googlenet_v1 result discrepancy

[Thesane]

I used the python Xdnn compiler to compile googlenet_v1 caffemodel to generated fpga version but when I benchmark the the generated one I only get 2793 FPS max throughput vs. the precompiled googlenet_v1 (under examples/deployment_modes/data) which is 4230 FPS for throughput mode. I wonder what was the parameters used for this version or if it was based on a different model?