Artificial Intelligence - Page 23

NVIDIA had a few hiccups with its original Blackwell-based GB200 NVL72 AI servers, with cloud service providers (CSPs) struggling to ink deals for new "Blackwell Ultra" GB300 AI servers, wanting more "mature" AI GPU solutions instead.

In a new report from Ctee and post on X by @Jukanlosreve, we're hearing that major CSPs like Microsoft and others are moving away from ordering GB300, and after the sour taste that GB200 AI servers left in their mouths, they're ordering mature HGX 8-card series products. Industry insiders analyzed that GB300 samples wouldn't be made available for customer testing until Q4 2025 at the earliest, meaning we could expect a delay for GB300 into 2026.

Microsoft has been rumored to cut its computing power at its data centers, with supply chain insiders reporting that Microsoft was one of the first companies to get their hands on NVIDIA's new GB200 NVL72 AI servers, but "due to poor yields, its customers are forced to join the testing process. Among them, the complexity of deploying the system was far beyond expectations. The installation process took about 5 to 7 working days, and instability and system crashes often occured".

It looks as though SK hynix has sold out of most of its HBM memory chips for 2026, with all of its 2025 supply exhausted... and with NVIDIA unstoppable with Blackwell Ultra GB300 and its HBM3E, and its next-gen Rubin R100 with HBM4, the South Korean memory leader is swamped, but in the best way possible.

SK hynix held its annual shareholders meeting a couple of days ago, with CEO Kwan Noh-jung revealing that the company has been wrapping up sales discussions of its HBM memory chips for 2026. Kwak said: "We will finalize discussions with customers on next year's HBM volumes by the first half of this year to further strengthen sales stability".

The company recently sent out samples of its next-generation HBM4 memory chips, with mass production expected in the second half of 2025, too. Kwak added: "Despite rising uncertainties in the global economy, major tech companies are increasing investment to gain leadership in AI. Demand for HBM is expected to increase sharply due to demand for GPUs and other chips".

Since OpenAI rolled out its new image generation feature in ChatGPT thousands of users are taking to social media platforms to share recreations of iconic events in history, movies, TV shows in the art-style of Japanese animation studio, Studio Ghibli.

The tsunami of Studio Ghibli-style images began when OpenAI launched what it describes as the "most advanced image generator yet," which includes capabilities of producing "precise, accurate, photorealistic outputs." For those who are unfamiliar with Studio Ghibli, the animation studio is behind renowned films such as Spired Away, How's Moving Castle, and My Neighbor Totor, all of which are directed by Hayao Miyazaki.

The new addition to OpenAI's GPT4o model enables users to recreate images in the style of the animation studio, which has been abundantly used to recreate some of the most viral memes ever created on the internet.

As outlined in a recent 10-K filing, computer maker Dell has been shring in terms of staff volume, with 25,000 of its employees no longer on the payroll as of January 2025.

As reported by Business Insider, the company said that it had 133,000 staff as of February 2023. According to the recent filing, the figure currently stands at 108,000 global employees as of January 31, 2025. This represents a total reduction of 25,000, marking a 10% shrinkage every year. Altogether, the company's workforce has reduced by 19% over the two-year period.

In August 2024, Dell significantly restructured its sales divisions, laying off over 20,000 workers. The laid-off staff were told the restructure was to prepare for "the world of AI". The company also began issuing return-to-office mandates last year, eventually ordering all staff within 90 minutes of Dell offices to return full-time as of January 2025.

Concerns about job displacement continue to mount, as OpenAI, Anthropic, and the broader market continue to roll out new models seemingly on the daily. Bill Gates has been relatively vocal about AI's potential to replace humans in most markets and professions. However, according to the Microsoft co-founder, there's a few exceptions.

Coding is widely considered one of the first professions on the chopping block due to AI. Major tech leaders, including NVIDIA's Jensen Huang, OpenAI's Sam Altman, and Salesforce CEO Marc Benioff, have each been insistent that programming will cease to exist when AI reaches a particular stage. However, Gates takes a different stance, explaining that human oversight is essential for identifying and correcting errors, refining algorithms, and helping to improve AI development.

Apple has reportedly placed a large order for $1 billion worth of NVIDIA's new GB300 NVL72 AI servers, powered by its beefed-up Blackwell Ultra AI GPUs according to Loop Capital.

