Apple Debuts M6 and M5 Ultra Chips With Major Performance and AI Gains
August 25, 2026
Apple unveiled two new chips, the M6 and the M5 Ultra, on August 25, 2026.
The M6 is Apple's first chip built on a 2-nanometer process, with a 12-core CPU (2 super cores, 4 performance cores, 6 efficiency cores) and a 12-core GPU with Neural Accelerators.
The M6 introduces a dual 16-core Neural Engine delivering up to 2x the peak compute of prior generations, and Apple describes it as delivering the fastest single-threaded chip performance available.
The M5 Ultra uses next-generation UltraFusion technology to fuse four dies into a quad-die design for the first time in Apple's M-series, with up to 36 CPU cores and 80 GPU cores.
The M5 Ultra supports up to 512GB of unified memory at 1.2TB/s bandwidth, 50% more than the M3 Ultra; the M6 ships in the new Mac mini and the M5 Ultra in the new Mac Studio.
Intel Details Three New Processor Designs for Agentic AI at Hot Chips 2026
August 24, 2026
Intel presented three new processor architectures at Hot Chips 2026, spanning data-center to edge deployments for agentic AI.
Diamond Rapids, the enterprise server chip, offers up to 256 cores, 1.28GB of last-level cache, 16 memory channels at 12800 MT/s, and 128 PCIe Gen6/CXL 3.0 lanes, built on Intel 18A-P with Foveros Direct 3D packaging.
Crescent Island, Intel's next-generation datacenter GPU, is built for inference economics with 32 Xe cores, 256 XMX engines, and up to 480GB of LPDDR5X memory in a 350-watt air-cooled design.
Wildcat Lake (Intel Core Series 3) brings AI to mainstream laptops and edge devices via the Intel 18A process, pairing performance and efficiency cores with Xe3 graphics and an NPU rated up to 17 TOPS.
Intel CTO Pushkar Ranade said agentic AI is reshaping chip design "from the transistor through full system architecture," and the new chips lean on Intel Foundry technology and the UCIe open chiplet interconnect standard.
Penn State Engineers Built an Ultra-Efficient Computer Memory Device
August 17, 2026
Penn State researchers combined synthetic, silver-doped DNA sequences with crystalline perovskite semiconductor thin films to build a bio-hybrid memristor (memory resistor).
The device retains a record of the direction electrical current previously flowed even after power is removed, mimicking how neurons store information.
The memristor reportedly consumes roughly 100 times less power than traditional storage devices while offering higher storage density than devices like flash drives.
It operated reliably at under 0.1 volt, kept working at temperatures approaching 250°F, and stayed functional at room temperature for more than six weeks.
The work, published in Advanced Functional Materials and covered by a pending patent, targets low-power neuromorphic computing for AI.
Aalto University Debuts First Cyclic Quantum Heat Engine Built Into a Superconducting Circuit
August 14, 2026
Aalto University (Finland) physicists led by Mikko Möttönen built the first cyclic quantum heat engine inside a superconducting circuit, combining a transmon qubit, a resonator, and a quantum refrigerator.
Near absolute zero, the device repeatedly converted small amounts of heat into measurable positive work via a quantum version of the Otto thermodynamic cycle.
A single tunable quantum-circuit refrigerator supplied both heating and cooling, simplifying the design versus engines needing separate hot/cold reservoirs.
Published in Nature Communications, seen as a step toward autonomous on-chip qubit readout without long microwave cable runs.
Researchers say this could cut both cost and noise as quantum computers scale toward Finland's goal of ~1,000 logical qubits by 2035.
Sub-nanometer interface engineering opens a path for transistors to surpass silicon's limits
August 9, 2026
NYCU researchers addressed a persistent atomic-interface bottleneck limiting atomically thin semiconductors.
They deposited an ultrathin epitaxial aluminum layer onto monolayer MoS2, then oxidized it into a ~0.42nm aluminum oxide layer.
A high-k hafnium oxide dielectric went on top of that layer.
The aluminum oxide layer both smooths the growth surface and buffers harmful interactions with the MoS2 channel.
Result: ~1nm equivalent oxide thickness and 0.45 mS/μm max transconductance — suggesting future gains depend on atomic-interface control, not just new materials.
Samsung Details Next-gen Memory Chips Aimed at Future AI Infrastructure
August 4, 2026
Samsung unveiled industry-first concept models of zHBM and zNAND-O at FMS 2026, representing its vision for next-generation 3D memory architectures for AI and high-performance computing.
The zHBM design stacks memory vertically directly above AI accelerators to minimize the physical distance data must travel between processor and memory.
Samsung says zHBM could deliver roughly eight times the performance of HBM5, more than 10 times the memory density, and triple the energy efficiency of existing solutions.
The company's new V10 BV-NAND uses wafer-bonding technology to exceed 400 layers, delivering about 58% higher memory density than the prior generation along with faster read, write and I/O performance.
Samsung also presented a broader AI memory roadmap spanning HBM4E samples, HBM5 models, LPDDR5X-PIM and enterprise storage aimed at data centers.
AMD Expands Ryzen AI Lineup to Push On-Device AI Compute to Personal Devices
August 1, 2026
AMD unveiled updates to its Ryzen AI client platforms at Advancing AI 2026, aimed at running AI models directly on personal devices rather than in the cloud.
The Ryzen AI Halo developer platform offers 128GB of unified memory and can run models up to 200 billion parameters; it went up for pre-order in June 2026 with a free year of Hugging Face PRO.
The newly announced Ryzen AI Max PRO 400 Series processors provide up to 192GB of unified memory and support models up to 300 billion parameters, with developer platforms launching later in 2026.
AMD expanded partnerships with Hugging Face (optimized open models/toolkits for Ryzen AI Halo) and Cisco (a new open client AI ecosystem collaboration).
AMD SVP Jack Huynh framed the push as AI moving "from a cloud-only experience to something deeply personal, local and always available."
News archive:
Contact us
tamos.projects@gmail.com
© 2021-2026 Computers.lt