“SambaNova's proprietary SM40 and SM50 technology is claimed to be 5 to 10 times faster on the decode phase of inference when compared to an NVIDIA GPU.”