Latest
A technique using "Lossless Scaling" to split rendering and frame generation across separate GPUs is gaining attention. Unlike legacy technologies such as SLI, this method allows GPUs from different generations to be combined to distribute rendering load, offering a promising new approach to affordable performance gains by repurposing older hardware.
Etched announced a $300M raise that included participation from SK hynix, and has begun initial production of inference clusters combining low-voltage compute with shared memory. Supply agreements and comparable performance figures have not yet been disclosed.
Claude Opus 5 matches Opus 4.8's API base pricing while approaching Fable 5's practical performance. Migration requires attention to the now-default thinking feature and unsupported APIs.
ChatGPT Voice has come to the desktop versions of Work and Codex, making it possible to start or interrupt tasks and coordinate multiple agents by voice. Permission settings and the separate accounting of voice time versus task execution will shape adoption decisions.
AMD layers Skills, a CLI, and Hyperloom's autonomous optimization on top of the ROCm Core SDK, connecting everything from environment setup to performance tuning through agents. The current implementation is a Tech Preview, with constraints remaining on supported GPUs and ROCm versions.
Reports suggest Intel could handle wafer manufacturing and advanced packaging for NVIDIA's next-generation AI platform Feynman. We examine what's confirmed in official materials versus what remains an unsettled contract scope.
Alphabet posted its first-ever free cash flow deficit ($5.855 billion) since going public, in the Q2 earnings reported on July 22, 2026. This piece breaks down how 69% of its $112.1 billion net income came from the after-tax impact of $99 billion in unrealized gains on stakes in SpaceX and an unnamed private company, and how $44.9 billion in capital expenditures outpaced the company's actual earning power.
Geekbench 7, released by Primate Labs, switches its baseline machine to an AMD system and introduces modern workloads that mimic actual gaming and AI processing. Because only some tests count toward the multi-core score, careful examination of each score is essential to gauge a chip's true performance.
Intel has committed fully to launching mass production of 14A in 2028. The 2028 timing itself isn't new—what's changed is the management decision to invest in PDK development and manufacturing capacity. External customer adoption and yield verification in 2027 are the next gates.
By harnessing a frustrated state in which electrons in a quantum material compete against one another, researchers have succeeded in controlling the electron arrangement—switching it from a stripe pattern to a checkerboard pattern—using only a minute magnetic field. This achievement points toward a path for developing next-generation computing and spintronic devices that operate with extremely low power consumption.
