Company pushes 800 VDC to feed power-hungry GPUs and cut copper

Nvidia has admitted that alternating current is no longer up to the job in AI data centres where inferencing surges and bandwidth spikes are battering power infrastructure.

Snapdragon slinger gets Humain and Saudi cash to back its first data centre silicon

Qualcomm saw its shares spike by 20 per cent after it revealed its first AI-focused data centre chips in a desperate attempt to chip away at Nvidia’s vice grip on the market.

Repair specialist says Nvidia's flagship card is a modder’s nightmare

Nvidia's RTX 5090 Founders Edition has been labelled a disaster by repair experts after a failed attempt to add water cooling rendered the GPU completely unfixable.

Outfit still on a tightrope

Shares in Troubled Chipzilla surged after the struggling semiconductor outfit posted better than expected revenue, signalling tentative progress in its long and painful turnaround.

Start-up inks deal worth tens of billions to secure computing power

Anthropic has struck a deal to grab hold of one million Google Cloud chips to train and run its artificial intelligence models, tightening its bond with one of its biggest investors.

Starcloud plans orbiting GPU boxes claiming 100x compute and free solar juice

Nvidia's AI silicon is headed for orbit as startup Starcloud prepares to launch a data centre in space using H100 GPUs stuffed into a 60-kilogram Starcloud-1 satellite.

SOCAMM2 modules promise massive capacity, tiny footprint, and less power

Micron has rolled out what it claims is the industry’s beefiest low-power DRAM module, a 192GB SOCAMM2 designed for the power-hungry AI data centre crowd.

Scheduler squeezes more work from fewer H20 accelerators

Alibaba Cloud boffins have emerged from their smoke filled labs having found a way to make Nvidia’s costly GPUs actually earn their keep.

With 72GB of GDDR7 ECC memory

NVIDIA has silently introduced a new version of the RTX PRO 500 Blackwell GPU that comes with 72GB of GDDR7 memory.

Jensen Huang made the $100 billion deal a top priority

Nvidia boss Jensen Huang apparently panicked when rumours spread that OpenAI might start using Google’s custom silicon for its massive AI workloads.