Artificial Intelligence - Page 46

Microsoft recently announced it was recalling its Windows Recall feature, which uses artificial intelligence and screenshots to enable users to access different times in the past on their Copilot+ enabled PC.

For those who don't know, Windows Recall is one of Microsoft's AI-powered features. The company said it would be released alongside the selection of the new Copilot+ PCs. However, the feature was quickly criticized by many for its impact on user privacy and security, as Recall only works by taking screenshots of a user's desktop - indiscriminate of what is on the screen. This means any sensitive user information such as passwords, financial information, and private content would be captured and stored on the PC.

Security researchers were quick to criticize the feature, saying it was a hacker's dream to have a target's information all aggregated in one file directory. There is also the more simple criticism, which is the feature enables Microsoft to "watch" everything a user does on their device, even if the company claims that none of the screenshots are being uploaded to their servers and that the images are stolen entirely on the user's device.

SIGGRAPH 2024 begins on July 28, where NVIDIA CEO Jensen Huang will deliver a 'fireside chat' at the show, which is the 50th anniversary of SIGGRAPH, which first took place in 1974.

SIGGRAPH, or Special Interest Group on Computer Graphics and Interactive Techniques, is an annual conference that looks into the world of computer graphics. We're talking academics, developers, and GPU makes like NVIDIA and AMD. SIGGRAPH takes place on July 28, running through to August 1.

The official SIGGRAPH page explains Jensen: "Jensen Huang founded NVIDIA in 1993 and has served since its inception as president, chief executive officer, and a member of the board of directors. Huang has been elected to the National Academy of Engineering and is a recipient of the Semiconductor Industry Association's highest honor, the Robert N. Noyce Award; IEEE Founder's Medal; the Dr. Morris Chang Exemplary Leadership Award; and honorary doctorate degrees from National Chiao Tung University, National Taiwan University, and Oregon State University".

Miles Astray entered into a top photo competition with his work titled "F L A M I N G O N E," which received honors from a high-profile jury that included members from the New York Times, Getty Images, Phaidon Press, and more. However, the artist was disqualified after he won the category.

Notably, the competition was under the category of Artificial Intelligence-generated photos, and Astray's real photograph has now been deemed the "first real photo to win an AI award". The reason behind the artist entering a real photograph into the AI category was to demonstrate that people are unable to tell the difference between an AI-generated image and a real photograph and the ethical and societal implications of the emerging technology.

This isn't the first time AI and real photograph has caused a stir in competions, as AI-generated photos have made international headlines for winning photo competitions, which they weren't meant to be competing in Probably the most infamous of these stories is when the Sony World Photography Awards awarded Boris Eldagsen a prize in the Creative category of the 2023 Open competition for his AI-generated image.

Samsung Electronics has unveiled its latest foundry innovations and outlined its vision for the "AI era" during its Samsung Foundry Forum (SFF) event, which it holds annually at its Device Solutions America HQ in San Jose, California, USA.

The theme was "Empowering the AI Revolution" to which Samsung reinforced its process technology roadmap, including two new cutting-edge nodes -- SF2Z and SF4U -- as well as its integrated Samsung AI Solutions platform that holds the "unique strengths" of its Foundry, Memory, and Advanced Package (AVP) businesses.

Dr. Siyoung Choi, President and Head of Foundry Business at Samsung Electronics, said: "At a time when numerous technologies are evolving around AI, the key to its implementation lies in high-performance, low-power semiconductors. Alongside our proven GAA process optimized for AI chips, we plan to introduce integrated, co-packaged optics (CPO) technology for high-speed, low-power data processing, providing our customers with the one-stop AI solutions they need to thrive in this transformative era".

TSMC is absolutely dominating, and we all know it. The big AI chip players are eating up its 3nm production into 2026, and now upstream IC design companies are reporting price increases.

NVIDIA, AMD, Intel, Qualcomm, MediaTek, Apple, and Google are all working on getting chips made on TSMC's new 3nm process. Ctee reports that Qualcomm's new Snapdragon 8 Gen4 processor is made on TSMC's new N3E process, and its quotation has increased by 25% compared to the previous generation, which Ctee reports "does not rule out subsequent price increases."

