✔️ NVIDIA is revising its plans for developing new architectures, and by September, Rubin chips and Vera processors may already be available.
A year ago, it was expected that these new products would not be accessible before next year, but now the timeline has shifted: development is progressing faster, and the new Rubin graphics cards from the R100 series as well as Vera processors could be released as early as this fall. This acceleration has been made possible by the team speeding up the schedule—moving from a yearly update cycle to releases every six months.
Rubin R100 supports HBM4 memory, utilizes TSMC’s 3nm process technology, and implements CoWoS-L packaging. Additionally, this is the first series to adopt chiplet design, quadrupling the die area. Meanwhile, Vera will replace ARM’s Grace processors and will be based on a new generation of ARM cores, promising a significant performance boost. However, this shift in development schedule carries risks: the market may not have time to adapt to the new products, and previous models, as experienced with Blackwell, could face serious issues at launch.
✔️ Google Cloud announced a preview of new G4 virtual servers equipped with NVIDIA Blackwell GPUs.
The company introduced the G4 virtual machine system with RTX PRO 6000 Blackwell graphics cards, marking this as the first public cloud infrastructure deployment of such hardware. Each VM combines 8 GPUs, paired with two AMD Turin processors and 384 virtual cores, along with 1.5 TB of DDR5 memory. Additionally, network accelerators based on Titanium technology provide up to 400 Gbps throughput. Thanks to these components, the virtual machines deliver four times more computing power and six times higher memory bandwidth compared to previous generations.
These systems are well-suited for tasks related to artificial intelligence, rendering, or physics-based simulations. RT cores accelerate realistic ray tracing for high-quality graphics, while NVIDIA’s Dynamo framework aids in working with generative models.
These new virtual machines are planned to be integrated into the AI Hypercomputer system and connected with existing Google Cloud services. They are expected to become available later this year.
✔️ OpenAI’s open-source model with weights has been postponed to a later date.
CEO Sam Altman announced via the X platform that the release of their new language model, originally scheduled for late June, has been delayed until the end of summer. Although the model was initially expected to feature reasoning capabilities, the timelines are now flexible due to unexpected progress by the research team, which requires more time. Altman described these results as “very promising.”
✔️ Disney and Universal Studios jointly filed a lawsuit against Midjourney.
Major film studios have taken legal action against the AI service developer Midjourney. In their lawsuit, they claim that the company used copyrighted images—including characters like Darth Vader and Minions—without permission, even after warnings. The court in California has already registered the case, and both sides are seeking damages, injunctive relief to prevent further use of the intellectual property, and a trial by jury.
Midjourney has not publicly commented on the situation.
✔️ ChatGPT suffered an unexpected defeat against an old-fashioned chess game.
The GPT-4 language model was tested against a classic game from 1977 released for the Atari 2600 console and surprisingly lost even at the early levels. Engineer Robert Caruso conducted the experiment via an emulator and observed that the neural network made gross mistakes—forgetting piece placements, confusing pieces, and failing to handle simple tactical moves.
The Atari software, running on a 1.19 MHz touchscreen with analysis limited to 1–2 moves ahead, easily defeated the neural network. Attempts to modify the piece design or simplify the game did not help—GPT-4 continued to err and promised to win the next game but ultimately resigned.
This interesting comparison illustrates that even the most modern neural networks are still far from mastering game tactics and strategies accessible to casual players who have learned the classics.