Artificial Intelligence - Page 43

TSMC has announced its development of next-generation Compact Universal Photonic Engine (COUPE) technology, which will support the "explosive growth" in data transmission associated with the "AI boom."

Taiwan Semiconductor Manufacturing Company (TSMC) unveiled its newest semiconductor process, advanced packaging, and 3D IC technologies for the next-generation of AI innovations with silicon leadership at TSMC's 2024 North America Technology Symposium.

At the event, TSMC unveiled its new A16 technology, which packs industry-leading nanosheet transistors with an innovative backside power rail solution for production in 2026, with "greatly improved logic density and performance." TSMC also introduced its new System-on-Wafer (TSMC-SoW) technology, which is an innovative solution that delivers "revolutionary performance to the wafer level" in addressing the future AI requirements for hyperscaler datacenterrs.

SK hynix has pledged another 20 trillion won (around $14.6 billion USD or so) in preemptively responding to "soaring demand" for HBM memory, which is a key part of AI GPUs.

The South Korean memory giant plans to complete the construction of its new fab in November 2025. Geared up and ready for early mass production, the fab will make major investments in enhancing AI memory market leadership and contributing to the domestic economy in South Korea.

SK hynix's board of directors has approved the plan, which will see the company building the new M15X fab in Cheongju, North Chungcheong Province, for a new DRAM production base, with an investment of around 5.3 trillion won (around $3.9 billion USD or so) for fab construction.

NVIDIA has announced that Japan's new quantum supercomputer will be powered by NVIDIA platforms for accelerated and quantum computing.

Japan's National Institute of Advanced Industrial Science and Technology (AIST) is building a hybrid cloud system of quantum computers and supercomputers called ABCI-Q. Quantum computers are still capable of making a lot of errors if they're operating solo, with supercomputers needing to solve the mistakes and make those complex operations smoother.

NVIDIA is providing the AI GPUs for the new ABCI-Q quantum supercomputer and quantum computing software through its cloud service. NVIDIA will provide over 2000 of its H100 AI GPUs in 500+ nodes interconnected by NVIDIA Quantum-2 InfiniBand, the world's only fully offloadable, in-networking computing platform.

Here's some good news for those of you with an older Android smartphone wanting to use Google's Gemini assistant, as the AI is now available for earlier versions of the mobile platform - plus a new 'conversation mode' is inbound for the app.

Previously, you needed a handset running Android 12 or better to use the Gemini app, but now, that has been pushed back to include phones with Android 10 or 11 installed on them.

Leaker @AssembleDebug on X (formerly Twitter) made this clear, as reported by Android Authority (via TechRadar).

NVIDIA announced its next-gen GB200 server AI platform at its GTC (GPU Technology Conference) last month. Quanta Computer received its first order for the new GB200 server chips, which will enter mass production in September 2024.

UDN reports that Quanta has confirmed it received its first order of NVIDIA GB200 servers, calling it the "most powerful AI chip on Earth" and that mass production begins in September, and will increase quarter by quarter into 2025. UDN reports that according to industry expectations, Quantua is more optimistic about the momentum of server shipments this year than originally expected.

This is because four of the major cloud service providers (CSPs), Microsoft, Google, Amazon AWS, and Meta, are all ordering NVIDIA GB200-powered AI servers, which Quanta will make.

Samsung has just announced that its next-generation HBM4 memory is in development, and will make its first appearance in 2025 with next-gen speeds, capacities, and features.

The South Korean everything giant posted a blog explaining its next-gen HBM4 is in development, with its current HBM3E "Shinebolt" memory as its flagship HBM. HBM3E "Shinebolt" memory has 36GB capacities using 24Gb DRAM and up to 9.8Gbps bandwidth, with the memory technology supporting 12-Hi stacks and 2.5D advanced packaging technology.

HBM4 is Samsung's next step. If the company uses the same 24Gb modules, we should expect to see 16-Hi stacks. This will allow up to 256GB of HBM4 on future-gen AI GPUs with speeds that should hit 10Gbps or more. Samsung is surely cooking up denser DRAM modules, where we should see 24Gb+ that will drive total capacities of HBM4-powered AI GPUs through the roof.

The Internal Revenue Service (IRS) is planning on implementing artificial intelligence-powered systems into its workflow, according to IRS Commissioner Danny Werfel.

The statements made by the CEO of the IRS were made on April 17 at the UiPath on Tour: Public Sector Event in Washington, D.C, where Werfel explained that the IRS is looking into implementing AI in two ways: one to assist American taxpayers and the other and tools that IRS agents can use to discover undetected tax cheats.

The CEO described the potential of AI tools to the IRS as "night vision goggles" that would enable the agency to "unlock and see and spot the issues" in case of the "most complicated, largest taxpayers" across the US. As for AI being implemented in a way that would help Americans, the IRS CEO said the agency is developing an IRS chatbot that taxpayers will be able to interact with virtually and ask questions regarding the taxation process.

A newly filed lawsuit against Amazon alleges the company broke its own copyright policy to prevent it from falling out the global generative artificial intelligence race that numerous big tech companies are embarking on.

