Adapteva announces 1,000 Teraflops accelerator chip for deep learning

april

[EDIT: April’s Fool’s Joke. This was a reaction to the heap of opportunistic DL hardware popping up these days. Should have been patently obvious from text below but just in case…]

We are pleased to announce our latest project at Adapteva simply called “The Brain”. This new disruptive technology will be designed in 3nm and has the following projected performance metrics:

  • 500,000 single precision floating point cores
  • A fixed programmable architecture for matrix-matrix multiplication
  • Zero KB of data and program memory
  • Total performance of 1,000 Teraflops
  • Total power consumption of 1,000W
  • Patent pending supersonic water cooling technology

Analyst Charles Smith at the Silicon Analysts has stated: “This is a smart move by Adapteva. Instead of wasting silicon space on memory and programmability, they are focusing on what matters, raw floating point performance. It’s a well known fact that all interesting problems can be reduced to dense matrix-matrix multiplication.”

This announcement marks a new chapter for Adapteva. We look forward to hearing more about your plans for our disruptive new chip.

Sincerely,

Andreas Olofsson

Chief Entertainment Officer, Adapteva, Inc.

[ April fools… couldn’t resist. Hopefully you read all the way to the end. And yeah, I am still working on a new chip, in case you were wondering]


 

Andreas Olofsson is the founder of Adapteva and the creator of the Epiphany architecture and Parallella open source computing project. Follow Andreas on Twitter.