Latest
Seoul National University's Ouroboros reuses ViT computation results by leveraging motion information from video. With Jetson Orin's peak figures as the benchmark, reproducibility against complex motion and device differences will be the key factor for practical adoption.
SK Group Chairman Chey Tae-won warned that soaring memory prices could shrink device demand, and called for expanded supply ahead of a projected surge in AI-related customer demand in 2027. This article examines SK hynix's 1,100 trillion won investment strategy and the time lag before mass production comes online.
Colossal, known for its de-extinction ambitions, is reportedly in talks to raise funding at a valuation of $20 billion to $30 billion. With revenue figures undisclosed, the profitability of its broader biotech business becomes the next key factor to watch.
At SIGGRAPH 2026, NVIDIA presents 21 research works linking 3D environment generation, physics computation, motion generation, and Unitree G1 control. Here's a look at what code has been released and what verification remains before practical deployment.
A 10-core AMD engineering sample believed to be Medusa Point from the Zen 6 generation scored 33.4% higher than the Zen 5-generation average of same-core products in Geekbench multi-core. However, power consumption and clock conditions remain unconfirmed.
Terahertz waves are valuable for imaging and high-speed communication, but the sheer size and cost of equipment have long blocked widespread adoption. A UCLA research team has developed a technology that repurposes quantum well structures—a standard building block in optical communications—to integrate the core functions onto a single chip, opening the door to dramatic system miniaturization and mass production.
Reports indicate that three major memory makers, including Samsung, have scaled back in-house development of CXL controllers. Product development continues, with a shift toward a division of labor that incorporates externally designed control chips.
Germany's retail DDR5 price index rose to 448% in July 2026, up 4.48 times from July 2025. Out of 20 categories, 18 saw their cheapest in-stock price increase month over month.
Microsoft will adopt AMD Helios for Azure's large-scale inference infrastructure. The introduction of the 72-GPU rack has been confirmed, but pricing and availability for ND MI455X v7 remain unannounced.
Reports indicate Google's "Frozen v2" hardwires part of Gemini's architecture, targeting 6-10x the inference throughput per watt compared to TPUs, with deployment possibly as early as 2028. Keeping pace with model updates remains a constraint.
