Search
South Korea’s FuriosaAI Releases Third-Gen Inference Accelerator: 32-Fold Surge in Compute Power and Memory Bandwidth
2 day ago /
Read about 0 minute
Author：小编
South Korean AI semiconductor pioneer FuriosaAI, in partnership with Broadcom, has revealed the technical specifications of its third-generation AI inference accelerator. The new accelerator integrates two 2nm compute dies, a dedicated I/O die, and 12 stacks of 48GB HBM4(E) memory. It leverages a fully connected Switchless Unidirectional Ring (SUE) topology and supports PCIe Gen7 connectivity. Compared to its predecessor, the physical footprint of this accelerator has expanded nearly eightfold, delivering a 32-fold increase in computing power and HBM memory bandwidth, along with a 12-fold growth in HBM memory capacity.
-
C114 Communication Network -
Communication Home
7 X 24 Track global technological trends