The lawsuit was filed last week in a Los Angeles state court by Dr Viviane Ghaderi, an AI researcher, who claims to have worked in Amazon's Alexa and Large Language Model (LLM) departments. Ghaderi claims she was promoted several times in both of these departments, but was allegedly demoted and eventually fired following her return back from maternity leave.

Ghaderi has placed many allegations against Amazon, claiming the company has discriminated against her, is inherently sexist towards her, retaliation, wrongful termination, and many other claims. More specifically, Ghaderi says when she returned to work, she was tasked with an LLM project, the underlying technology powering impressive tools such as OpenAI's ChatGPT, and her role was flagging instances of the LLM violating Amazon's own copyright policy.

Google is set to shake up the weather prediction industry with the release of SEEDS, or the Scalable Ensemble Envelope Diffusion Sampler AI model.

The new Large Language Model (LLM) is designed to provide accurate weather predictions much cheaper and faster than traditional weather prediction tools, specifically, weather events such as hurricanes or heat waves that can have a potentially devastating impact on regions. So, how does it work? Predicting the weather inherently is difficult due to the multitude of variables that are at play, with current forecasting being good enough for conditions such as local temperature, it gets progressively more difficult the further out into the future those predictions.

It gets even harder to predict the occurrence of an extreme weather event as that is a culmination of all the variables in typical weather predictions plus random variables that would induce an extreme weather event. The accuracy of predicting an extreme weather event is currently extremely low, with Google explaining that a model needs to incorporate 10,000 weather predictions for an event to have a 1% likelihood of occurrence.

Last week, LG was in Milan, Italy, showcasing its latest built-in kitchen appliances, including a Signature Kitchen Suite with a built-in oven, 'free zone induction hob,' and downdraft hood. Now, you might be wondering why we're suddenly bringing you news about a new all-in-one oven do-dad - that's because AI powers it.

LG calls it Gourmet AI. The oven can identify what is being cooked through video recognition and automatically select the suitable cooking mode from 130 recipes. With Gourmet AI, the oven also keeps tabs on the cooking process, watching how pizza or a steak is 'browning' and then alerting users via LG's ThinQ app to let them know what they were cooking is done.

There's also real-time video monitoring so that you can watch those cookies cook, and time-lapse recordings are also available - which is a pretty cool feature for an oven.

Humans are only so good at detecting details in art, which is why researchers have decided to train an AI algorithm to detect what humans can't and only suspect.

A team of researchers from the UK and US created a custom algorithm using a pre-trained architecture called ResNet50, which was developed by Microsoft. The team took advantage of a common machine learning technique to improve the algorithm, feeding it with authenticated paintings by the famous artist Raphael. The goal was to teach the AI to identify with a high degree of accuracy brushstrokes in paintings that were created by Raphael's hand and the faces of people within those paintings.

One painting, in particular, from Raphael, has been a point of debate among scholars. The Madonna della Rosa, or Madonna of the Rose, was painted on canvas from 1518 to 1520. Scholars have been debating for many years if the painting was an original work by the famed artist or if its creation involved another hand, specifically with the face of St Joseph. Some scholars have argued that St Joseph's face isn't as meticulously crafted as the Raphael's other pieces of work, leading them to believe it was done by another painter.

NVIDIA has just announced optimizations across all of its platforms to accelerate Meta Llama 3, Meta's latest-generation large language model (LLM).

The new Llama 3 model combined with NVIDIA accelerated computing provides developers, researchers, and businesses with innovation across various applications. Meta engineers trained their new Llama 3 on a computing cluster featuring 24,576 NVIDIA H100 AI GPUs linked through the NVIDIA Quantum-2 InfiniBand network; with support from NVIDIA, Meta tuned its network, software, and model architectures for its flagship Llama 3 LLM.

To further advance the state-of-the-art generative AI, Meta recently described plans to scale its AI GPU infrastructure to an astonishing 350,000 NVIDIA H100 AI GPUs. That's a lot of AI computing power, a ton of silicon, probably a city's worth of power, and an incredible sum of money on AI GPUs ordered by Meta from NVIDIA.

Meta has a nice surprise today: its latest large language model (LLM), Llama 3. The company explains its new Llama 3 8B contains 8 billion parameters, while its Llama 3 70B features 70 billion parameters.

Meta promises a gigantic increase in performance over the previous Llama 2 8B and Llama 2 70B models, with the company claiming that Llama 3 8B and Llama 3 70B are some of the best-performing generative AI models available today, trained on two custom-built 24,000 GPU clusters.

The company trained its new Llama 3 model on over 15 trillion tokens that were all collected from "publicly available sources," and that Met'a training dataset is 7x larger than what was used for Llama 2, and that includes 4x more code.

AI startup TensorWave is one of the first with a publically deployed setup powered by AMD Instinct MI300X AI accelerators, and its CEO says they're a far better option than NVIDIA's dominant Hopper H100 AI GPU.

TensorWave started racking up AI systems powered by AMD's just-released Instinct MI300X AI accelerator, which it plans to lease the MI300X chips at a fraction of the cost of NVIDIA's Hopper H100 AI GPU. TensorWave plans to have 20,000 of AMD's new Instinct MI300X AI accelerators before the end of the year across two facilities, and has plans to have liquid-cooled systems online in 2025.