Another issue is that semiconductor competitor Samsung led the mass production of 3nm chips using the GAA (Gate All Around) process on June 22, but the first-gen N3 node "SF3E" was not successful at all. The yield rates of Samsung's in-house Exynos 2500 chips dropped, and Google Tensor processors are all made by Samsung. The current fourth-generation Tensor SoC uses Samsung's new 4nm process, but the rumored fifth-generation Tensor chip will switch to TSMC's new 3nm process.

The world of AI chips hasn't stopped, and won't stop, with advanced packaging production capacity in short supply.

TSMC's new CoWoS plant in Nanke Chiayi Park is entering the "environmental review stage," reports UDN, with TSMC now purchasing equipment to help speed up the construction of the new CoWoS advanced packaging facility. At the same time, the two new CoWoS advanced packaging factories were planned to be built in Nanke Chiayi Park, but they're "not enough," reports UDN.

TSMC reportedly sent personnel down to southern Taiwan to scout for land for a third CoWoS advanced packaging facility. UDN reports that the Chiayi County Government previously announced that TSMC's new advanced packaging plant will be located in Nanke Chiayi Park, covering an area of around 20 hectares. The first CoWoS advanced packaging plant was planned to take up around 12 hectares of space, will be completed by 2026, and will create 3000 jobs.

NVIDIA has had a mega-successful last 12+ months riding the AI wave, with data center GPU shipments hitting 3.76 million units according to semiconductor analyst firm TechInsights.

NVIDIA has an absolutely dominant 98% market share in AI GPU shipments for 2023, with 3.76 million AI chips shipped. However, if you add AMD and Intel AI processor shipments, the total number only goes to 3.85 million. This means that AMD and Intel shipped only 710,000 AI accelerators, compared to the 3.76 million shipped by NVIDIA alone.

This also means that NVIDIA has 98% market share of AI GPU revenue, with $36.2 billion -- that's 3x the revenue of 2022, which stood at $10.9 billion -- while AI processor alternatives are coming from Google with its TPUs, AMD with its Instinct AI accelerators, Intel's Gaudi AI accelerators, and CPUs that feature NPUs for AI workloads.

Just before the weekend, Microsoft announced it was making changes to the controversial incoming Windows 11 features that will be exclusive to the company's Copilot+ PCs.

That feature is Windows Recall, which is a Copilot+ PC tool that enables users to access different times on their PC through screenshots the PC is continuously snapping. These snapshots are of the users desktop and don't hide any potential sensitive information, and are also stored in a single folder. Security researchers and the public raised serious concerns with Microsoft about the potential of hackers gaining access to the folder and the ramifications that could follow.

Microsoft has since responded and said it is adding additional security measures and encryption to Recall, making it much harder for a bad actor to gain access to Recall's data stored on a user's PC. Now, Microsoft has quietly pulled the build of Windows 11 24H2 update, which contains the Recall feature and was being tested in its Release Preview channel - usually the last stop before a Windows update is rolled out the public.

Apple unveiled its new ChatGPT-powered "Apple Intelligence" during its WWDC 2024 event, with a slew of new AI features... with smartphone competitor Samsung taking to its X account and posting this:

Samsung said: "Adding 'Apple' doesn't make it new or groundbreaking. Welcome to AI" taking a stab at Apple's just-announced Apple Intelligence. Apple Intelligence will launch with iOS, iPadOS, and macOS later this year... but most of these AI-powered features are already on plenty of Android-powered smartphones.

Samsung wasn't finished with the 'mean tweets' with another post on X that read "Moving my icons wherever I want since 2010... isn't that cool?" If you thought it was just Samsung jumping in on mocking Apple and its new ChatGPT-powered Apple Intelligence, you'd be wrong... smartphone maker Nothing also got on the mocking train, with this simple post on X:

The PC has been reborn -- at least according to Qualcomm -- with the wave of Copilot+ PCs with NPUs for AI workloads, and the biggest winner of them all? TSMC.

TSMC will manufacture Intel's next-gen Core Ultra 200 series "Lunar Lake" CPUs, AMD's new Ryzen AI 300 series APUs, the new Ryzen 9000 series "Zen 5" processors, and Qualcomm's new Snapdragon X series processors.

