- AI-Energy requirements could be lowered with large single-sink chips
- Researchers say these can overcome the limitations that GPUs face
- Cerebras and Tesla are already using these huge chips with special cooling systems to control heat
Engineers at the University of California Riverside are investigating a new approach to artificial intelligence hardware that could tackle both performance and sustainability.
In a peer-reviewed paper published in the journal Unitexamined the team the potential of wafer -scale -accelerators -giant computer chips that work on whole silicon slices rather than the little chips used in today’s GPUs.
“Wafer-scale technology represents a big leap forward,” said Mihri Ozkan, professor of electric and computer technology at UCR and lead author of the paper. “It allows AI models with trillion of parameters to run faster and more efficiently than traditional systems.”
Like monorails
These chips, cerebras’ wafer-scale engine 3 (WSE-3), which we have previously covered, contain up to 4 trillion transistors and 900,000 AI-focused cores on a single device. Another wafer-scale processor, Teslas Dojo D1, houses 1.25 trillion transistors and close to 9,000 cores per day. Module.
Processors remove delays and energy losses that are common in systems where data raises between multiple chips.
“By keeping everything on a slice, you avoid delays and power losses from chip-to-chip communication,” Ozkan said.
Traditional GPUs are still important due to their lower cost and modularity, but when the AI models grow in size and complexity, chips begin to encounter performance and energy carriers.
“AI Computing is not just about speed anymore,” Ozkan explained. “It’s about designing systems that can move huge amounts of data without overheating or ingestion of excessive electricity.”
WAFER-scale systems also have important environmental benefits. For example, Cerebras’ WSE-3 can perform up to 125 quadrillion operations per day. Second while using far less energy than GPU setups.
“Think of GPUs like busy highways – effective but traffic jams waste energy,” Ozkan said. “Wafer-scale engines look more like monorails: direct, effective and less polluting.”
However, a major challenge is still – the ancient question of heat. Wafer-scale chips can consume up to 10,000 watts of power, which almost all turns into heat, requiring advanced cooling systems to prevent overheating and maintain performance.
Cerebras uses a glycol -based refrigerator built into the chip, while Tesla has a fluid system that spreads coolant evenly over the surface of the chip.
Via Tech Xplore
