What is AI hardware? How GPUs and TPUs give artificial intelligence algorithms a boost

Were you unable to attend Transform 2022? Check out all of the summit sessions in our on-demand library now! Watch here.

Most computers and algorithms — including, at this point, many artificial intelligence (AI) applications — run on general-purpose circuits called central processing units or CPUs. Though, when some calculations are done often, computer scientists and electrical engineers design special circuits that can perform the same work faster or with more accuracy. Now that AI algorithms are becoming so common and essential, specialized circuits or chips are becoming more and more common and essential. 

The circuits are found in several forms and in different locations. Some offer faster creation of new AI models. They use multiple processing circuits in parallel to churn through millions, billions or even more data elements, searching for patterns and signals. These are used in the lab at the beginning of the process by AI scientists looking for the best algorithms to understand the data. 

Others are being deployed at the point where the model is being used. Some smartphones and home automation systems have specialized circuits that can speed up speech recognition or other common tasks. They run the model more efficiently at the place it is being used by offering faster calculations and lower power consumption. 

Scientists are also experimenting with newer designs for circuits. Some, for example, want to use analog electronics instead of the digital circuits that have dominated computers. These different forms may offer better accuracy, lower power consumption, faster training and more. 


MetaBeat 2022

MetaBeat will bring together thought leaders to give guidance on how metaverse technology will transform the way all industries communicate and do business on October 4 in San Francisco, CA.

Register Here

What are some examples of AI hardware? 

The simplest examples of AI hardware are the graphical processing units, or GPUs, that have been redeployed to handle machine learning (ML) chores. Many ML packages have been modified to take advantage of the extensive parallelism available inside the average GPU. The same hardware that renders scenes for games can also train ML models because in both cases there are many tasks that can be done at the same time. 

Some companies have taken this same approach and extended it to focus only on ML. These newer chips, sometimes called tensor processing units (TPUs), don’t try to serve both game display and learning algorithms. They are completely optimized for AI model development and deployment. 

There are also chips optimized for different parts of the machine learning pipeline. These may be better for creating the model because it can juggle large datasets — or, they may excel at applying the model to incoming data to see if the model can find an answer in them. These can be optimized to use lower power and fewer resources to make them easier to deploy in mobile phones or places where users will want to rely on AI but not to create new models. 

Additionally, there are basic CPUs that are starting to streamline their performance for ML workloads. Traditionally, many CPUs have focused on double-precision floating-point computations because they are used extensively in games and scientific research. Lately, some chips are emphasizing single-precision floating-point computations because they can be substantially faster. The newer chips are trading off precision for speed because scientists have found that the extra precision may not be valuable in some common machine learning tasks — they would rather have the speed.

In all these cases, many of the cloud providers are making it possible for users to spin up and shut down multiple instances of these specialized machines. Users don’t need to invest in buying their own and can just rent them when they are training a model. In some cases, deploying multiple machines can be significantly faster, making the cloud an efficient choice. 

How is AI hardware different from regular hardware? 

Many of the chips designed for accelerating artificial intelligence algorithms rely on the same basic arithmetic operations as regular chips. They add, subtract, multiply and divide as before. The biggest advantage they have is that they have many cores, often smaller, so they can process this data in parallel. 

The architects of these chips usually try to tune the channels for bringing the data in and out of the chip because the size and nature of the data flows are often quite different from general-purpose computing. Regular CPUs may process many more instructions and relatively fewer data. AI processing chips generally work with large data volumes. 

Some companies deliberately embed many very small processors in large memory arrays. Traditional computers separate the memory from the CPU; orchestrating the movement of data between the two is one of the biggest challenges for machine architects. Placing many small arithmetic units next to the memory speeds up calculations dramatically by eliminating much of the time and organization devoted to data movement. 

Some companies also focus on creating special processors for particular types of AI operations. The work of creating an AI model through training is much more computationally intensive and involves more data movement and communication. When the model is built, the need for analyzing new data elements is simpler. Some companies are creating special AI inference systems that work faster and more efficiently with existing models. 

Not all approaches rely on traditional arithmetic methods. Some developers are creating analog circuits that behave differently from the traditional digital circuits found in almost all CPUs. They hope to create even faster and denser chips by forgoing the digital approach and tapping into some of the raw behavior of electrical circuitry. 

What are some advantages of using AI hardware?

