| From: Frank Sully | 10/11/2021 8:02:38 AM | | | | | | Qualcomm snaps up self-driving software provider to poach business from Nvidia
The semiconductor company seals a deal to acquire part of Veoneer, beating out auto parts supplier Magna.
BY CHRISTIAAN HETZNER
October 04, 2021
Chipmaker Qualcomm took its boldest step yet in the race to develop self-driving technology, swooping in to snag a prized asset out from under the nose of a major auto parts supplier.
Qualcomm and New York–based private equity firm SSW Partners agreed to acquire Veoneer in a deal that values the Swedish componentry manufacturer’s shares at $4.5 billion in total, an 18% premium over the July offer from rival North American bidder Magna.
Best known for its dominance in smartphone chips, Qualcomm aims to carve out the prime piece for itself: Veoneer’s self-driving software development unit, Arriver, as part of a plan to best rivals like Nvidia. The leftovers will go to SSW Partners, which intends to sell off the remaining assets over time to competitors.
fortune.com |
| | NVIDIA Corporation (NVDA) | Stock Discussion ForumsShare | RecommendKeepReplyMark as Last Read |
|
| From: Frank Sully | 10/11/2021 12:48:10 PM | | | | | | Microsoft and Nvidia team up to train one of the world’s largest language models
Kyle Wiggers @Kyle_L_Wiggers
October 11, 2021
Image Credit: raindrop74 / Shutterstock
Microsoft and Nvidia today announced that they trained what they claim is the largest and most capable AI-powered language model to date: Megatron-Turing Natural Language Generation (MT-NLP). The successor to the companies’ Turing NLG 17B and Megatron-LM models, MT-NLP contains 530 billion parameters and achieves “unmatched” accuracy in a broad set of natural language tasks, Microsoft and Nvidia say — including reading comprehension, commonsense reasoning, and natural language inferences.
“The quality and results that we have obtained today are a big step forward in the journey towards unlocking the full promise of AI in natural language. The innovations of DeepSpeed and Megatron-LM will benefit existing and future AI model development and make large AI models cheaper and faster to train,” Nvidia’s Paresh Kharya and Microsoft’s Ali Alvi wrote in a blog post. “We look forward to how MT-NLG will shape tomorrow’s products and motivate the community to push the boundaries of natural language processing (NLP) even further. The journey is long and far from complete, but we are excited by what is possible and what lies ahead.”
Training massive language models
In machine learning, parameters are the part of the model that’s learned from historical training data. Generally speaking, in the language domain, the correlation between the number of parameters and sophistication has held up remarkably well. Language models with large numbers of parameters, more data, and more training time have been shown to acquire a richer, more nuanced understanding of language, for example gaining the ability to summarize books and even complete programming code.
To train MT-NLG, Microsoft and Nvidia say that they created a training dataset with 270 billions of tokens from English-language websites. Tokens, a way of separating pieces of text into smaller units in natural language, can be either words, characters, or parts of words. Like all AI models, MT-NLP had to “train” by ingesting a set of examples to learn patterns among data points, like grammatical and syntactical rules.
The dataset largely came from The Pile, a 835GB collection of 22 smaller datasets created by the open source AI research effort EleutherAI. The Pile spans academic sources (e.g., Arxiv, PubMed), communities (StackExchange, Wikipedia), code repositories (Github), and more, which Microsoft and Nvidia say they curated and combined with filtered snapshots of the Common Crawl, a large collection of webpages including news stories and social media posts.
Above: The data used to train MT-NLP.
Training took place across 560 Nvidia DGX A100 servers, each containing 8 Nvidia A100 80GB GPUs.
When benchmarked, Microsoft says that MT-NLP can infer basic mathematical operations even when the symbols are “badly obfuscated.” While not extremely accurate, the model seems to go beyond memorization for arithmetic and manages to complete tasks containing questions that prompt it for an answer, a major challenge in NLP.
It’s well-established that models like MT-NLP can amplify the biases in data on which they were trained, and indeed, Microsoft and Nvidia acknowledge that the model “picks up stereotypes and biases from the data.” That’s likely because a portion of the dataset was sourced from communities with pervasive gender, race, physical, and religious prejudices, which curation can’t completely address.
