Have you tested it on much (x100, x1000 times) more complicated scenes? Ten objects is very simple task for GPU and is definately not a benchmark for it. Try to lower overall rate to 200 fps or less and see.
Thanks for reply.
I tested it on 500 the same models,
and the results were not change.
DrawIndexed x500 times got 440 fps, and DrawInstanced got 134 fps.
Sorry, I forgot to mention that: the model used in the DrawInstanced method is extracted from a texture in geometry shader(GS). Is this the bottleneck? How can I get the time cost of Texture.SampleLevel()?