The main advantage is speed. It is not uncommon for some benchmarks to show that GPUs are more than 100 times or even 200 times faster than a CPU. Not all models and all algorithms, though, will speed up that much, and some benchmarks are only 10 to 20 times faster. A few algorithms aren’t much faster at all. 

One advantage that is growing more important is the power consumption. In the right combinations, GPUs and TPUs can use less electricity to produce the same result. While GPU and TPU cards are often big power consumers, they run so much faster that they can end up saving electricity. This is a big advantage when power costs are rising. They can also help companies produce “greener AI” by delivering the same results while using less electricity and consequently producing less CO2. 

The specialized circuits can also be helpful in mobile phones or other devices that must rely upon batteries or less copious sources of electricity. Some applications, for instance, rely upon fast AI hardware for very common tasks like waiting for the “wake word” used in speech recognition. 

Faster, local hardware can also eliminate the need to send data over the internet to a cloud. This can save bandwidth charges and electricity when the computation is done locally. 

What are some examples of how leading companies are approaching AI hardware?

The most common forms of specialized hardware for machine learning continue to come from the companies that manufacture graphical processing units. Nvidia and AMD create many of the leading GPUs on the market, and many of these are also used to accelerate ML. While many of these can accelerate many tasks like rendering computer games, some are starting to come with enhancements designed especially for AI. 

Nvidia, for example, adds a number of multiprecision operations that are useful for training ML models and calls these Tensor Cores. AMD is also adapting its GPUs for machine learning and calls this approach CDNA2. The use of AI will continue to drive these architectures for the foreseeable future. 

As mentioned earlier, Google makes its own hardware for accelerating ML, called Tensor Processing Units or TPUs. The company also delivers a set of libraries and tools that simplify deploying the hardware and the models they build. Google’s TPUs are mainly available for rent through the Google Cloud platform.

Google is also adding a version of its TPU design to its Pixel phone line to accelerate any of the AI chores that the phone might be used for. These could include voice recognition, photo improvement or machine translation. Google notes that the chip is powerful enough to do much of this work locally, saving bandwidth and improving speeds because, traditionally, phones have offloaded the work to the cloud. 

Many of the cloud companies like Amazon, IBM, Oracle, Vultr and Microsoft are installing these GPUs or TPUs and renting time on them. Indeed, many of the high-end GPUs are not intended for users to purchase directly because it can be more cost-effective to share them through this business model. 

Amazon’s cloud computing systems are also offering a new set of chips built around the ARM architecture. The latest versions of these Graviton chips can run lower-precision arithmetic at a much faster rate, a feature that is often desirable for machine learning. 

Some companies are also building simple front-end applications that help data scientists curate their data and then feed it to various AI algorithms. Google’s CoLab or AutoML, Amazon’s SageMaker, Microsoft’s Machine Learning Studio and IBM’s Watson Studio are just several examples of options that hide any specialized hardware behind an interface. These companies may or may not use specialized hardware to speed up the ML tasks and deliver them at a lower price, but the customer may not know. 

How startups are tackling creating AI hardware

Dozens of startups are approaching the job of creating good AI chips. These examples are notable for their funding and market interest: 

  • D-Matrix is creating a collection of chips that move the standard arithmetic functions to be closer to the data that’s stored in RAM cells. This architecture, which they call “in-memory computing,” promises to accelerate many AI applications by speeding up the work that comes with evaluating previously trained models. The data does not need to move as far and many of the calculations can be done in parallel. 
  • Untether is another startup that’s mixing standard logic with memory cells to create what they call “at-memory” computing. Embedding the logic with the RAM cells produces an extremely dense — but energy efficient — system in a single card that delivers about 2 petaflops of computation. Untether calls this the “world’s highest compute density.” The system is designed to scale from small chips, perhaps for embedded or mobile systems, to larger configurations for server farms. 
  • Graphcore calls its approach to in-memory computing the “IPU” (for Intelligence Processing Unit) and relies upon a novel three-dimensional packaging of the chips to improve processor density and limit communication times. The IPU is a large grid of thousands of what they call “IPU tiles” built with memory and computational abilities. Together, they promise to deliver 350 teraflops of computing power. 
  • Cerebras has built a very large, wafer-scale chip that’s up to 50 times bigger than a competing GPU. They’ve used this extra silicon to pack in 850,000 cores that can train and evaluate models in parallel. They’ve coupled this with extremely high bandwidth connections to suck in data, allowing them to produce results thousands of times faster than even the best GPUs.  
  • Celestial uses photonics — a mixture of electronics and light-based logic — to speed up communication between processing nodes. This “photonic fabric” promises to reduce the amount of energy devoted to communication by using light, allowing the entire system to lower power consumption and deliver faster results. 

