Search
Exclusive: DeepSeek R2 Unveils Major Performance Leap Thanks to Autonomous AI Infrastructure
2025-04-27 /
Read about 0 minute
Author：小编
Renowned AI expert @iruletheworldmo has disclosed that following the triumphant debut of the R1 model, DeepSeek is poised to introduce the R2 model, marking a substantial performance upgrade. Leveraging Huawei's Ascend 910B chip cluster, the forthcoming Atlas 900, and DeepSeek's proprietary distributed training framework, this latest iteration boasts an impressive accelerator utilization rate of 82%, delivering a staggering 512 PetaFLOPS of FP16 performance—approximately half an exaFLOP of computing power.
-
C114 Communication Network -
Communication Home
7 X 24 Track global technological trends
