Latest
London-based Spectral Compute has published its own benchmarks showing its CUDA-compatible compiler SCALE running 5.94x faster than HIP on AMD GPUs, raised $6 million in November 2025, and joined the NVIDIA Inception Program in June 2026. This piece examines where the evidence stands, contrasted against the ZLUDA precedent.
A research team including KAIST automatically classified over 120,000 MoS₂ flakes and evaluated 1,615 FETs. By statistically capturing differences in current flow and switching performance across layer counts, they built a data foundation for 2D semiconductor design.
Meta has removed, across all accounts, the reference feature of its AI tool "Muse Image," which could generate images of Instagram users without their consent, just three days after launch amid a concentration of criticism from multiple directions. This is an examination of what marks the third failure of consent-related design at Meta since 2025.
Bloomberg reports Apple will skip M6 Pro and M6 Max and move up M7 to boost on-device AI performance, with a major memory bandwidth increase planned. The shift could make upgrade timing for higher-end Macs less predictable.
SEALSQ and GlobalFoundries will jointly develop reusable post-quantum cryptography IP and CryoCMOS for quantum control. The US government's compliance deadline of end-2030 is pushing forward the race for certification and mass production after standardization.
Claude Code Desktop's browser pane now supports interaction with external websites. While it bridges development and research within a single screen, permissions are separated through clean profiles and per-site approvals.
Nanya Technology has laid out a preliminary plan to invest over NT$200 billion in 2027. The company aims to channel cash generated from surging prices into its new fab, ramping up to a monthly capacity of 30,000 wafers by 2028.
We examine the fast-charging degradation mechanism reported by Live Science against Geotab's data from 22,700 EVs and a test of 40 smartphones. EVs show a 12-point gap over an 8-year projection, while smartphones show just 0.5 points over 500 cycles—an order-of-magnitude difference in degradation scale that this article explains.
Muse Spark 1.1 tied GLM-5.2 at an overall score of 51 in an independent evaluation, taking the lead in coding scores and estimated task cost. The difference between closed APIs and open weights will shape adoption decisions.
The helium supply risk that Intel's CEO flagged in June 2026 is not directly connected to China's export ban three weeks later. What links the two is the attack on Qatar in March and the resulting production halt. This article traces the timeline of the global supply network to unpack why China—a country with just a 1.6% production share that depends on imports—resorted to such an unusual measure.