Is there anything that AI hardware can’t do? 

For the most part, specialized hardware does not execute any special algorithms or approach training in a better way. The chips are just faster at running the algorithms. Standard hardware will find the same answers, but at a slower rate.

This equivalence doesn’t apply to chips that use analog circuitry. In general, though, the approach is similar enough that the results won’t necessarily be different, just faster. 

There will be cases where it may be a mistake to trade off precision for speed by relying on single-precision computations instead of double-precision, but these may be rare and predictable. AI scientists have devoted many hours of research to understand how to best train models and, often, the algorithms converge without the extra precision. 

There will also be cases where the extra power and parallelism of specialized hardware lends little to finding the solution. When datasets are small, the advantages may not be worth the time and complexity of deploying extra hardware.

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.

Repost: Original Source and Author Link


Cohere partners with Google Cloud to train large language models using dedicated hardware

Google Cloud, Google’s cloud computing services platform, today announced a multi-year collaboration with startup Cohere to “accelerate natural language processing (NLP) to businesses by making it more cost effective.” Under the partnership, Google Cloud says it’ll help Cohere establish computing infrastructure to power Cohere’s API, enabling Cohere to train large language models on dedicated hardware.

The news comes a day after Cohere announced the general availability of its API, which lets customers access models that are fine-tuned for a range of natural language applications — in some cases at a fraction of the cost of rival offerings. “Leading companies around the world are using AI to fundamentally transform their business processes and deliver more helpful customer experiences,” Google Cloud CEO Thomas Kurian said in a statement. “Our work with Cohere will make it easier and more cost-effective for any organization to realize the possibilities of AI with powerful NLP services powered by Google’s custom-designed [hardware].”

How Cohere runs

Headquartered in Toronto, Canada, Cohere was founded in 2019 by a pedigreed team including Aidan Gomez, Ivan Zhang, and Nick Frosst. Gomez, a former intern at Google Brain, coauthored the academic paper “Attention Is All You Need,” which introduced the world to a fundamental AI model architecture called the Transformer. (Among other high-profile systems, OpenAI’s GPT-3 and Codex are based on the Transformer architecture.) Zhang, alongside Gomez, is a contributor at, an open AI research collective involving data scientists and engineers. As for Frosst, he, like Gomez, worked at Google Brain, publishing research on machine learning alongside Turing Award winner Geoffrey Hinton.

In a vote of confidence, even before launching its commercial service, Cohere raised $40 million from institutional venture capitalists as well as Hinton, Google Cloud AI chief scientist Fei-Fei Li, UC Berkeley AI lab co-director Pieter Abbeel, and former Uber autonomous driving head Raquel Urtasun.

Unlike some of its competitors, Cohere offers two types of English NLP models, generation and representation, in Large, Medium, and Small sizes. The generation models can complete tasks involving generating text — for example, writing product descriptions or extracting document metadata. By contrast, the representational models are about understanding language, driving apps like semantic search, chatbots, and sentiment analysis.

To keep its technology relatively affordable, Cohere charges access on a per-character basis based on the size of the model and the number of characters apps use (ranging from $0.0025-$0.12 per 10,000 characters for generation and $0.019 per 10,000 characters for representation). Only the generate models charge on input and output characters, while other models charge on output characters. All fine-tuned models, meanwhile — i.e., models tailored to particular domains, industries, or scenarios — are charged at two times the baseline model rate.

Large language models

The partnership with Google Cloud will grant Cohere access to dedicated fourth-generation tensor processing units (TPUs) running in Google Cloud instances. TPUs are custom chips developed specifically to accelerate AI training, powering products like Google Search, Google Photos, Google Translate, Google Assistant, Gmail, and Google Cloud AI APIs.

“The partnership will run until the end of 2024 with options to extend into 2025 and 2026. Google Cloud and Cohere have plans to partner on a go-to-market strategy,” Gomez told VentureBeat via email. “We met with a number of Cloud providers and felt that Google Cloud was best positioned to meet our needs.”

