobile. This may mirror the strategy Intel used for Broadwell, where the desktop versions of the CPU were few and far between, and the mobile and server parts dominated that family — but Intel also said later that not doing a Broadwell desktop release was a mistake and that the company had goofed by skipping the market. Whether that means Intel is keeping an Ice Lake desktop launch under its hat or if the company has decided skipping desktop again does make sense this time around is still unclear.

Cooper Lake’s focus on AI processing implies that it isn’t necessarily intended to go toe-to-toe with AMD’s upcoming 7nm Epyc. AMD hasn’t said much about AI or machine learning workloads on its processors, and while its 7nm chips add support for 256-bit AVX2 operations, we haven’t heard anything from the CPU division at the company to imply a specific focus on the AI market. AMD’s efforts in this space are still GPU-based, and while its CPUs will certainly run AI code, it doesn’t seem to be gunning for the market the way Intel has. Between adding new support for AI to existing Xeons, its Movidius and Nervana products, projects like Loihi, and plans for the data center market with Xe, Intel is trying to build a market for itself to protect its HPC and high-end server business — and to tackle Nvidia’s own current dominance of the space.

Now Read:




10 minutes mail – Also known by names like : 10minemail, 10minutemail, 10mins email, mail 10 minutes, 10 minute e-mail, 10min mail, 10minute email or 10 minute temporary email. 10 minute email address is a disposable temporary email that self-destructed after a 10 minutes. https://tempemail.co/– is most advanced throwaway email service that helps you avoid spam and stay safe. Try tempemail and you can view content, post comments or download something