Artificial Intelligence - Page 49

AMD and NVIDIA are both pumping out new generations of AI hardware, with the two GPU giants eating all of TSMC's advanced packaging capacity for not just 2024, but also 2025.

In a new report from UDN, the site reports that TSMC will contract advanced packaging production capacity of its in-house CoWoS and SoIC for this and next year, as AI GPU hardware orders aren't slowing down. TSMC expects revenue contributed by server AI processors to more than double this year, and once NVIDIA's new Blackwell B200 AI GPUs are being pumped out later this year, 2025 is going to be a bananas year for TSMC.

In response to the insatiable AI demand, TSMC is now actively expanding its advanced packaging production capacity, with industry estimates that TSMC's CoWoS monthly production capacity will be between 45,000 and 50,000 by the end of 2024. This is a gigantic 3x increase from the 15,000 in 2023, and it's expected to hit 50,000 in 2025.

AMD has announced that Nscale is the world's first AMD technology-focused cloud provider, using the company's new Instinct MI300X AI accelerators.

Nscale is a vertically integrated GPU cloud company spun out of Arkon Energy, with a 300MW data center and hosting business in the United States. Nscale is based in Northern Norway, with its N1 Glomfjord site featuring some of the cheapest renewable energy on the planet, making the N1 Glomfjord site one of the most cost-effective LLM and AI training hubs in the world.

Some of the key benefits and features of the Nscale cloud:

For quite some time we have been hearing how iOS 18 is expected to be Apple's biggest update ever to iOS, with reports even stating that internally Apple is referring to iOS 18 as "biggest update ever".

The main reason iOS 18 is so monumental is because it will be the time Apple implements artificial intelligence into its operating system. Rumors of this have been circulating for months and well documented by Bloomberg's Mark Gurman, who has reported Apple will be introducing a suite of new AI features that will include a revamped version of Siri, heavy AI integration into Apple's apps such as Apple Music, iWork apps like Keynote and Pages, and more overall customization.

Now, AppleInsider has learned more details about the AI implementation, with "people familiar with the software" telling the publication about the generative AI capabilities of the Ajax Large Language Model (LLM), Apple's generative AI model. The insiders said Ajax will power a feature that enables text summarization, which covers anything from webpages to messages. The feature will quickly read the on-screen text and provide summarized key points on the analyzed text.

X has rolled out an update that enables its Premium users to receive AI-generated summaries of news and topics trending on the platform.

The new feature called "Stories on X" has become available to Premium subscribers, and according to a post from the company's engineering account, the new feature will appear within the Explore tab and is entirely curated by X's GrokAI tool. So, how does it work? Grok has been leveraged to identify the most popular content trending on X, which can be news stories and any general public discussion that has gained a lot of attention.

Grok then digests the popular content and generates a summary. Users exploring the new feature have posted screenshots to X showcasing the design and layout, with some screenshots showing AI-generated summaries for stories such as Apple's earnings report, aid to Ukraine, and "Musk, Experts Debate National Debt," which was a summary of an online discussion between Musk and other prominent X users.

SK hynix has had an absolutely stellar last 12 months riding the ever-growing AI wave, with the South Korean memory giant expecting over $10 billion in revenue from HBM alone by the end of 2024.

The news is coming from South Korean outlet TheElec, which sums up that SK hynix has sold out of its 2024 supply of HBM memory, and is already close to selling out its 2025 supply of HBM. NVIDIA's current H100 uses HBM3, while its new H200 and next-gen B200 AI GPUs both use HBM3E memory, provided by SK hynix.

SK hynix is staying ahead of its HBM competitors in Samsung and Micron, with plans to provide samples of its new 12-stack HBM3E this month, with mass production of the new HBM memory chips expected in Q3 2024 according to SK hynix CEO Kwak Noh-jung at a press conference on Thursday.

NVIDIA's new DGX GB200 AI server is expected to enter mass production in the second half of this year, with the volume expected to hit 40,000 units in 2025.

NVIDIA and Quanta are the two suppliers of NVL72 and NVL36 cabinets, respectively, with the NVL72 packing 72 GPUs and 36 Grace CPUs. Each of the AI server cabinets cost 96 million NTD (around $3 million USD or so).

