The Korean unicorn packages four NPU dies with 144 GB of HBM3E and bets on an open software stack to serve large language models at lower cost per token.
The Korean unicorn packages four NPU dies with 144 GB of HBM3E and bets on an open software stack to serve large language models at lower cost per token.
The 120 MHz GD32E512 is optimized for high-speed optical modules and the 72 MHz GD32E252 takes on low-speed optical modules.
The 120 MHz GD32E512 is optimized for high-speed optical modules and the 72 MHz GD32E252 takes on low-speed optical modules.
The new XpressConnect PCIe 6.0 and CXL 3.1 Retimers reign in signal integrity challenges in AI data centers and other…
The new XpressConnect PCIe 6.0 and CXL 3.1 Retimers reign in signal integrity challenges in AI data centers and other high-bandwidth installations.
At Computex, Intel described the CPU as the control plane for agentic workloads, pairing new processors and network…
At Computex, Intel described the CPU as the control plane for agentic workloads, pairing new processors and network controllers with fresh details on its inference GPU.
The 88-core Vera processor introduces Spatial Multithreading and a claimed 1.8x task-completion lead over x86 processors.
The 88-core Vera processor introduces Spatial Multithreading and a claimed 1.8x task-completion lead over x86 processors.
The bottleneck limiting the next generation of artificial intelligence isn't compute power—it's the wire connecting the…
The bottleneck limiting the next generation of artificial intelligence isn't compute power—it's the wire connecting the chips. French startup Scintil Photonics thinks it has the answer.
Broadcom’s high-integration Wi-Fi 8 & NPU-accelerated 50G PON gateway SoCs aim to build a cohesive, 50 Gbps broadband…
Broadcom’s high-integration Wi-Fi 8 & NPU-accelerated 50G PON gateway SoCs aim to build a cohesive, 50 Gbps broadband access ecosystem for the AI-infused home.
Learn how innovative technologies like protection switches will be crucial in transitioning to high-power delivery of USB-C ports.
Learn how innovative technologies like protection switches will be crucial in transitioning to high-power delivery of USB-C ports.
The new version of FPGA AI Suite is designed to accelerate trained AI models into FPGAs
The new version of FPGA AI Suite is designed to accelerate trained AI models into FPGAs
Learn how liquid cooling eliminates system airflow, creating a hidden thermal bottleneck for 'left-behind' components…
Learn how liquid cooling eliminates system airflow, creating a hidden thermal bottleneck for 'left-behind' components like memory and SSDs. Targeted micro-cooling is required to restore system balance.
Learn how Korean AI technology company Rebellions uses proprietary dataflow NPUs, chiplets, and HBM to deliver…
Learn how Korean AI technology company Rebellions uses proprietary dataflow NPUs, chiplets, and HBM to deliver high-efficiency, scalable AI inference for modern data centers.
Learn how The Netherlands is leveraging billion-euro investments and European partnerships to scale integrated photonics…
Learn how The Netherlands is leveraging billion-euro investments and European partnerships to scale integrated photonics from research into a world-leading, industrially mature production ecosystem.
GaN-based power delivery board targets 96.5% efficiency and 2,100 W/in3 for NVIDIA's next-generation 800 VDC data center…
GaN-based power delivery board targets 96.5% efficiency and 2,100 W/in3 for NVIDIA's next-generation 800 VDC data center architecture.
In this exclusive interview, the Seattle startup says its software-ML-hardware co-design—announced today—converts…
In this exclusive interview, the Seattle startup says its software-ML-hardware co-design—announced today—converts off-the-shelf FPGA servers into high-efficiency AI inference engines.
Learn how monolithic ceramic cooling using Selective Laser Reaction Sintering (SLRS) eliminates the "thermal wall" in…
Learn how monolithic ceramic cooling using Selective Laser Reaction Sintering (SLRS) eliminates the "thermal wall" in high-density power electronics by creating single-piece, reliable heat exchangers.
Learn how GaN technology enables a low-cost, low-profile 6 kW, 800 VDC to 12.5 VDC converter using an ISOP LLC topology.…
Learn how GaN technology enables a low-cost, low-profile 6 kW, 800 VDC to 12.5 VDC converter using an ISOP LLC topology. This meets the design needs of next-gen, MW-scale AI server infrastructure.
Optical computing is the "New Moore's Law" for AI. It solves electronic scaling limits, offering higher speed, lower…
Optical computing is the "New Moore's Law" for AI. It solves electronic scaling limits, offering higher speed, lower power, and efficiency gains proportional to problem size for matrix operations.
Learn why the best silicon is useless without the right AI developer tools.
Learn why the best silicon is useless without the right AI developer tools.
In the final part of this series, we zero in on three photonics technologies expected to be essential for demanding…
In the final part of this series, we zero in on three photonics technologies expected to be essential for demanding computing and communication systems of the future.
Announced today, Broadcom’s Gen 8 128G Fibre Channel switch portfolio is claimed as the first quantum-safe, SAN…
Announced today, Broadcom’s Gen 8 128G Fibre Channel switch portfolio is claimed as the first quantum-safe, SAN architecture with 128G speed, AI-powered SAN management, and quantum-safe cryptography.
Facebook
Google
GitHub
Linkedin