Qualcomm's new Snapdragon X Elite is powered by TSMC's new N4 process node, some of the Tiles on Intel's new Lunar Lake chips are made on TSMC N3B, while AMD's new Zen 5-based processors will be made on TSMC's 4nm process node.

NVIDIA's next-gen GB200-based AI server cabinets are expected to ship in "small quantities" in Q4 2024 at the earliest, with GB200 AI server cabinets to be fully unleashed in 2025.

In a new report from Ctee, we're learning that there will be a "slight gap" in the delivery time between each factory and the customer and that it "will not be available until next year." The outlet continues, saying that shipments can be expected, with multiple Taiwanese manufacturers "striving to increase profitability as their revenue scale increases".

NVIDIA's new GB200 Superchip is a monster, with the company supplying the front-end AI chip modules, Switchboards, power modules, chassis/cabinets, and more. It's expected to see the NVIDIA GB200 Superchip enter mass production in Q3 2024, with "legal persons" talking to Ctee, with the outlet reporting: "Hon Hai, the parent company of Hongbai, and Wistron, as well as Delta Electronics, Lite-On, Shengming Electronics, Yingguang, etc., shipments of GB200-related products will increase in the second half of the year and will further demonstrate strong growth momentum".

Apple has just unveiled its new Apple Intelligence system; in a new partnership with OpenAI, Apple Intelligence will integrate ChatGPT directly into the heart of Siri.

Apple Intelligence will help people with their everyday lives, create images and text, and privacy improvements, which is an important part for Apple. Apple Intelligence is Apple's multimodal, cross-platform approach to the future of AI computing, with generative AI features including text and image creation, an improved Siri assistant, and more.

The company says that everything you do withing the ChatGPT-powered Apple Intelligence won't be shared with anyone, including Apple. Most of the new Apple Intelligence features run locally on your device, but if more computing power is required, then it will go to the cloud. It features a new Power Cloud Compute standard that protects your data and privacy.

Well, well, well... it looks like Chinese companies that haven't been able to buy high-end AI GPUs like NVIDIA's AI stomping H100 AI GPU, are buying and renting AI GPUs but using them on US soil to skirt around US sanctions banning high-end AI chips from the country.

The Information reports that ByteDance, the Chinese company that owns TikTok, has been renting out the best AI GPUs from NVIDIA from Oracle to use for AI. The news outlet reports from two anonymous sources that China Telecom has been making similar moves with other cloud computing providers.

Another source said that cloud companies Alibaba and Tencent have been reportedly talking with NVIDIA about setting up data centers in the United States, in order to get their hands on the best AI GPUs. The Information reports that two, unnamed (but smaller) cloud companies in the US had declined to rent NVIDIA's current-gen H100 AI GPUs because it "seemed to go against the spirit" of US sanctions.

GIGABYTE had some new GeForce hardware on display at Computex, but the dual-slot GeForce RTX 4070 Ti SUPER AI TOP GPU wasn't built to run the latest games. Sporting a blower-fan cooling setup in a stylish silver and metallic shroud is about powerful and secure AI training and tools on a local setup without tapping into the cloud.

It's a complete solution from GIGABYTE. On the software side, you get the AI TOP Utility, AI TOP Software, and AI TOP Tutor. Using popular and powerful open-source models, you can train and query generative AI based on local data sources using powerful GIGABYTE AI TOP hardware that will include additional gear like workstation motherboards, memory, and power supplies.

AI TOP software already supports open-source LLMs, with GIGABYTE explaining that its software and hardware have been designed for professional and home use. This would be awesome to have in a room somewhere to play around with and create different AIs.

The president of SK hynix and the chairman of SK Group met with TSMC chairman and CEO C.C. Wei to discuss, and solidify their partnership in the AI semiconductor business in Taipei, just two months after SK hynix and TSMC shook hands on next-gen HBM4 memory and next-gen semiconductor packaging technology.

The two technology leaders discussed the progress and details of their on-going partnership, with TSMC contract manufacturing SK hynix's HBM memory chips, which are the key part to NVIDIA's dominance in the AI GPU market with its Hopper H100, H200, and Ampere A100 series AI GPUs.