The new NVIDIA DGX NVL72 is the AI server with the most computing power, and thus, the highest unit price. Inside, the DGX NVL72 features 72 built-in Blackwell-based B200 AI GPUs and 36 Grace CPUs (18 servers in total with dual Grace CPUs and 36 x B200 AI GPUs per server) with 9 switches. The entire cabinet is designed by NVIDIA in-house, and cannot be modified, it is 100% made, tested, and provided by NVIDIA.

SK hynix has announced that almost all of its HBM volume for 2025 has been sold out, as demand for AI GPUs continues to skyrocket.

During a recent press conference, the South Korean memory giant announced plans to invest on its new M15X lab in the Cheongju and Yongin Semiconductor Cluster in Korea with advanced packaging plants in the US.

SK hynix selling out of most its 2025 HBM volume is pretty crazy, as we're not even half way through the year, and NVIDIA's beefed-up H200 AI GPU with HBM3E isn't quite here yet, and its next-gen Blackwell B200 AI GPUs with HBM3E will be launching later this year... yet SK hynix is selling HBM like they're hotcakes.

NVIDIA has just updated its ChatRTX AI chatbot with support for new LLMs, new media search abilities, and speech recognition technology. Check it out:

The latest version of ChatRTX supports more LLMs including Gemma, the latest open, local LLM trained by Google. Gemma was developed by the same research and technology that Google used to create Gemini models, and is built for responsible AI development.

ChatRTX now supports ChatGLM3, an open, bilingual (English and Chinese) LLM based on the general language model framework. The updated version of ChatRTX now lets users interact with image data through Contrastive Language-Image Pre-training from OpenAI. CLIP is a neural network that, as NVIDIA explains, through training and refinement will learn visual concepts from natural lanage supervsion -- a model that recognizes what the AI is "seeing" in image collections.

NVIDIA isn't going to be left out with the coming AI PC revolution, with AMD and Intel already releasing NPU-powered processors for AI workloads, NVIDIA has just announced the "Premium AI PC" with RTX hardware, compared to the "Basic AI PC" with an NPU on the CPU.

NVIDIA compares that a Premium AI PC with an RTX GPU for AI workloads can provide gigantic performance uplifts for AI TOPS of between 100 and a whopping 1300+ TOPS (int8 + FP8 workloads) compared to just 10-45 TOPS on current, and next-gen CPUs with beefed-up NPUs.

Intel and AMD will combine the TOPS performance of the CPU + GPU + NPU, because if they used the NPU alone, the AI power is nowhere near as great. The more the NPU is used, the more resources that are taken away from other key parts of the chip: ya know, like the CPU and GPU.

OpenAI introduced the 'memory' feature for its ChatGPT AI back in February 2024, but now that functionality is available to all users - well, paying subscribers, anyway.

To recap, what this feature does is to give ChatGPT the ability to remember elements of your previous chats with the AI.

It can then refer back to those memories to make things easier or more convenient in future queries, while making chats with the AI seem a more human-like experience.

The boss of Microsoft's AI division has said that people really need to change their perspective of the emergence of artificial intelligence, and that a different way of thinking about the new technology is consider it a new "species".

DeepMind co-founder, and current CEO of Microsoft AI, Mustafa Suleyman, held a TED talk last week where he explained his vision for a world that will be mostly driven by artificial intelligence. Suleyman spoke to the crowd and said that it may be useful to think of AI as "something like a new digital species", with his prediction being that humans will come to view AI as "digital companions" and "new partners in the journeys of all our lives."

Suleyman's idea is that artificial intelligence will become so engrained into our everyday lives, much like smartphones have, that we will begin to consider AI-powered pieces of tech, such as Siri, more of a virtual being rather than a tool. This change is perspective will be enabled by the AI's language model that will be designed to enable a much more fluent interaction and response.

The price of NVIDIA's Hopper H100 AI GPU in the Chinese black market is coming down "rapidly," with spot prices falling more than 10% in the short term.

In a new report from UDN, the market is watching the impact of this wave of selling on the overall supply and demand, and whether the normal market for AI GPUs will be affected. NVIDIA's new H200 AI GPU is about to be released, upgraded with HBM3E memory, meaning that the price of AI servers with H100 AI GPUs on the black market in mainland China have plummeted in anticipation of H200.

At the time of replacing the old H100 AI GPUs with newer H200 AI GPUs, scalpers who previously had lots of stock of H100 AI GPUs are trying to get rid of them all -- and sometimes, in large quantities -- which is quickly bringing the price down on the black market.