Cohere’s decision to partner with Google Cloud reflects the logistical challenges of developing large language models. For example, Nvidia’s recently released Megatron 530B model was originally trained across 560 Nvidia DGX A100 servers, each hosting 8 Nvidia A100 80GB GPUs. Microsoft and Nvidia say that they observed between 113 to 126 teraflops per second per GPU while training Megatron 530B, which would put the training cost in the millions of dollars. (A teraflop rating measures the performance of hardware, including GPUs.)

Inference — actually running the trained model — is another challenge. On two of its costly DGX SuperPod systems, Nvidia claims that inference (e.g., autocompleting a sentence) with Megatron 530B only takes half a second. But it can take over a minute on a CPU-based on-premises server. While cloud alternatives might be cheaper, they’re not dramatically so — one estimate pegs the cost of running GPT-3 on a single Amazon Web Services instance at a minimum of $87,000 per year.

Cohere rival OpenAI trains its large language models on an “AI supercomputer” hosted by Microsoft, which invested over $1 billion in the company in 2020, roughly $500 million of which came in the form of Azure compute credits.

Affordable NLP

In Cohere, Google Cloud — which already offered a range of NLP services — gains a customer in a market that’s growing rapidly during the pandemic. According to a 2021 survey from John Snow Labs and Gradient Flow, 60% of tech leaders indicated that their NLP budgets grew by at least 10% compared to 2020, while a third — 33% — said that their spending climbed by more than 30%.

“We’re dedicated to supporting companies, such as Cohere, through our advanced infrastructure offering in order to drive innovation in NLP,” Google Cloud AI director of product management Craig Wiley told VentureBeat via email. “Our goal is always to provide the best pipeline tools for developers of NLP models. By bringing together the NLP expertise from both Cohere and Google Cloud, we are going to be able to provide customers with some pretty extraordinary outcomes.”

The global NLP market is projected to be worth $2.53 billion by 2027, up from $703 million in 2020. And if the current trend holds, a substantial portion of that spending will be put toward cloud infrastructure — benefiting Google Cloud.


VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more

Become a member

Repost: Original Source and Author Link


Oculus Go unlock software released for hardware freedom

An Oculus Go unlocked OS build was released today, courtesy of John Carmack. After a prolonged period of time and a whole lot of work in securing the rights to release the software, Carmack and crew made this build a reality. It’s a surprise, honestly, now that Oculus is owned by Facebook. It IS a welcome surprise, in any case.

The Oculus Go software unlock is relatively simple to enact. The most difficult part of the process will be your decision to push the software, given the agreement you must make with Oculus. You’ll effectively be agreeing that they are no longer responsible for any part of the device’s software or any future software updates.

SEE TOO: Our original Oculus Go review

Per Oculus, “this process will not work on any other device or OS,” and “please note that unlocking your device is not reversible and you will no longer receive OTA updates.” So why would you want to initiate such an unlock?

Because once you unlock you Oculus Go, you can install any software you like. Per Oculus, this process will give you access to the OS build so that you might “repurpose Oculus Go for more things today.”

Per Carmack, this process “opens up the ability to repurpose the hardware for more things today, and means that a randomly discovered shrink wrapped headset twenty years from now will be able to update to the final software version, long after over-the-air update servers have been shut down.”

Take a peek at the Unlocking Oculus Go page at the Oculus for Developers website. There you’ll be taken through the steps that are required to unlock the device and open your software door to the future. And if you happen to find a very awesome use for this headset and would like to share, let us know! Or just let me know over on Twitter – let’s chat!

Repost: Original Source and Author Link


What Does GPU Stand For? A Look at a Vital Piece of Hardware

When you’re shopping around for a new laptop or desktop, it can be helpful to know a little about the important internal components that help such devices perform as well they do. This way, you can figure out if the computer you’re considering purchasing has the right specs for how you intend to use the machine once you take it home.

In this guide, we’ll be taking a quick look at one of these internal computing components: GPUs. We’ll go over what its abbreviation means, how they can (in some cases) be different from graphics cards, and a few other details you’ll want to keep in mind while shopping for your next computer.

What does GPU stand for?

Martin Katler/Unsplash

GPU stands for “graphics processing unit.” Generally speaking, it’s a type of processor that handles and speeds up graphics rendering (the generation of the images you’ll see on computer screens). This is especially important for computing tasks such as 3D rendering and gaming. So if you’re buying a PC or laptop for gaming or content creation projects, you’ll need to pay attention to the kinds of GPUs such devices have to offer because it can affect the quality of what you see on screen and how fast certain operations are completed.

