In the latest round of MLPerf Training results, Graphcore’s Bow offers a modest improvement and Habana’s Gaudi2 triples performance over its predecessor, vaulting past Nvidia’s A100 on one benchmark.
New Intel Atom processors have 4 to 24 CPUs and target enterprise networking and SMB storage. They extend the Snow Ridge product line and largely replace the Denverton family.
AlphaICs is sampling a test version of its Gluon edge-inference accelerator chip. Given moderate performance, the company plans changes to improve the production version, with an architectural overhaul planned for the second generation.
MediaTek and Qualcomm announced midlife kickers for their premium-smartphone processors, delivering the Dimensity 9000+ and Snapdragon 8+ Gen 1, respectively, and showing that not all 4nm processes are the same.
Ray tracing in smartphones takes a big step with Arm’s new Immortalis-G715 GPU, while the Mali-G715 version brings new features and better performance to mainstream devices.
The startup’s technology can create a hundred optical data carriers from a single laser, greatly reducing the cost and power of high-bandwidth optical links that use wavelength-division multiplexing.
Arm’s new Makalu CPUs, branded Cortex-X3 and Cortex-A710, deliver performance and power-efficiency gains, respectively. The IP vendor also updated Cortex-A510 and the DSU-110.
French startup SiPearl has received additional funding to complete the development of its Rhea supercomputing processor, which has changed considerable from its original specifications.
NXP’s S32Z2 and S32E2 processors integrate multiple low-level automotive functions into a single SoC that preserves isolation between virtualized electronic control units (ECUs) from software through I/Os.
MIPS Inc. has developed its first licensable CPUs implementing the RISC-V instruction set by repurposing older MIPS-compatible cores. The P8700 and I8500 outperform most other RISC-V designs.
Silicon Labs has added a small deep-learning accelerator to its newest wireless microcontrollers, reducing the power required for inference of tiny AI models.
Server processors, AI accelerators, and switch chips are consuming more power and generating more heat, some as much as 950W. Data centers must change to accommodate these hot chips.
Broadcom and Marvell now offer automotive Ethernet switch chips capable of 55Gbps and 37Gbps. Usable in both zonal and domain architectures, they support the move to automotive Ethernet and its rising bandwidths.
The startup is close to taping out its “universal” processor, branded Prodigy, which aims to deliver industry-leading performance on server, supercomputer, and AI workloads.
The CPU performance of Qualcomm’s mid-premium smartphone processor is similar to last year’s, but its capabilities in total make for a substantial upgrade.
