Multiverse, Qualcomm Target Data Centers With AI Models Up to 93% Faster
Updated
Updated · Daily Tribune News · Aug 5
Multiverse, Qualcomm Target Data Centers With AI Models Up to 93% Faster
3 articles · Updated · Daily Tribune News · Aug 5
Summary
Multiverse Computing said its optimized AI models will be tailored for Qualcomm Dragonfly AI200 and AI250 accelerators to boost data-center performance while cutting power use.
The pitch is higher capacity on existing hardware: model compression reduces compute and memory needs, letting operators handle more inference requests or run more models without adding accelerators.
At Mobile World Congress in March, a compressed open-source LLM on Qualcomm hardware delivered up to 93% faster responses, 44% higher throughput, 45% lower memory use and 21% lower power consumption with no accuracy loss.
A second on-premises RAG chatbot demo for confidential financial documents ran up to 35% faster with 54% higher throughput, while trimming memory by 45% and power by 14%.
The partnership aligns with a broader industry push to scale AI workloads without matching increases in hardware, energy use and data-center expansion.