Why FPS when running inference on CPU is 4 times faster than when running on GPU?

[brm738]

<style> </style>

Training Time (s)	Testing Time (s)	Inference Throughput (fps)	device	model_name	dataset.category
52.4286046	12.37639093	0.950309898	gpu	patchcore	bottle
61.79791188	14.6184566	1.391339307	gpu	patchcore	cable
Training Time (s)	Testing Time (s)	Inference Throughput (fps)	device	model_name	dataset.category
1748.111946	27.97528219	4.048704866	cpu	patchcore	bottle
1710.906662	47.00997496	4.217795942	cpu	patchcore	cable

Is it because parallelism isn't needed for inference?