Skip to content

Artificial Intelligence Hardware - Category Home

Artificial Intelligence Hardware - Topics

Why do AI accelerator cards use 12V-2x6 power instead of EPS connectors? Why do dual-socket AI servers need so many PCIe lanes? How much of an AI cluster's cost is chassis and power delivery? Intel Loihi vs IBM TrueNorth vs SpiNNaker: what does each chip really compute? Does liquid cooling change how many cards an AI rack can hold? Can a local AI workstation be useful without a data center card? Why do accelerator cards hold resale value better than gaming GPUs? How much airflow does a passively cooled data center accelerator need? Why does backpropagation not map onto neuromorphic hardware? Spiking neural networks cut power use but training them is still hard What should you inspect before buying a used AI accelerator? Does ECC on accelerator memory matter for long training runs? Why does the software toolchain hold neuromorphic hardware back more than the silicon? Always on keyword spotting at microwatts on a neuromorphic chip What breaks first when an AI server runs flat out for weeks? Can photonics replace the copper interconnect inside a GPU cluster? Memristor crossbar arrays doing multiply accumulate where the weights live How many neurons and synapses actually fit on one neuromorphic die? Why do custom AI accelerator startups struggle to beat a general-purpose GPU? Brain inspired is a marketing word until you check the learning rule How do you split one physical AI card between several users? Optical computing: which problem does light actually solve better than transistors? What really caps how many accelerator cards fit in one chassis? How many watts does the self-driving computer draw and where does that heat go OAM modules instead of PCIe cards: what does the board gain? How do you power a four-card workstation on a household circuit?