Gyrfalcon starts shipping AI chip
Startup Gyrfalcon introduced the Laceli AI Compute Stick, based on the same ASIC, early in 2018 to compete with the Neural Compute Stick introduced by Intel in 2017 (see Gyrfalcon takes on Intel over AI power consumption).
The Lightspeeur 2801S is the first instantiation of Gyrfalcon’s APiM processing-in-memory architecture. The use of memory as a processing element can be used to minimize the data movement that is a large part of the power consumption of conventional processors. The chip has roughly 28,000 parallel computing cores and does not require external memory for AI inference.
The Lightspeeur 2801S achieves 2.8 tera operations per second (TOPS) while consuming 300mW yielding 9.3TOPS/W in a package measuring 7mm by 7mm by 2.8mm. For heavier loads the ASIC be set in arrays from 2 to 32 chips per board.
“Balancing the cost-performance-energy equation has been a challenge for developers looking to bring AI-enabled equipment to market at scale,” said Lin Yang, chief scientist at Gyrfalcon, in a statement. “By deploying APiM and MPE on a standard, commoditized ASIC, GTI is enabling our customers to bring innovative, AI-enabled devices to the masses.”
Gyrfalcon was founded in 2017 by a group of entrepreneurs and engineers that came from such companies as AMD, Spansion, TDK, Legend Silicon, Omnivision, Cadence, C-Cube and C2 Microsystems.
Related links and articles: