Skip to content

TVM Benchmark

Jia Guo edited this page Dec 13, 2018 · 10 revisions
MobileFacenet@TVM MobileFacenet@NCNN
Macbook r2015 2.2G@4 Threads 6.4ms No SIMD optimization on x86
Firefly Rk 3399@1xA72 69.24ms 72ms
Firefly Rk 3399@Mali GPU[fp32] 70.32ms [x]
Firefly Rk 3399@Mali GPU[fp16] 45.20ms [x]
Raspberry Pi 3b@1xA53 308.24ms 163.41ms
Tips: Now, MXNet PReLu layer will cause compile error when use TVM on arm devices.
Clone this wiki locally