GoKawiil
Tech News
AMD has submitted patches to its open-source Linux kernel driver adding GDDR7 memory identification along with new IP blocks including IH 8.0 and NBIF 7.10, according to Phoronix. Current Radeon RX 9000-series cards use GDDR6, making this the clearest driver-level sign yet of preparation for a next-generation GPU architecture, widely expected to be RDNA 5.
Digital Realty's chief technology officer Chris Sharp says the data centre industry is moving away from copper for internal networking, though not electrical power or cooling systems. A typical 100MW facility uses around 400 tonnes of copper overall, with up to 20 tonnes used specifically in the wiring that connects servers together. Sharp argues this copper wiring between GPUs and CPUs is now a bottleneck limiting performance.
ReBarUEFI is a UEFI DXE driver that can be injected into a system's firmware to enable Resizable BAR on boards that don't officially support the feature. It works by patching the PCI resource allocation process during boot and is paired with a companion tool, ReBarState, to set the BAR size, with a separate project called NvStrapsReBar covering older NVIDIA Turing GPUs.
Nscale, a London-based cloud provider focused on AI infrastructure, has filed for an initial public offering on the New York Stock Exchange under the symbol NSCL. Its prospectus shows revenue jumped over 1,250% to $140.6 million in the first half of 2026, though the company posted a $1.02 billion net loss for the period and reports $56.4 billion in remaining performance obligations. Nscale rents out Nvidia GPUs to clients including OpenAI, Anthropic and Microsoft, and one undisclosed customer made up more than half of its first-half revenue.
Robotics startup Vsim, founded by former Nvidia Isaac Sim engineers Lu and Storey, has built a high-performance training simulator optimized for modern GPU chips rather than older algorithms from the 1970s and 80s. The resulting software is efficient enough to run directly on the hardware of their robot Freddo, letting it simulate tens of thousands of possible near-future scenarios in real time while moving.
Speaking at a UK summit alongside King Charles III, Nvidia CEO Jensen Huang predicted the company would sell twice as many chips in the coming year compared to this year. This follows Nvidia's recent guidance of roughly 70% revenue growth for its fiscal year ending January 2028, reaching about $673 billion, a figure that already pushed analyst estimates higher.
Leaker Kepler_L2 says most of Nvidia's next-gen RTX 60 series and AMD's RDNA 5-based Radeon 10000 lineup won't launch until 2028, with AMD's midrange AT2 chip as the sole exception still on track for 2026. The claim contradicts an earlier report from Moore's Law is Dead suggesting Nvidia's RTX 6090 could arrive in early 2027, and AMD board partners have reportedly given inconsistent timelines ranging from late 2027 to early 2028.
USB-C hubs can substitute for docking stations by adding extra ports to a laptop, but they draw power and bandwidth directly from the host computer rather than a dedicated outlet or bus. This makes them less reliable for power-hungry or high-bandwidth devices like external SSDs, HD webcams, external GPUs, and multiple monitors.
Mexican federal police, Navy and state forces raided a hidden cryptocurrency mining site in Tlaola, Puebla, seizing roughly 300 GPUs, 80 medium-voltage terminations, a transformer and eight satellite antennas. Investigators believe the operation was illegally drawing power from the nearby Necaxa hydroelectric dam, and it marks the fourth such facility uncovered in the region since early last year.
At the Goldman Sachs Communicopia + Technology conference, Nvidia CEO Jensen Huang repeated the company's guidance that revenue could climb 70% year-over-year next year, building on roughly $400 billion expected this fiscal year to around $680 billion. He described Nvidia's GPU systems as massive, multimillion-dollar installations requiring airplanes to ship, and noted 27% month-over-month order growth for its combined Grace CPU-Blackwell GPU systems.
Inception has released Mercury 2.5, an update to its diffusion-based language model that the company says is 40% more intelligent than Mercury 2 while keeping the same low-latency, low-cost performance. The model runs at over 1,100 tokens per second on standard NVIDIA GPUs, supports a 260K-token context window, and launches at an 80% discount, priced at $0.04 per million input tokens and $0.15 per million output tokens.
AMD published results from experiments running speculative decoding in vLLM on its MI300X and MI355X Instinct GPUs using the ROCm software stack. The team compared five drafting approaches—native MTP, Gemma 4 MTP, EAGLE-3, DFlash, and DSpark—which differ in how draft tokens are generated and how they receive signals from the target model.