NVIDIA opened its first Asia-based AI R&D center in Taiwan two years ago, investing NT$24.3 billion ($715 million USD or so) and employing 400 people there. But now, the company is reportedly considering a second R&D center in Taiwan.

In a new post by the China Times, the Technology Department of the Ministry of Economic Affairs, said on April 26 that NVIDIA's current research and development (R&D) is "going smoothly" and it will "assist our manufacturers in AI application development". They're reportedly happy with the quality of talent in Taiwan, and are considering setting up a second R&D center in Taiwan.

The first R&D center that NVIDIA built in Taiwan mainly works on AI chip research and development, as well as GPUs. The R&D center also helped build Taiwan's biggest supercomputer -- Taipei-1 -- with the Taiwan government providing a NT$6.7 billion ($205 million USD or so) subsidy for NVIDIA's first R&D center project.

NVIDIA is expected to generate $40 billion of revenue from AI GPU sales this year according to Bloomberg analysts, with competitors AMD and Intel in their dust.

AMD is estimated to generate around $3.5 billion in revenue from its new Instinct MI300X AI accelerator, while Intel will enjoy some loose change with its Gaudi AI accelerator, with analysts estimating just $500 million in sales. Both pale in comparison to $40 billion from NVIDIA, which is 80x more than Intel's estimated $500 million.

These numbers were discussed by Bloomberg Technology during an interview with Intel CEO Pat Gelsinger, where he had no real answer to how Intel's new Gaudi AI accelerator will battle NVIDIA. When NVIDIA is making 80x your revenue -- and keep in mind, this is just AI GPU sales, not GeForce graphics cards or any other NVIDIA product -- it's hard to compete against the dominant force in AI.

Apple has removed three apps from the App Store after it was found that they were using generative AI to take photos uploaded by users and then turn them into nonconsensual nudes of their subjects.

The apps were first reported to Apple by 4040 Media earlier this week after ads promoting them were found in Meta's Ad Library. Other ads pointed to similar services that were web-based, but three were apps that were available for download via the App Store. Upon being informed, Apple asked for more information before removing the apps entirely. However, the apps were only removed once 404 Media had provided links to them, suggesting that Apple didn't know which apps were violating its rules.

Since the ads were discovered Meta has removed the offending ads from its platform, but apps like this are already proving to be a problem for many. The 4040 Media report notes that these so-called "undress" apps are already being used in schools across the United States. Just weeks ago two middle school students were arrested in Florida while another case saw several students from a Washington state high school investigated by police after using an app to remove the clothes of someone in a photo.

NVIDIA CEO Jensen Huang gave a hand-written note that he wrote for SK Group Chairman Chey Tae-won, which said: "To our partnership and creating the future of AI and humanity together".

Chey posted the photo of him with Jensen on his personal Instagram account, with SK Group's memory affiliate -- SK hynix -- supplying NVIDIA with its ultra-fast HBM3 and new HBM3E memory for its AI GPUs. NVIDIA uses HBM3 on its industry-leading Hopper H100 AI GPU while throwing on ultra-fast HBM3E memory on its new Blackwell B200 AI GPU.

In the future, we can expect HBM4 to appear in 2025 while being used on future-gen AI GPUs in 2026 and beyond, and SK hynix is at the center of that. SK hynix recently announced a $14.6 billion investment on a new memory chip fab plant to meet the "soaring demand" of HBM, while it plans to work with TSMC on next-gen HBM4 memory, which will be used on future-gen AI GPUs from NVIDIA.

OpenAI is the first company to receive the powerful new NVIDIA DGX H200 GPU for accelerating generative AI, which was hand-delivered to CEO Sam Altman and president and co-founder Greg Brockman by NVIDIA CEO Jensen Huang. And for those interested in the leather jacket count, two out of three (Greg Brockman and Jensen Huang, natch) were donning stylish black leather jackets.

Jokes aside, the DGX H200 is NVIDIA's and the world's most powerful AI GPU hardware. Jensen Huang signed the hardware with a simple message: "To advance AI, computing, and humanity."

The DGX H200 includes the new H200 Tensor Core GPU, with the Hopper-based hardware equipped with 141GB of HBM3e GPU memory and speeds of up to 4.8TB/s. Designed to accelerate AI workloads, the DGX H200 is more efficient than its DGX H100 predecessor and substantially more powerful.

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.