SambaNova Achieves Record Speed with DeepSeek-R1 AI Model Deployment

SambaNova Systems announced the world's fastest deployment of the DeepSeek-R1 671B large language model. The company achieved 198 tokens per second per user using only 16 custom-built chips, replacing the 40 racks of 320 Nvidia GPUs typically required. According to SambaNova, their SN40L RDU chip makes their platform the fastest for running DeepSeek. They anticipate increasing the speed to five times faster than the latest GPU speed on a single rack and offering 100 times the capacity for DeepSeek-R1 by year-end. SambaNova's reconfigurable dataflow architecture offers a more efficient solution, delivering three times the speed and five times the efficiency of leading GPUs. DeepSeek-R1 is now available on SambaNova Cloud, with API access offered to select users.