Are GPUs and graphics cards the same thing?

Colloquially, GPU is often used interchangeably with “graphics card,” but it can also mean the actual GPU at the heart of the graphics card, alongside the memory and other components — in the same way that CPU refers to the central processor, not the entire PC around it. Technically, a GPU is part of a graphics card. The GPU does the graphics rendering, while the graphics card provides the power and access to high-speed memory it requires. It also connects it to the other parts of the computer it needs access to, like the CPU, system memory, and storage, to complete its tasks.

What does iGPU mean?

An iGPU is an integrated graphics processing unit. It’s a type of GPU that’s found on the same chip as a CPU. Almost all Intel processors (save its F-series models) include integrated or onboard graphics, so they have an iGPU. AMD only includes iGPUs on its APU model processors, typically found in laptops.

With iGPU’s, the main processor shares memory with the GPU, which results in benefits such as a cooler machine and less power consumption. iGPUs are great for normal everyday computing tasks like surfing the web and productivity tasks, but if you need a powerhouse machine for serious gaming or video editing, you may want to look at a graphics card, or dGPU, instead.

What does dGPU stand for?

A dGPU is a “discrete GPU.” A dGPU is not merged with the main processor and doesn’t share memory with the main processor either. It has its own memory, so it can be a full-size graphics card in a desktop or a discrete graphics chip within a laptop. Typically, dGPUs offer better performance and can handle more intensive tasks. But this also means they consume more power, and machines with dGPUs will produce more heat. Laptops can have dGPUs, but you’ll mostly see them in desktop computers.

How do I know if I have an iGPU or dedicated graphics card?

Generally speaking, you can search for your machine’s GPU information via its device settings. For example, on Windows 10, you can do so by navigating to Device Manager and then selecting Display Adapters, which should expand to show you a list of GPUs that your device has. You can Google the name of the GPUs to find out if they’re graphics cards or iGPUs.

You can also find your PC’s GPU info and other specs in several other ways, including third-party programs. Check out our guide to how to check your PC’s specifications on Windows 10 for more info.

Now that you’ve learned a little more about GPUs, you may want to continue doing a little more research before you commit to buying your next PC. If you need more information, you should peruse our best laptops guide.

Editors’ Choice

Repost: Original Source and Author Link


Blaize raises $71M for AI edge hardware

All the sessions from Transform 2021 are available on-demand now. Watch now.

Blaize, a company developing AI edge computing platforms for automotive, enterprise, and computer vision markets, today announced that it raised $71 million in series D funding led by Franklin Templeton and Temasek, with participation from Denso and other new and existing backers. The company says that the funds will be used to support its go-to-market and R&D efforts.

The pandemic has accelerated the adoption of edge computing, or computation and data storage that’s located close to where it’s needed. According to the Linux Foundation’s State of the Edge report, digital health care, manufacturing, and retail businesses are particularly likely to expand their use of edge computing by 2028. This is largely because of the technology’s ability to improve response times and save bandwidth while enabling less constrained data analysis.

Blaize initially focused on “vision processors” to speed up sensor fusion tasks before expanding to datacenters, edge infrastructure, and enterprise client devices. The company’s 16-core chips — which deliver 16 TOPS of AI inference horsepower — enable concurrent execution of AI models while supporting heterogeneous compute-intensive workloads.


Blaize’s chips leverage graph computing and dynamic streaming to minimize non-computational data movement. Models can be built to integrate functions such as image signal processing, represented as graphs. And developers can build multiple models on a single architecture through to runtime.

Blaize’s cofounders — Dinakar Munagala, Ke Yin, Satyaki Koneru, and Val Cook — were at Intel prior to starting the company, developing several generations of GPUs. In 2010, working out of a spare bedroom, they eventually gained support and funding from venture, angel, and strategic investors including Dado Banatao.

“Customers are choosing Blaize … because we address their unmet need for products purpose-built for the requirements of edge AI,” Munagala told VentureBeat via email. “We win by achieving a better match for customer requirements with a balance of performance, power, latency, and cost … Programmability — and the flexibility that it gives customers — is key in many accounts as well, as is our advanced software, which transforms productivity for faster return on investment from edge AI deployments.”