Jeff Tatarchuk said that AMD's new Instinct MI300X AI GPU in "just raw specs, the MI300X dominates H100," and he's not wrong. But that's the original H100 with 40GB and 80GB HBM3 options, while a beefed-up H200 with a much larger 141GB of ultra-fast HBM3e memory with up to 4.8TB/sec of memory bandwidth, but once Blackwell B200 is here, NVIDIA owns the AI GPU game with 192TB of HBM3e at 8TB/sec memory bandwidth for the B200 AI GPU.

A fresh jaw-dropping initiative for AI, this one from Microsoft, runs like this - give AI a photo of someone, and an audio clip of their voice, and it'll mock up a video clip of that person chatting away.

So, from just an image and a minute long chunk of audio of the person's voice - which is what was used to create the example videos Microsoft shares - you get these fully realized talking faces, complete with lifelike facial expressions, head and eye movements, full lip-syncing, and so on.

In short, the whole works needed for it to look like a totally realistic video of said person espousing their thoughts on any given topic, all generated in real-time by the AI.

Foxconn and NVIDIA are teaming up to battle the future of AI, with Foxconn chairman Liu Yangwei saying yesterday that we'll see the progress of what the companies have been working on, later this year.

The two companies have teamed up for three major platforms: AI factories, AI smart manufacturing, and AI smart electric vehicles. Foxconn is expected to ramp up its support with NVIDIA in building next-gen Blackwell GB200 AI servers, which was announced at the 2024 Taipei International Automotive Electronics Show.

The Foxconn executive flew to the United States to personally attend NVIDIA's recent GTC (GPU Technology Conference), where he said that the two companies would be working on AI factories and that "everyone will soon see some construction on NVIDIA's AI factories".

NVIDIA's new wave of Blackwell AI GPUs will boost demand for TSMC's advanced CoWoS packaging capacity by over 150% in 2024.

In a new report from TrendForce, the next-gen Blackwell AI GPU platform that includes the B200 and B100 AI GPUs -- as well as the GB200, which also features NVIDIA's in-house Arm-based Grace CPU -- all use CoWoS packaging from TSMC, and projections suggest millions of shipments of high-end B200 AI GPUs are expected in 2025 making up close to 50% of NVIDIA's high-end GPU market.

NVIDIA will launch its new GB200 and B200 in the second half of 2024, with upstream wafer packaging needing to adopt a much more complex, and high-precision CoWoS-L technology, which will make validation and testing take more time. On top of that, more time is needed to optimize the B-series for AI server systems in things like network communication and cooling performance.

Intel is now talking about its new augmented intelligence tool developed in-house by Intel engineers, system-on-chip (SoC) designers now aren't waiting 6 weeks to see if they hit the sensor sweet spot, thanks to AI they're getting these answers in just a few minutes.

The new AI-powered tool developed by the Augmented Intelligence team led by Dr. Olena Zhu, senior principal engineer and AI solution artitect inside of Intel's Client Computing Group (CCG), helped Intel's system architecture factor thousands of variables into future silicon designs, with it being just one of many where Intel teams are using AI to optimize various workloads.

Engineers are required to analyze complex, concurrent workloads that activate the CPU core, input/output (I/O), and other system functions precisely to determine the location of thermal hotspots on a processor. The engineers also have to decide where to place very tiny sensors-each is barely larger than the tip of your average pin, says Intel-which makes things more complicated again.

Intel is under continued pressure from US restrictions -- and could reportedly, like NVIDIA -- that would see Team Blue gimping their AI processors for the Chinese market.

This isn't a new move, with NVIDIA modifying its AI GPUs to meet US restrictions into China, and now it seems Intel could do it soon with "special edition" versions of its new Gaudi 3 AI accelerators. These two new gimped AI accelerators from Intel are rumored to be launched at the end of June, and the end of September, reports TrendForce.

Intel announced its new Gaudi 3 AI accelerator recently, and in the Gaudi 3 white paper it was discovered that Intel said it would be launching two special edition versions of Gaudi 3 that will be specifically for the Chinese market. There will be two hardware variants launched later this year: the HL-328 OAM-compatible Mezzanine card, and the HL-388 PCIe Accelerator Card. TrendForce reports that the HL-328 will launch on June 24, and the HL-388 on September 24.

NVIDIA has been making cut-down AI GPUs to circumvent US export restrictions in China for months now, but it appears modified Ampere A100 AI GPUs are also making the rounds there.

A new NVIDIA A100 7936SP -- not the real name; this name is based on its CUDA core count -- has the same GA100 GPU as the regular A100 -- but with 15% more CUDA cores. The regular A100 has 124 SMs (Streaming Multiprocessors), while the new A100 7936SP has 128 SMs in total.

There's also more HBM2e memory on the new A100 7936SP, with 96GB of HBM2e memory, compared to just 80GB of HBM2e on the standard A100 AI GPU. With the additional 4 SMs, the new A100 7936P has a 15% increase in SM, CUDA, and Tensor Core counts, which should relate to at least 15% more AI performance across the board.