In a paper, the Middlebury Institute of International Studies’ Center on Terrorism, Extremism, and Counterterrorism claims that GPT-3 and similar models can generate “informational” and “influential” text that might radicalize people into far-right extremist ideologies and behaviors. A group at Georgetown University has used GPT-3 to generate misinformation, including stories around a false narrative, articles altered to push a bogus perspective, and tweets riffing on particular points of disinformation. Other studies, like one published by Intel, MIT, and Canadian AI initiative CIFAR researchers in April, have found high levels of stereotypical bias from some of the most popular open source models, including Google’s BERT and XLNet and Facebook’s RoBERTa.
Microsoft and Nvidia claim that they’re “committed to working on addressing [the] problem” and encourage “continued research to help in quantifying the bias of the model.” They also say that any use of Megatron-Turing in production “must ensure that proper measures are put in place to mitigate and minimize potential harm to users” and follow tenets such as those outlined in Microsoft’s Responsible AI Principles.
“We live in a time where AI advancements are far outpacing Moore’s law. We continue to see more computation power being made available with newer generations of GPUs, interconnected at lightning speeds. At the same time, we continue to see hyperscaling of AI models leading to better performance, with seemingly no end in sight,” Kharya and Alvi continued. “Marrying these two trends together are software innovations that push the boundaries of optimization and efficiency.”
The cost of large models
Projects like MT-NLP, AI21 Labs’ Jurassic-1, Huawei’s PanGu-Alpha, Naver’s HyperCLOVA, and the Beijing Academy of Artificial Intelligence’s Wu Dao 2.0are impressive from an academic standpoint, but building them doesn’t come cheap. For example, the training dataset for OpenAI’s GPT-3 — one of the world’s largest language models — was 45 terabytes in size, enough to fill 90 500GB hard drives.
AI training costs dropped 100-fold between 2017 and 2019, according to one source, but the totals still exceed the compute budgets of most startups. The inequity favors corporations with extraordinary access to resources at the expense of small-time entrepreneurs, cementing incumbent advantages.
For example, OpenAI’s GPT-3 required an estimated 3.1423 floating point operations per second (FLOPS) of compute during training. In computer science, FLOPS are a measure of raw processing performance typically used to compare different types of hardware. Assuming OpenAI reserved 28 teraflops — 28 trillion floating point operations per second — of compute across a bank of Nvidia V100 GPUs, a common GPU available through cloud services, it’d take $4.6 million for a single training run. One Nvidia RTX 8000 GPU with 15 teraflops of compute would be substantially cheaper — but it’d take 665 years to finish the training.
A Synced report estimated that a fake news detection model developed by researchers at the University of Washington cost $25,000 to train, and Google spent around $6,912 to train a language model called BERT that it used to improve the quality of Google Search results. Storage costs also quickly mount when dealing with datasets at the terabyte — or petabyte — scale. To take an extreme example, one of the datasets accumulated by Tesla’s self-driving team — 1.5 petabytes of video footage — would cost over $67,500 to store in Azure for three months, according to CrowdStorage.
The effects of AI and machine learning model training on the environment have also been brought into relief. In June 2020, researchers at the University of Massachusetts at Amherst released a report estimating that the amount of power required for training and searching a certain model involves the emissions of roughly 626,000 pounds of carbon dioxide, equivalent to nearly 5 times the lifetime emissions of the average U.S. car. OpenAI itself has conceded that models like Codex require significant amounts of compute — on the order of hundreds of petaflops per day — which contributes to carbon emissions.
In a sliver of good news, the cost for FLOPS and basic machine learning operations have been falling over the past few years. A 2020 OpenAI surveyfound that since 2012, the amount of compute needed to train a model to the same performance on classifying images in a popular benchmark — ImageNet — has been decreasing by a factor of two every 16 months. Other recent research suggests that large language models aren’t always more complex than smaller models, depending on the techniques used to train them.
