The AI Inference Revolution Is Here
Multiplication is slow and expensive. Addition is fast and cheap. Our Napier chip stores numbers as logarithms, so it never multiplies. It only adds. That's how we squeeze more tokens out of a megawatt than anyone else.
Why this matters for talent
Recogni's cofounder being featured in IEEE Spectrum's look at the inference hardware revolution, discussing a chip architecture that avoids multiplication in favor of cheaper addition, is a technical credibility signal aimed at attracting chip design and machine learning hardware engineers to a company competing against much larger, better funded inference chip makers. Recogni is not currently tracked on the board's company count under that name, so candidates should check the company's own careers page.
11 likes · 1 comments · 0 shares · Read the original LinkedIn post · full article. Curated by Larry Sherwood Jr. from company announcements; analysis above is my own.