AI accelerators

“Edge AI” describes architectures in which AI models are processed locally, on devices at the edge of the network. According to Markets and Markets, the global edge AI software market is anticipated to grow from $590 million in 2020 to $1.83 billion by 2026. Deloitte estimates more than 750 million edge AI chips that perform tasks on-device have been sold to date, representing $2.6 billion in revenue.

Startups AIStorm, Hailo, Esperanto Technologies, Quadric, Graphcore, Xnor, and Flex Logix, among others, are developing chips customized to accelerate edge AI workloads. But Blaize claims it’s differentiated by its no-code software, which implements “edge-aware” transfer learning and optimizations to achieve high accuracy post-model-compression.

“On the supply side, [our software] helps hardware vendors to multiply their value. On the demand side, it allows consumers of the hardware to construct apps without a legion of data scientists,” Munagala said. “This is critical on the path to proliferation of AI at the edge, as is our … hardware offering that customers can use to build solutions for their specific needs. For edge AI deployments, users need both efficient hardware and the ability to enable business technologists to develop and deploy applications, rather than requiring the hiring of costly AI data science teams.”


Blaize claims that it’s generated revenue since 2017, two years before it emerged from stealth. The company’s first commercial product began shipping in Q4, and Blaize has “multiple” customer engagements with automotive OEMs and Tier 1 suppliers, industrial systems integrators, independent software vendors, and end users.

“In the electric vehicle automotive market, low power to enable the key product feature of range extension is key. [And] in [the] retail shoplifting use case, Blaize [has been] favored for achieving the necessary performance and accuracy at one-fifth the power, while running multiple apps utilizing programmability,” Munagala said. “Overall, we have been able to successfully cope with the disruptions to business as usual that occurred due to the pandemic. As with most companies, our supply chain is affected. To mitigate the challenge, we proactively placed orders for products more than 18 months ahead. Some of our customer projects were temporarily affected by the pandemic, but now we are seeing a rebound.”

Blaize, which is headquartered in El Dorado Hills, California, has over 300 employees. It’s raised $155 million to date.


VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more

Become a member

Repost: Original Source and Author Link


Lucata raises $11.9M to accelerate graph analytics with specialized hardware

All the sessions from Transform 2021 are available on-demand now. Watch now.

Graph analytics startup Lucata today announced that it raised $11.9 million in series B funding, bringing its total raised to nearly $30 million. Notre Dame, Middleburg Capital Development, Blu Ventures, Hunt Holdings, Maulick Capital, Varian Capital, Samsung Ventures, and Irish Angels participated in the round, which CEO Michael Mallick says will be put toward commercializing the company’s computing architecture for graph analytics and AI use cases.

Graph analytics is a set of techniques that allows companies to drill down into the interrelationships between organizations, people, and things. The applications span cybersecurity, logistics, neural networks, natural language processing, and ecommerce, but one increasingly popular use case is fraud detection. For large credit card issuers, financial fraud can cost tens of billions of dollars a year. If these companies could run real-time graph analytics on large graph databases, some experts assert, they could detect fraud hours sooner than what’s possible today.

New York-based Lucata offers a hardware platform — Pathfinder — that ostensibly enables organizations to better support large graph analytics workloads. The company leverages “migrating threads” to conduct high-performance, “multi-hop” analytics, including on databases with over 1 trillion vertices. Organizations can use existing graph database software or custom solutions to analyze deep connections on expanded graphs.

“Lucata was founded in 2008 as Emu Technology by Peter Kogge, Jay Brockman, and Ed Upchurch. The company was [started] to commercialize migrating thread technology, which was developed and patented by the founders to address the scale and performance limitations of traditional computing architectures for big data,” Maulick told VentureBeat via email. “Migrating thread technology enables the creation of shared RAM and CPU pools that allow users to process monolithic big data datasets in real-time with no data pruning or database sharding.”

Graph technology

According to Maulick, current machine learning and AI model training on large, sparse datasets often leverage approaches that can skew the results. One method is to reduce the size of the dataset by pruning — i.e., deleting — significant amounts of data during loading that are thought to be unimportant. The other technique is to “shard” the loaded data into smaller subsets of data, which the model training process sequentially processes.

Bias or skew can creep into the models if important data is deleted during pruning. But with Lucata’s technology, Maulick argues, users can avoid this by loading entire datasets into a single RAM image, leading to improved accuracy during training.