Maria Antoniak, a natural language processing researcher and data scientist at Cornell University, says when it comes to natural language, it’s an open question whether larger models are the right approach. While some of the best benchmark performance scores today come from large datasets and models, the payoff from dumping enormous amounts of data into models is uncertain.
“The current structure of the field is task-focused, where the community gathers together to try to solve specific problems on specific datasets,” Antoniak told VentureBeat in a previous interview. “These tasks are usually very structured and can have their own weaknesses, so while they help our field move forward in some ways, they can also constrain us. Large models perform well on these tasks, but whether these tasks can ultimately lead us to any true language understanding is up for debate.”
venturebeat.com |
| | NVIDIA Corporation (NVDA) | Stock Discussion ForumsShare | RecommendKeepReplyMark as Last Read |
|
| From: Frank Sully | 10/11/2021 9:31:46 PM | | | | | | Telstra-UOW Hub to deliver Australian AIOT solutions
October 12, 2021
Image credit: The University of Wollongong.
The University of Wollongong (UOW) and technology giants Telstra, Microsoft and NVIDIA are uniting to position the Illawarra as a global leader in the uptake and creation of the Artificial Intelligence of Things (AIOT) via a Telstra-UOW Hub. This will help Australia access a world market estimated to reach $35.8 billion by 2023.
The university’s SMART Infrastructure Facility will establish the Telstra-UOW Hub for AIOT Solutions to deliver innovative, cost-effective and fit-for-purpose AIOT solutions for communities, enterprises and governments.
Telstra looks forward to working with the other partners to help develop pilot projects for local customers, Telstra group executive of Product and Technology Kim Krogh Andersen said.
“Combined with Telstra 5G, the potential applications of these transformative technologies including edge compute, Digital Twins, IoT, AI, ComputerVision and drones are profound,” Anderson said.
“That’s why we’re excited to put them into the hands of some of the smartest people across different industries, while helping them create the best environment to build and test solutions. I’d like to thank Dr Iain Russell, who led the engagement into the Illawarra community and our key partners to help establish the consortium.”
Under the federal government’s Strategic University Reform Fund (SURF), the Telstra-UOW Hub will receive $1.7 million to give Australia international traction and use applied research to increase productivity and create jobs.
Its focus will be research in several AIOT technologies including AI-enabled video analytics, sensor-embedded edge computing for optimised data fusion and transmission, as well as industrial digital twin technology.
Once established, the Telstra-UOW Hub will strive to improve the adoption of AIOT technologies by Illawarra industries, attracting and retaining a skilled workforce in AI and IoT technologies. It will develop a regional ecosystem of SMEs focused on AIOT products and services and ensure greater sovereignty in Australia, in sensitive sectors such as AI and IoT technologies.
“Our pioneering work in accelerated computing to solve problems normal computers cannot solve will benefit both the research and production capabilities of the projects being targeted by the Telstra-UOW Hub,” NVIDIA country manager – Enterprise Sudarshan Ramachandran said.
“Simultaneously, the work undertaken at the hub ties into our strategies for such applications as ethical AI, secure end-to-end communications and edge computing to build smarter, secure and safer cities.”
Four initial pilot projects have been identified for immediate design and testing, with ten projects expected to be supported during the initial two years of the collaboration. The AIOT Solutions will be supported by cellular LPWAN (NB-IOT) or LTE/5G technologies, as well as cloud computing such as Microsoft Azure.
“We see the UOW, through this hub, taking an innovative lead for the Illawarra region, the community and its members,” Microsoft spokesperson Steven Miller said.
“Combining Microsoft’s Azure ecosystem and Telstra’s 5G network will enable UOW to build high impact end-to-end digital solutions, and we look forward to this exciting partnership.”
The four projects will focus on intelligent manufacturing, smart transport, smart logistics and resilient infrastructure. They will be tested at scale with the support of key end-users such as Bluescope Steel, Premier Illawarra, Remondis, and Wollongong City Council.
The Telstra-UOW Hub will connect a task force of 30 AIOT experts from industry and academia, UOW vice-chancellor Professor Patricia M. Davidson said.
“At the UOW, we believe it is important for our academic leaders, researchers and developers to work with industry and community groups to deliver practical solutions with long-term benefits,” Davidson said.
