Lamb Labs unveils AI chip plan targeting 20,000 tokens a second
Lamb Labs says an 8B FPGA prototype runs below 10 watts, while its 63x efficiency claim lacks a reproducible benchmark.
By Ryan Merket · Published
Primary source: X
Why it matters
Sub-watt inference could move capable language models into robots and consumer devices, but Lamb Labs must translate FPGA experiments and internal metrics into tested silicon.
Lamb Labs, founded by Niki Kotecha and Thomas Lanning, said in a thread on X on August 4th that its planned custom inference chips could generate more than 20,000 tokens per second and deliver 63 times the "Intelligence per Watt" of traditional GPUs.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.