Loop Capital's note explained that Apple is "officially" in the large server cluster Gen AI game, with Super Micro and Dell being Apple's key server partners. Apple's continued Siri AI stumbles might have seen an internal shift that is putting even more importance on AI, and that requires the best AI hardware on the market right now: NVIDIA's new GB300 NVL72 AI servers.

NVIDIA's new GB300 NVL72 AI servers cost around $3.7 million to $4 million each, so we're looking at around 250 new NVIDIA GB300 NVL72 AI servers, made by both Super Micro and Dell. This should see Apple with a Gen AI LLM cluster, which should change things up for Apple as the company seems to be having issues with its AI/ML (machine learning) versus GenAI.

A new breakthrough artificial intelligence-powered detection models has caught one of the most challenging cancers to catch, Endometrial cancer, with reports now stating accuracy with the mode has reached up to 99% in its detection rate.

The new AI model named ECgMLP processes visual data, and according to reports, it set a new precedent for AI cancer detection as it is capable of taking the data given to it by doctors, filter out any unnecessary noise, and highlight areas within the data that it deems as a potential risk for cancer. In the case of the data being given to the AI, it would be scans of areas of the human body, and having an AI be able to thoroughly go over multiple scans quickly and accurately will considerably speed up the rate of cancer detection and its accuracy.

For example, ECgMLP was able to detect Endometrial cancer, also known as uterine cancer, which is particularly hard to detect early as most of its symptoms can be subtle or mimic other conditions. The model uses digital pattern recognition techniques to evaluate areas of tissue and then provide a diagnostic prediction. Reports indicate most other automated systems for detecting endometrial cancer have about 80% accuracy, but ECgMLP beats those systems by approximately 20%, making it one of the most accurate AI cancer detection models created so far.

Micron's new SOCAMM, a modular LPDDR5X memory solution, was developed in collaboration with NVIDIA to support its just-announced GB300 Grace Blackwell Superchip announced at GTC 2025 this week.

Micron's new modular SOCAMM memory modules provide over 2.5x higher bandwidth at the same capacity as RDIMMs, and they're super-small, occupying just one-third the size of the industry-standard RDIMM form factor. Thanks to LPDDR5X, the new SOCAMM modules use one-third the power of regular DDR5 DIMMs, while SOCAMM placements of 16-die stacks of LPDDR5X memory enable a 128GB memory module: the highest-capacity LPDDR5X memory solution, which is perfect for training large AI models and more concurrent users on inference workloads.

Raj Narasimhan, senior vice president and general manager of Micron's Compute and Networking Business Unit, explained: "AI is driving a paradigm shift in computing, and memory is at the heart of this evolution. Micron's contributions to the NVIDIA Grace Blackwell platform yields significant performance and power-saving benefits for AI training and inference applications. HBM and LP memory solutions help unlock improved computational capabilities for GPUs".

Razer has unveiled its latest AI-powered venture, Project Ava, which aims to become the smartest gaming companion ever created.

Originally announced at CES 2025, ProjectAva provides features that aim to assist the player with every aspect of the experience in-game. In-game walkthroughs allow the user to ask questions and get advice on how to progress through missions while in the game. One-click PC optimization claims to offer expertise with optimizing your game's graphics settings. Real-time esports coaching offers competitive analysis and insights to improve the player's performance in multiplayer experiences.

The focus on competitive play is the key differentiator for Razer's gaming copilot. The video example showcases the assistant pointing out enemy rotations in a game of League of Legends, and real-time instruction on how to counter a boss in Black Myth: Wukong. The focus on multiplayer, e-sports style applications is the backbone of the copilot's pitch. While the demonstrations are impressive, the real question is whether any of it will actually work.

Internal studies from OpenAI and MIT have been exploring how chatting to ChatGPT may affect the emotional experience of its users. In recent findings - which draw on over 40 million ChatGPT interactions and four weeks of human observation - there were some interesting takeaways.

The study from MIT found that chatbot interactions, whether through text or voice, were associated with higher levels of loneliness, reduced socialization, and emotional dependence. It highlights that participants who already trusted the chatbot, or were prone to emotional attachment in human relationships, were more likely to experience these effects. While the impact was less severe in ChatGPT's voice mode, the study found that both personal and general conversations could lead to emotional dependence and feelings of loneliness.

OpenAI's study, however, found that emotional conversations with ChatGPT were the exception rather than the norm. The researchers highlighted that such interactions were prominent for 'only a small group of the heavy Advanced Voice Mode users' that were studied. Meaning, that while there are links to adverse emotional effects found in the MIT study, they only made up a small subset of the broader population.