“Our mission is to help improve society through research, collaboration and innovation, and we have a long history of being at the forefront of deploying technology within the Illawarra.
“The Telstra-UOW Hub has the potential to create AIOT solutions with worldwide applications that will be refined and tested locally. By working with Telstra, Microsoft and Nvidia, we can position the region as a leader in the uptake and creation of AIOT-enabled solutions,” she said.
The hub will provide the Illawarra and Australia with a unique chance to move from early adopters to leading innovators in the AIOT space, according to SMART Infrastructure Facility director Professor Pascal Perez said.
“This is an incredible opportunity and I would like to thank Dr Johan Barthelemy, who leads SMART’s IoT Hub and Digital Living Lab, for driving the collaboration with our technology industry partners,” Perez said.
“It is through this collaboration and the support of regional enterprise that this operation will succeed.”
pacetoday.com.au |
|
| | NVIDIA Corporation (NVDA) | Stock Discussion ForumsShare | RecommendKeepReplyMark as Last Read |
|
| From: Frank Sully | 10/11/2021 9:51:59 PM | | | | | | By 2028, the Artificial Intelligence Chip Market will be worth $ 109.83 Billion, at a CAGR of 40.17%, according to Verified Market Research
BY CLAIRE REID
OCTOBER 11, 2021
Verified Market Research published a study titled "JERSEY CITY, N.J.", Oct. 11, 2021 /PRNewswire/ -- Verify Market research published " "A Report on the Verification of Market Size and Growth in JERSEE CITIES," released yesterday. Artificial Intelligence Chip Market is a rapidly growing industry. By End-User (Healthcare, Manufacturing, Automotive, Retail), By Technology (Machine Learning, Predictive Analysis, Natural Language Processing), and By Geography. According to Verified Market Research, the Global Artificial Intelligence Chip Market was valued at USD 7.37 billion in 2020 and is projected to reach USD 109.83 billion by 2028, growing at a CAGR of 40.17% from 2021 to 2027.
Global Artificial Intelligence Chip Market Overview
The AI investment landscape experienced another year of steady growth in 2019, with the US taking the lead, with a US$22.7 billion record. Despite many setbacks and a shift in interest and priorities that can result in uncertainty, the market did not suffer any fatigue. Enterprises are increasingly waking up to the AI value. Not only are businesses actively adopting automation to automate repetitive tasks, ensure compliance, and improve customer experience, but they are also partnering with machine learning platforms and acquiring AI startups and talent to build data pipelines, develop proprietary AI models,, as well as manage their machine-learning development and operation lifecycle.
Different key players have been working on developing a separate platform; Mythic's platform has the advantage of processing digital/analog calculations in memory, resulting in enhanced performance, accuracy, and power life. Furthermore, the rise in demand to integrate video surveillance & AI and the increase in government expenditure for cybersecurity solutions integrated with real-time analytics e'AI are expected to boost the artificial intelligence chip's growth. The rise in investments in AI firms is what drives the global AI Chips Market expansion.
However, there are relatively higher prices for AI chips and a slumber of skilled workers with the ability to use AI-based systems, particularly in developing countries. These factors may adversely affect the global AI Chips Market growth.
Key Developments in the Field of Key Advancements in Research Artificial Intelligence Chip Market (AIICP Market) is a growing market for Artificial intelligence chips.
Kneron, a world-renowned full-stack edge AI solutions provider, will launch its advanced AI chipset, the "Knerson KL 720 SoC", in August 2020. The goal is to provide a comprehensive and cost-effective AI chipsets suite for devices around the world.
Nvidia Corporation, a global corporation that manufactures graphics processors, mobile technologies, and desktop computers, will introduce the EGX Jetson Xavier NX and PGX A100 as part of its upcoming VGX Edge AI platform in May 2020. The goal is to provide secure AI processing and high-performance at the edge.
In September 2019, Alibaba Group Holding Limited introduced a new AI-based chipset, "Hanguang 800", which provides advanced computing capabilities on the cloud. This chip can accelerate machine learning tasks and improve the customer experience, resulting in increased customer satisfaction.
In September 2019, Apple Inc. released its A11, A12, and A13 Bionic Chips for high-performance processors that use core CPUs coupled to GPUs as accelerators.
https://www.google.com/amp/s/list23.com/amp/229649-by-2028-the-artificial-intelligence-chip-market-will-be-worth-109-83-billion-at-a-cagr-of-40-17/
|
| | NVIDIA Corporation (NVDA) | Stock Discussion ForumsShare | RecommendKeepReplyMark as Last Read |
|
| From: Frank Sully | 10/11/2021 10:27:20 PM | | | | | | What The Experts Say On NVIDIA: October 10, 2021
- B of A Securities has decided to maintain their Buy rating on NVIDIA, which currently sits at a price target of $275.0.
- Susquehanna has decided to maintain their Positive rating on NVIDIA, which currently sits at a price target of $250.0.
- Keybanc has decided to maintain their Overweight rating on NVIDIA, which currently sits at a price target of $260.0.
|
| | NVIDIA Corporation (NVDA) | Stock Discussion ForumsShare | RecommendKeepReplyMark as Last Read |
|
| From: Frank Sully | 10/13/2021 7:26:00 PM | | | | | | Google’s DeepMind is one of the world’s premier AI laboratories.
A DeepMind AI AlphaGo beat Go master player Lee Sedol by a decisive score of 4 to 1 in a 5 game match in 2016 and stunned the world. In particular, it awakened China from an AI slumber and prompted a series of 5 year AI plans leading to Chinese AI world dominance in 2030.
Last year DeepMind stunned the world again by solving protein folding, said to be the most significant AI achievement ever. See message below:
Message 33514697
Last year DeepMind turned another corner and posted it’s first ever profit.
This year DeepMind announced that it was turning to robotics AI. It just released details of it’s first effort, a shape-stacking robotic AI, which is perhaps on the intelligence order of a kindergarten student. But ya gotta start somewhere! See message below, including a six and a half minute video.
Message 33529245 |
| | NVIDIA Corporation (NVDA) | Stock Discussion ForumsShare | RecommendKeepReplyMark as Last Read |
|
| From: Frank Sully | 10/13/2021 8:24:10 PM | | | | | | Supermicro Expands GPU System Portfolio for AI, HPC, Cloud Workloads
October 13, 2021 by staff
SAN JOSE, Oct. 12, 2021 — Supermicro has announced new systems based on NVIDIA Ampere architecture GPUs and 3rd Gen Intel Xeon Scalable processors with built-in AI accelerators (Supermicro X12 series).
These servers are designed for demanding AI applications where low latency and high application performance are essential. The 2U NVIDIA HGX A100 4-GPU system is suited for deploying modern AI training clusters at scale with high-speed CPU-GPU and GPU-GPU interconnect. The Supermicro 2U 2-Node system reduces energy usage and costs by sharing power supplies and cooling fans, reducing carbon emissions, and supports a range of discrete GPU accelerators, which can be matched to the workload. Both of these systems include advanced hardware security features that are enabled by the latest Intel Software Guard Extensions (Intel SGX).
“ Supermicro engineers have created another extensive portfolio of high-performance GPU-based systems that reduce costs, space, and power consumption compared to other designs in the market,” said Charles Liang, president and CEO, Supermicro. “With our innovative design, we can offer customers NVIDIA HGX A100 (code name Redstone) 4-GPU accelerators for AI and HPC workloads in dense 2U form factors. Also, our 2U 2-Node system is uniquely designed to share power and cooling components which reduce OPEX and the impact on the environment.”
NVIDIA HGX A100 server is based on the 3rd Gen Intel Xeon Scalable processors with Intel Deep Learning Boost technology and is optimized for analytics, training, and inference workloads. The system can deliver up to 2.5 petaflops of AI performance, with four A100 GPUs fully interconnected with NVIDIA NVLink, providing up to 320GB of GPU memory to speed breakthroughs in enterprise data science and AI. The system is up to 4x faster than the previous generation GPUs for complex conversational AI models like BERT large inference and delivers up to 3x performance boost for BERT large AI training.
