Tektronix introduced the 7 Series DPO oscilloscopes; the flagship DPO718AX is the first 8-channel instrument with up to 25 GHz bandwidth per channel and 125 GS/s sampling. Inside is a 12-core AMD EPYC Embedded 3351, 96 GB RAM, Nvidia T1000 GPU, and a 15.6-inch touchscreen. Price starts at $178,000 for the 4-channel version and $308,000 for the 8-channel version.
South Korean startup MilliTrack introduced the AMR+ system, which supplements autonomous mobile robot navigation with millimeter radio waves where cameras and LiDAR fail. Claimed accuracy is 3 cm at distances up to 30 m with latency under 100 ms; launch is planned for March 2027.
AMD released the Ryzen 7 7700X3D, an eight-core Zen 4 processor with 96 MB 3D V-Cache for $329. Frequencies are reduced to 4.0/4.5 GHz versus 4.2/5.0 GHz on the 7800X3D, TDP 120 W, AM5 socket, DDR5 and PCIe 5.0 support.
San Francisco startup Rune has introduced RELIC, a modular compute unit that connects directly to operating solar plants without grid connection or construction. The company raised $40 million Series A, bringing total funding to $53.5 million. The first installation operates at a 200 MW solar plant in Texas.
A study on IEEE Xplore analyzed 443,156 HDDs from four brands using Backblaze data from 2013–2025. Toshiba had more failures than Seagate (107% of baseline), while HGST and Western Digital were more reliable — 39–41% and 51–52% respectively.
Tests by YouTube channel Geekerwan showed liquid nitrogen cooling gives Apple A20 Pro only +0.76% single-core and +2.49% multi-core in Geekbench 6. The shift to WMCM packaging and improved vapor chamber already provide high performance.
China has begun mass production of immersion DUV lithography machines through state-owned Shanghai Aishengna Electronic Technology Group. Plans call for 5 machines in 2026 and 20 by 2027, reducing reliance on ASML, which holds 90% of the lithography equipment market.
Startup Perseus Defense tested its Mk-III interceptor, costing about $2,500, at the Yuma Proving Ground in Arizona. The interceptor destroys small drones by kinetic impact rather than signal jamming, during the Falcon Peak 26.2 exercise with NORAD and JIATF-401.
A Geekbench 7 result for Apple's M6 Pro chip has appeared online: 18 cores (6+12), 4.78 GHz frequency, 48 GB memory. Single-core score is 4150, multi-core 37,565, higher than M5 Pro (3692 and 33,595). The entry's authenticity is unconfirmed.
-xl-xl.jpg)
Geekbench 7 results for Apple's M6 chip, manufactured on a 2nm process, have appeared: 22,783 points in multi-core and 4,071 in single-core, versus 17,948 and 3,641 for M5. Apple calls M6 the world's fastest single-core CPU; the chip, along with M5 Pro, will be in an updated Mac Mini on September 22.
ASUS is releasing beta BIOS updates for AM5 motherboards (X870, B850, X670, B650) based on AGESA ComboAM5 PI 1.3.0.1d, optimizing timings and latency for DDR5 modules using CXMT chips. Exact performance figures were not disclosed.
TSMC co-COO Y.J. Mii said AI is useful for coding and chip design but cannot help develop advanced process technologies like A14 (1.4nm) due to lack of data on new variables. He compared AI to a three-year-old Superman and called finding executives for global teams the main challenge to expansion.
A modder ran DLSS 5 Neural Rendering on integrated Intel Arc 140V graphics in a Core Ultra 7 256V processor via a Vulkan layer on Linux, without CUDA or NGX. The technology reconstructs a 71-block U-Net on XMX matrix engines, but performance drops to 3–5 FPS at 720p.
Researchers at Shibaura Institute of Technology developed a touch interface printed with a standard laser printer on transparent PVC. Sliding a finger over the toner pattern generates electrical signals via the triboelectric effect, enabling direction, speed, letter, and user recognition.