Google is continuing its push into artificial intelligence being injected into as many of its apps and services as possible, and the latest piece of Google software to get a touch of AI is Gmail.

The search engine giant has announced via a new blog post that Gmail is being upgraded with a "smarter search feature powered by AI." The blog post explains the new and improved Gmail will show users the "most relevant results faster," through what is called "smarter search." This upgrade is intended to help users find information buried in their inbox much faster, and with the power of AI users will now be presented with emails that aren't just based on keywords.

According to the blog post, up until now, Gmail has been showing searched for emails based on keywords and chronological order, but now the search feature will factor in elements such as recency, most-clicked emails and frequent contacts. With the inclusion of these new factors, emails users are looking for are more likely to appear at the top of your search, which will, in turn, save users significant time.

The former boss of Intel attended NVIDIA's AI-focused GTC 2025 conference in San Francisco, California. In a 'Live at NVIDIA GTC' video appearance, ex-Intel CEO Pat Gelsinger shared some words about NVIDIA's meteoric rise in recent years, thanks to the current AI boom. Aimed at NVIDIA CEO Jensen Huang, Geslinger says, "he got lucky with AI" - and it's not a dig.

The fortunes or fall of Intel under Pat Gelsinger's stewardship have been widely reported. In 2025, the once-dominant force in consumer and enterprise processor hardware is now playing catch-up to companies like NVIDIA and AMD. Regarding NVIDIA being the leader in AI, Pat Gelsinger also talked about Intel's failed Larrabee project, which attempted to bring GPU-like acceleration for things like AI to a traditional x86 CPU.

"The CPU was the king of the hill, and I applaud Jensen for his tenacity of just saying, 'No, I am not trying to build one of those'; I am trying to deliver against the workload starting in graphics," said Gelsinger.

Various firsts for AI are inevitably going to emerge, and here's a new one: the world's first newspaper written by an AI.

Yes, it's happened, courtesy of an Italian newspaper, Il Foglio, which has produced an AI-generated edition as highlighted by The Guardian.

Certainly as far as we're aware, this is a first of its kind, but this isn't a case of the paper's owners letting AI write the entire thing. Rather, it's a four-page supplement which is written by AI.

SK hynix has started sampling its next-gen 12-Hi HBM4 memory modules to customers -- namely NVIDIA, which will be using HBM4 on its new Rubin AI GPUs -- the world's first HBM4.

During NVIDIA's recent GTC 2025 event, AI memory partner SK hynix announced it was unveiling its new 12-Hi HBM4, 12-Hi HBM3E (for NVIDIA's new GB300 AI GPU), and new SOCAMM memory modules.

SK hynix's' new HBM4 memory uses its Advanced MR-MUF process, pushing capacities up to 36GB which makes it the highest amongst 12-Hi HBM products. We know that NVIDIA's new Rubin and Rubin Ultra AI GPU platforms will use HBM4 memory, thanks to its high speeds we have up to 2TB/sec for the first time, a 60% performance jump over HBM3E

NVIDIA has just unveiled its new RTX PRO 6000 "Blackwell" graphics card, based on the same GB202 GPU and GDDR7 memory that the RTX 5090 uses, but with a whopping 96GB of GDDR7 memory.

The new NVIDIA RTX PRO 6000 Blackwell GPU features the GB202 GPU with 24,064 cores inside of 188 SMs, compared to the 21,760 cores and 170 SMs inside of the same GB202 GPU in the GeForce RTX 5090. The full-fat GB202 GPU features 192 SMs, so we're not even maxed out here with the new RTX PRO 6000.

We do have the full 600W of TDP support through a single 12V-2x6 power connector, offering 25W more TDP than the 575W on the RTX 5090. With all of this in tow, NVIDIA's new RTX PRO 6000 Blackwell GPU features 4000 AI TOPs, 125 TFLOPS of FP32 compute performance, and 380 TFLOPS of RT performance (a 19% increase over the RTX 5090 in all performance figures).

NVIDIA has just announced that its next-gen Feynman GPU architecture will succeed Rubin and Blackwell, and will debut with ''next-gen HBM" memory in 2028.

At its GPU Technology Conference (GTC 2025) event, NVIDIA unveiled its latest AI GPU architecture roadmap, extending beyond the just-detailed Rubin and Rubin AI GPUs. NVIDIA has now confirmed that its next-gen AI GPU will be codenamed Feynman, after renowned American theoretical physicist, Richard Phillips Feynman, who is known for his work on quantum mechanics

On his Wikipedia page, Feynman is explained as "an American theoretical physicist. He is best known for his work in the path integral formulation of quantum mechanics, the theory of quantum electrodynamics, the physics of the superfluidity of supercooled liquid helium, and in particle physics, for which he proposed the parton model".