Under the agreement announced back in April 2024, the two companies promised to use TSMC's advanced manufacturing process in the base die of next-gen HBM4 memory, which will enter mass production in 2025, one year ahead of schedule.

Micron is taking a big HBM memory chip business opportunity according to new reports, where the US-based company plans to secure 20-25% of the HBM3E market share by August 2025.

HBM is one of the hottest markets right now thanks to the insatiable demand for more and more AI GPU computing power, with HBM makers like SK hynix, Samsung, and Micron all fighting over each other to be the best. As it stands, SK hynix is far out in the lead, but Micron is making progress, and so too is Samsung which is now working with AMD on providing its HBM3E for their new Instinct MI325X AI accelerator which sports 288GB of HBM3E memory.

Micron said the progress in its HBM3E could be contributed through its advanced packaging and design capabilities, as well as the integration of its own processes. Micron is also working on next-generation HBM4 memory, which we'll see debut in 2025 and beyond for next-gen AI GPUs.

AMD announced its new Instinct MI series AI accelerators at Computex 2024 earlier this week, but now we're hearing the company is cozying up to Samsung to get its fix of HBM memory for its upcoming AI accelerators.

During AMD's Computex 2024 keynote speed, CEO Dr. Lisa Su met with Seoul Economic Daily to talk about their HBM collaboration with South Korean giant Samsung. Samsung has now been confirmed to provide AMD with its fifth-generation HBM memory, also known as HBM3E.

On the same day, AMD announced its upcoming Instinct MI325X AI accelerator, which will feature a monster 288GB of HBM3E memory, that we now know will be provided by Samsung. Back at ISSCC 2023, the world's largest semiconductor technology conference held in San Francisco, California, AMD CEO Dr. Lisa Su announced it was working with Samsung on developing HBM-PIM (processing-in-memory).

NVIDIA had quite a lot to announce at Computex 2024, but one of the biggest teases are future-gen AI GPU architectures including something we've been hearing rumbles for a while now: Rubin.

NVIDIA CEO Jensen Huang revealed the next-gen Rubin GPU architecture, named after American astronomer Vera Rubin, who made significant contributions to the understanding of dark matter in the universe, while also pioneering work on galaxy rotation rate.

The thing is, NVIDIA hasn't even got its new Blackwell AI GPUs out into the wild yet, but we're already seeing Rubin announced. NVIDIA's new Blackwell B100 and B200 AI GPUs will arrive later this year, but the company has supercharged versions that will feature 12-Hi HBM3E memory stacks across 8 sites versus the 8-Hi HBM3E memory stacks across 8 existing products, we should see these beefed-up Blackwell AI GPUs arrive in 2025.

AMD had some big reveals at Computex 2024 including its next-gen Zen 5 architecture, but new AI accelerators were also teased. This includes the Instinct MI325X "CDNA 3", the MI350X "CDNA 4" and the next-gen MI400 "CDNA Next" AI accelerators for data centers, and the cloud.

AMD's new Instinct MI325X AI accelerator will use the same CDNA 3 architecture as the just-launched Instinct MI300 series AI accelerators, with a monster 288GB of HBM3E memory with up to 6TB/sec of memory bandwidth. We'll see 1.3 PFLOPs of FP16, and up to 2.6 PFLOPs of FP8 compute performance, with up to 1 trillion parameters per server.

AMD compared its upcoming Instinct MI325X AI accelerator against NVIDIA's new beefed-up Hopper H200 AI GPU:

Tesla and SpaceX CEO Elon Musk has said that he wants to buy 300,000 of NVIDIA's new Blackwell B200 AI GPUs to train xAI by next summer.

In a new post on X, Musk said that the new B200 AI GPUs will be a (huge) upgrade to the X's current AI GPU cluster, which is powered by 100,000 previous-gen Hopper H100 AI GPUs.

NVIDIA's new Blackwell B200 AI GPU has around 4x the AI training performance of H100, and an incredible 30x the inference performance. NVIDIA's new Blackwell B200 AI GPU uses more power than H100... an important part to remember when you've got hundreds of thousands of power-hungry AI GPUs in an AI supercomputer cluster. But, the monster performance improvements are worth it.