In addition, the advanced thermal and cooling designs make these systems ideal for high-performance clusters where node density and power efficiency are priorities. Liquid cooling is also available for these systems, resulting in even more OPEX savings. Intel Optane Persistent Memory (PMem) is also supported on this platform, enabling significantly larger models to be held in memory, close to the CPU, before processing on the GPUs. For applications that require multi-system interaction, the system can also be equipped with four NVIDIA ConnectX-6 200Gb/s InfiniBand cards to support GPUDirect RDMA with a 1:1 GPU-to-DPU ratio.
The new 2U 2-Node is an energy-efficient resource-saving architecture designed for each node to support up to three double-width GPUs. Each node also features a single 3rd Gen Intel Xeon Scalable processor with up to 40 cores and built-in AI and HPC acceleration. A wide range of AI, rendering, and VDI applications will benefit from this balance of CPUs and GPUs. Equipped with Supermicro’s advanced I/O Module (AIOM) expansion slots for fast and flexible networking capabilities, the system can also process massive data flow for demanding AI/ML applications, deep learning training, and inferencing while securing the workload and learning models. It is also ideal for multi-instance high-end cloud gaming and many other compute-intensive VDI applications. In addition, Virtual Content Delivery Networks (vCDNs) will be able to satisfy increasing demands for streaming services. Power supply redundancy is built-in, as either node can use the adjacent node’s power supply in the event of a failure.
insidehpc.com |
| | NVIDIA Corporation (NVDA) | Stock Discussion ForumsShare | RecommendKeepReplyMark as Last Read |
|
| From: Frank Sully | 10/14/2021 6:34:33 PM | | | | | | TSMC is building a new chip factory in Japan, but warns of ‘tight’ supply through 2022
The factory would focus on producing the older chips that have the biggest supply issues
By Chaim Oct 14, 2021
/cdn.vox-cdn.com/uploads/chorus_image/image/69993593/1235442658.0.jpg)
Photo by Walid Berrazeg/SOPA Images/LightRocket via Getty Images
TSMC has announced plans to expand its chipmaking efforts with a new factory in Japan during its Q3 2021 earnings call, marking the latest expansion for what’s already the world’s largest chipmaker, via Reuters. But the news comes as the company also cautions that supply will be “tight” for chips throughout 2022 as a result on the ongoing shortage.The new plant is planned to focus on producing chips with older technologies, instead of the bleeding-edge processors that TSMC is best known for providing to companies like Apple, AMD, Nvidia, and Qualcomm. That’s particularly important, given that it’s those older chips — like those found in cars for controlling the airbags and seatbelts or the power management chip in an iPhone — that are the ones having the biggest supply issues.
But it’ll still be a while before the new fab comes online, with production not expected to start until “late 2024.” TSMC is also still finalizing plans for the new factory, with CEO C.C. Wei noting on the company’s earnings call that the new Japan fab still needs approval from TSMC’s board.
TSMC had previously cautioned in April that shortages could last through 2022. Wei noted then: “In 2023, I hope we can offer more capacity to support our customers. At that time, we’ll start to see the supply chain tightness release a little bit.” Intel CEO Pat Gelsinger has also echoed concerns about 2022 supply, telling the BBC in July that it would be “a year or two” before shortages end. And while AMD CEO Lisa Su was more optimistic when speaking at the 2021 Code Conference, she too noted that supply for at least the first half of the year will be “likely tight.”
TSMC is also reportedly raising its prices for its semiconductor products by around 10 percent for its advanced chips, and by about 20 percent for its less advanced products (the sort of chips that the new Japan factory is meant to help increase supply of) to try to both reduce demand and help fund its planned investments.
And investing to expand its capacity to meet that skyrocketing demand — and help prevent future shortages — is a key priority for the TSMC. The company has already announced plans to invest $100 billion through 2023 to increase its manufacturing capacity, which includes its planned $12 billion manufacturing hub in Arizona and the new Japan fab.
theverge.com |
| | NVIDIA Corporation (NVDA) | Stock Discussion ForumsShare | RecommendKeepReplyMark as Last Read |
|
| |