“Companies that would potentially benefit from using [our] Lucata computing architecture to improve the performance of their software include Redis Labs, TigerGraph, Neo4j, and other graph database vendors. In addition, software vendors and cloud providers that offer solutions which leverage common machine learning and AI processing frameworks such as PyTorch, TensorFlow and Apache Spark would potentially benefit from using the Lucata Pathfinder platform,” Maulick said.

Because Lucata’s hardware relies on DRAM chips that are in short supply, owing to the worldwide semiconductor shortage, the company anticipates its production schedule will be impacted going forward. But even with this being the case, year-over-year revenue from 2020 to 2021 is internally projected to grow 100%.

Lucata has a workforce of 20 people across its offices in Palo Alto, New York City, and South Bend, Indiana, which it expects will expand to 34 by 2021. “The pandemic impacted the work of our employees in our physical office in New York City but has not had a significant impact on those of our employees who work remotely,” Maulick said.


VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more

Become a member

Repost: Original Source and Author Link


Lambda raises $24.5M for AI-optimized hardware infrastructure

Join executive leaders at the Conversational AI & Intelligent AI Assistants Summit, presented by Five9. Watch now!

Lambda, an AI infrastructure company, this week announced it raised $15 million in a venture funding round from 1517, Gradient Ventures, Razer, Bloomberg Beta, Georges Harik, and others, plus a $9.5 million debt facility.  The $24.5 million investment brings the company’s total raised to $28.5 million, following an earlier $4 million seed tranche.

In 2013, San Francisco, California-based Lambda controversially launched a facial recognition API for developers working on apps for Google Glass, Google’s ill-fated heads-up augmented reality display. The API — which soon expanded to other platforms — enabled apps to do things like “remember this face” and “find your friends in a crowd,” Lambda CEO Stephen Balaban told TechCrunch at the time. The API has been used by thousands of developers and was, at least at one point, seeing over 5 million API calls per month.

Since then, however, Lambda has pivoted to selling hardware systems designed for AI, machine learning, and deep learning applications. Among these are the TensorBook, a laptop with a dedicated GPU, and a workstation product with up to four desktop-class GPUs for AI training. Lambda also offers servers, including one designed to be shared between teams and a server cluster, called Echelon, that Balaban describes as “datacenter-scale.”

“We initially created a facial recognition API, a deep learning-powered image editor called Dreamscope, and, because of the experience of running our own GPU infrastructure, decided to offer preconfigured workstations and servers,” Balaban told VentureBeat via email. “We provide the infrastructure — laptops, workstations, and servers — so that our customers can focus on training models and building value for their customers. Most companies get distracted by building huge infrastructure teams when they should be building huge machine learning teams to use infrastructure that’s easier to manage.”

Lambda Labs

Above: One of Lambda workstations.

Image Credit: Lambda Labs

Software plus hardware

A number of startups offer preconfigured hardware for AI development, including Graphcore. But Balaban says Lambda’s major differentiator is its software tools.

Every Lambda machine comes preinstalled with Lambda Stack, a collection of machine learning software development frameworks, including Google’s TensorFlow and Facebook’s PyTorch. Developers can update the frameworks with a single console command, and if they’ve trained the model on a local machine, they can copy it up to a Lambda server running in the cloud.


“Our customers include Apple, Intel, Microsoft, Amazon Research, Tencent, Kaiser Permanente, MIT, Stanford, Harvard, Caltech, and the Department of Defense,” Balaban said. “We have [thousands] of users, [and] most of the Fortune 500 and almost every major research university in the U.S. — as well as [many of] the research labs at the Department of Transportation and Department of Energy — use Lambda hardware and Lambda Stack.”

Balaban also claims that the over-40-employee company, which was founded in 2012, has been cash-flow positive since November 2013. It’s on a $60 million revenue run rate for 2021 and plans to have around 60 employees by the end of the year.


VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more

Become a member

Repost: Original Source and Author Link

Tech News

Square Bitcoin hardware wallet floated by Jack Dorsey

Jack Dorsey revealed that Square was considering creating a Bitcoin hardware wallet this week. He suggested that “if we do it”, they’ll build it “in the open, from software to hardware design, and in collaboration with the community.” Jack suggested that if they do it, they’ll do it with principals like “bitcoin is for everyone” and “no keys, no cheese.”