NVIDIA is hosting its GPU Technology Conference (GTC 2025) with an unveiling of its GB300 Blackwell Ultra, and a tease of its next-gen Rubin and Rubin Ultra AI GPUs, and its next-gen Vera CPUs... new ultra-fast AI platforms for AI computing deploying in 2026-2027.

NVIDIA has increased performance by 50% with its new GB300 "Blackwell Ultra" AI systems over GB200 AI servers, as well as increased memory capacity to 288GB HBM3E on GB300 from 192GB of HBM3E on GB200. The new GB300 and current GB200 AI platforms are made into up to NVL72 solutions, but next-gen Rubin GPUs will scale up to a far larger NVL144 platform.

The new NVIDIA Vera Rubin NVL144 platform will use two new chips, with the Rubin GPU using two Reticle-sized chips with up to 50 PFLOPs of FP4 performance and 288GB of next-gen, ultra-fast HBM4 memory. Alongside these chips will be an 88-core Vera CPU with a custom Arm architecture, 176 threads in total, with up to 1.8TB/sec of NVLINK-C2C interconnect.

NVIDIA has officially unveiled its beefed-up GB300 "Blackwell Ultra" NVL72 AI server, with its new GB300 AI GPU featuring 50% more performance over GB200, and a larger pool of 288GB of HBM3E memory.

NVIDIA's new GB300 "Blackwell Ultra" AI GPUs will continue the AI domination led by GB200, with memory capacity increased through using new 12-Hi HBM3E memory stacks, and even more compute power for AI workloads. NVIDIA is also coupling GB300 with the latest Spectrum Ultra X800 Ethernet switches (512-Radix).

The new NVIDIA GB300 NVL72 rack-scale solution has 1.5x more performance than an NVIDIA GB200 NVL72 AI server, as well as increasing Blackwell revenue opportunities by 50x for AI factories compared to AI servers built on the Hopper GPU architecture.

NVIDIA's GTC conference has officially kicked off, and Phison, one of the world's leading companies in controllers for NAND flash memory chips, has unveiled what it's describing as an affordable path to supporting a 1 trillion parameter AI model.

The race to the first 1 trillion parameter AI model is nearing completion, at least according to Phison, which predicts we will see the first 1 trillion parameter AI model before 2026. Phison has briefly outlined the progression of AI models, with a 69 billion parameter model unveiled in 2023 with Llama 2, then a big jump to the 405 billion parameter Llama 3.1 model in 2024, and then the 671 billion parameter DeepSeek R3 model in 2025. Keeping on this same trajectory, Phison expects the world's first 1 trillion parameter model will be unveiled before the end of 2025.

With the increase in the size of these AI models, the hardware needed to train and support them will also need to be increased, which is why Phison is looking at ways to reduce operational costs, and the company has a solution. Phison expects the operational costs of a 1 trillion parameter model will be approximately $3 million in raw GPU power, but the company has outlined a plan to alleviate the cost down to just $100,000. How? A combination of SSDs, Phison's aiDAPTIV+ software, and an NVIDIA GH200 Superchip.

The world's first Large Language Model (LLM) training laptop is being showcased at GTC 2025, and it's a creation between Phison and Maingear.

Phison Electronics, a world leader in NAND flash technology, has teamed up with Maingear to bring its aiDAPTIV+ software stack to Maingear's ML-series laptop, creating the first AI laptop PC capable of Large Language Model Operations. The new laptop, which is being demoed at GTC 2025 with an RTX 4090 laptop GPU, will feature an NVIDIA RTX 5090 laptop GPU when it is released for sale. As for its capabilities, Phison has announced it will be able to train and inference LLMs that are up to 8 billion parameters, while also having support for fine-tuning, which enables users to inject their own data into the AI model.

Why does this laptop exist? Phison is setting out to reduce the cost of running AI models by offloading the processing, which is currently on high-cost GPU memory, to low-cost NAND flash storage, and by doing this the operational cost of the AI model shrinks, enabling larger models to be trained on cheaper devices. By utilizing the power of the company's aiDAPTIV+ software, users will now be able to customize an AI model with their own data, all within a device that can fit in your backpack.