GoKawiil
Tech News
D-Matrix introduced Raptor, a new accelerator that stacks compute logic directly on top of DRAM dies to tackle memory bottlenecks in generative AI inference. The design aims to bridge the gap between SRAM's speed but tiny capacity and HBM's larger capacity but bandwidth and power limitations, following the company's earlier Corsair and JetStream products.
Amazon has cut the price of Crucial's 4TB T705 SSD to $502.51, down from nearly $700 a few weeks ago and its lowest price since February. That makes it the most affordable PCIe Gen 5 drive currently available, undercutting the Samsung 990 Evo Plus by about $60 despite offering true four-lane Gen 5 performance.
Newegg is selling a TeamGroup T-Force Delta RGB 32GB (2x16GB) DDR5-5200 memory kit for $389.99 using coupon code LUYRF228, down from $429.99. The kit runs at 5,200 MT/s with 40-40-40-76 timings, making it a mid-tier option rather than a top-speed choice for enthusiasts.
Apple is said to be considering ChangXin Memory Technologies (CXMT), a fast-growing Chinese DRAM maker, as a potential supplier for some of its China-related memory needs. CXMT has not confirmed Apple by name but says it is open to working with top global customers and claims its products can compete with established suppliers on performance and reliability. The move would give Apple added leverage in talks with existing suppliers Micron, Samsung, SK Hynix and Kioxia.
Samsung and TSMC, the world's two largest chipmakers, have confirmed plans to deploy ASML's High NA extreme ultraviolet lithography systems, machines that can cost roughly $400 million each and etch finer circuit patterns onto silicon wafers. Samsung plans to use the tools for DRAM production starting in 2028 and expand its scaling roadmap by 2030, while TSMC expects broader adoption driven by increasingly complex transistor designs needed for AI chips. Both companies join Intel as customers of the technology.
Computer memory listed as 16GB or 32GB RAM refers to DRAM, the temporary working memory that loses its contents once power is cut, requiring constant electrical refreshes to hold data while running. SRAM, used in processor caches like AMD's Ryzen 9 9950X3D with 144MB of cache, is also volatile but doesn't need refresh cycles and offers faster access due to its transistor-based latching design. Both differ fundamentally from non-volatile storage, which retains data without power.
South Korean court documents allege that Chinese memory maker CXMT ran an internal program called Project Hefei, led by a former Samsung DRAM development chief, to systematically acquire Samsung's manufacturing know-how starting in 2016. Prosecutors say the plan involved poaching Samsung engineers, obtaining a 620-step Process Recipe Plan detailing equipment and production methods, and using it to launch DRAM wafer production by 2018.
Taiwanese media report that Micron's two main unions are pressing for larger bonuses after the company's profits surged on AI-driven memory demand. Union chairman Lin Zherui pointed to Micron's record earnings, arguing compensation has stayed capped near 200% of salary even as SK Hynix and Samsung paid workers hundreds of thousands of dollars this year. Reports suggest union membership has grown sharply as frustration builds, raising the possibility of labor action.
SK hynix held a groundbreaking ceremony this week for its first U.S.-based HBM packaging and testing facility, located in West Lafayette, Indiana. The $4 billion site will package HBM stacks using DRAM wafers made in South Korea, with volume production of next-generation HBM memory expected in the second half of 2029, roughly a year later than originally planned.
Samsung unveiled a 16GB LPDDR5X memory package that runs computation directly inside the chip, achieving 614 GB/s of internal bandwidth between its DRAM banks and compute units. That figure is eight times higher than the 76.8 GB/s the same package can push through its external pins, according to a Tom's Hardware report from the conference.