What does “no keys, no cheese” mean? In this case, it means Square wants to make the actual “custody” of the Bitcoin as clear as possible. To do this, they’ll be attempting to simplify the process with some sort of “assisted self-custody” system.

They’ll also be attempting to “blend availability and security.” They’ll have to consider that safety failures stem from one of three types of events. As Jack suggested, these are availability failures (“sunken gold”), security failures (“pirated gold”), and discretionary actions (“confiscated gold”).

They’ll likely integrate this hardware wallet with their own Cash App. They’ll likely be creating a custom-built app, but “it doesn’t need to be owned by Square.” Jack suggested that they’re able to “imagine apps that work without Square and maybe also without permission from Apple and Google.” The entire Tweet string makes the case for the wallet – and asks whether it should exist in the first place.

This should prove interesting. Making an app that doesn’t rely on permission from Apple or Google AND works reliably on mobile devices – that’s not easy. That might not even be possible. Would you trust a hardware wallet for your Bitcoin that follows all the guidelines as listed above? Even if it was made by Square?

Repost: Original Source and Author Link

Tech News

Beats design at Apple now lead by Android hardware legend

Apple appears to have pinpointed one of the most influential Android hardware designers of the last decade in Scott Croyle. This guy was the head of HTC’s design team when they created the HTC One M7 – he was also a founder of the company Nextbit, creators of the Nextbit Robin. A report this week details the timing and details surrounding the designer’s joining Apple.

Below you’ll see the Nextbit Robin and an HTC One M7 (originally just called HTC One). These devices remain memorable pieces of hardware design even here, nearly a decade later. Now Croyle, one of the main keys behind this hardware design, is working with Apple.

FUN FACT: The HTC One (whose design team was led by Croyle), came with Beats Audio branding. This phone was released in early 2013, when HTC still owned stock in Beats. HTC sold their last shares of Beats by September of 2013. In the year 2014, Apple acquired Beats.

According to 9to5Mac, Scott Croyle “joined Apple last year specifically to oversee Beats product design.” It’s reported that Apple will continue to have the design firm Ammunition “create the look of Beats hardware products and company identity” while Croyle acts as “point person” between Beats and the design company.

It’s very likely that details like this appearing mean that we’ll see new Beats products in the very near future. The fact that Croyle is working with Apple specifically on Beats is a talking point that’s perfect for starting a fire in the minds of the public. We’re now reminded that Apple owns Beats, and Beats hasn’t released a brand new product for quite some time.

It’s likely there’ll be a new pair of Beats headphones in the near future, and it would not be shocking to find a new wireless speaker appearing soon, too. The Beats brand remains solid. The public still knows the logo, and it’s quite likely Apple will capitalize on the hype that will inevitably come with the release of a new Beats product before the end of the year 2021.

What sort of Beats brand hardware do you expect Apple will release next? Will it just be another pair of headphones with a slightly different shape from what’s come before, updated with Apple’s latest wireless chip tech? Or will it be something truly new?

Repost: Original Source and Author Link


Xbox Series X Console Purchase Pilot lets real people buy hardware

If you’re looking for an Xbox Series X or Xbox Series S, you’re not alone. The next-generation console series shortage is one of a wide variety of results of the COVID-19 pandemic and supply issues worldwide. Once supply was short, scalpers came to take advantage of the situation by buying up the Xbox Series X and S that were sold in the first few waves this year. Now, Microsoft is taking action with a “Console Purchase Pilot.”

The Console Purchase Pilot aims to put Xbox Series X and S units directly in to user hands, rather than through the grips of flippers. To do this, Microsoft will sell consoles to users who are currently Xbox Insider users on Xbox One. This way there’s a far better chance that the person buying the console is an actual end customer, rather than a 3rd-party scalper.

The program can be found via the Xbox Insider Hub on Xbox One. There, users will be able to register to participate. Not all registered users will be “selected” to “have a chance” to purchase an Xbox Series X or S. At the moment, it would seem that this is more of a test run than it is a “this should work for everyone” sort of situation.

You’ll see one announcement for this program in a Tweet from the official Xbox Insider Twitter account below. As noted, not all those who register will be selected.

The Xbox Series X and Xbox Series S have also been appearing in small supply every once in a while in online storefronts like Target, Walmart, Best Buy, and Gamestop. Take a peek at the timeline below for other recent updates to the Xbox Series X universe as well, and stay tuned as we report on the next tip on availability from confirmed sources.

Repost: Original Source and Author Link