AIThe Registerabout 2 hours ago
SambaNova beats GPU-only inference with H200s and RDUs
3 min read
Third-party testing by Artificial Analysis shows SambaNova's heterogeneous platform combining four Nvidia H200 GPUs and 16 SN50 RDUs achieves 763 tokens per second on MiniMax M2.7 at short context lengths. The system disaggregates prefill on GPUs from decode on RDUs. SambaNova closed a $1 billion Series F led by General Atlantic at an $11 billion valuation.
Level
Hype check
Tap to vote and see what everyone thinks.
Summary by ByteBrief