Toyota is developing the ELEY robot based on diffusion AI and quasi-direct-drive actuators, which learns factory skills from humans. From 2028, the company plans to release 400,000 units at 60 factories, but engineers acknowledge unresolved issues of reliability, precision, and data infrastructure.
On Newegg and other platforms, the RTX 5090 is selling for $9000–14000, though at launch in 2025 the card cost $1999. Most listings are from third-party sellers, mainly from Hong Kong, which is linked to artificial scarcity and hype around AI purchases.
UK companies heata and thermify install mini-servers in homes that heat water and supply warmth to heating. heata already operates in 100 homes with 3,000 on the waiting list; thermify is testing 480-computer blocks and promises savings up to 34% over 15 years.
Guangzhou robotics company Huaruizhipu unveiled Praxis One, a physical AI platform on the MediaTek Genio Pro 5100 chip. It combines multimodal and VLA workloads with motion control on one SoC, delivering ~106 TOPS and on-device Qwen3-7B inference.
The CAS Institute of Semiconductors and partners built a tunnel junction on sliding ferroelectric Cr/h-BN/3R-MoS2/graphene. At a 0.5 V read voltage, it achieved a tunnel electroresistance ratio of 1.9×10^7 and endurance over 10^11 switches. Results published in Science.
Kioxia unveiled the XL1, a 512 GB CXL 2.0 Type 3 device using XL-FLASH Gen 2 NAND in E3.S form factor. It delivers up to 2 million IOPS at 755 ns latency and lets servers scale memory without DRAM prices.
Chinese leaker ChannelGate claims TSMC notified clients of roughly 10% chip price increases in Q4 2026, potentially raising AMD Radeon GPU prices again. TSMC officially cited up to 10% increases only from early 2027; AMD and TSMC have not confirmed.
Meta plans to deploy its custom AI processor MTIA 450 (Arke) in data centers in the first half of 2027 to reduce inference costs. On September 1, TSMC delivered 12 chips to Meta; tests showed 2–3% deviation from simulations. The next generation MTIA 500 (Astrid) is expected by the end of 2027.
Honor announced the compact Tiangong AXB35 AI workstation at about 2 L (70 × 192 × 203 mm) in three versions: Standard, Pro, and Ultra. The Pro is available with Intel Core Ultra X7 358H and Arc B390 or with a 16-core AMD Ryzen AI Max+ 395, Radeon 8060S, and up to 128 GB LPDDR5X on a 256 GB/s bus; the Ultra is built on the Arm-based Nvidia RTX Spark platform.
MediaTek has unveiled its flagship Dimensity 9600 Pro chip, focused on agentic AI. It supports on-device execution of mixture-of-experts models up to 30 billion parameters. Vivo will partner in the AI smartphone race.
Shin-Pu Jen, president of IMAPS Taiwan, said that in 2026 advanced semiconductor packaging is shifting from individual technologies to system-level co-design for AI tasks. In Taiwan, co-packaged optics, chiplets, and glass substrates are increasingly used.
Growing memory demands boost interest in Compute Express Link (CXL) as a way to add capacity beyond locally attached memory. HBM remains the leader in bandwidth, while CXL wins on capacity.
Intel Foundry has moved ASML's High-NA EUV lithography into mass production. However, the challenge of stitching large chips from multiple exposures remains unresolved.
AMD published first official benchmarks for EPYC Venice server CPUs. According to the company, the 256-core flagship EPYC 9996 is 2.24 times faster than Nvidia Vera in SPEC CPU 2026 Integer Rate, and the 96-core version is about 20% faster per-core.
Geekbench 7 published the first graphics test results for Apple's M5 Ultra and M6 ahead of new Mac Studio and Mac mini launches on Tuesday. M5 Ultra scored up to 366,744 Metal points (+59% vs M3 Ultra), while M6 scored up to 93,217 (+34% vs M5).
A wave of leaks points to an imminent Intel Nova Lake announcement: ASRock Z990 motherboards appeared in a supply database, and an HP EliteBook X with an unknown Intel chip appeared in SiSoftware. Intel CEO Lip-Bu Tan said the launch is planned for the second half of 2026.
