search

u39kun / deep-learning-benchmark

Deep Learning Benchmark for comparing the performance of DL frameworks, GPUs, and single vs half precision
429 stars 66 forks source link

How is possible to work with FP16 in GTX 1080 Ti? #10

Closed diegodgs closed 6 years ago

u39kun commented 6 years ago

You can run FP16 calculations on 1080 Ti, and you get a marginal speed up over FP32. Of course, 1080 Ti does not have Tensor Cores, so you don't get the level of speed up that you do with Titan V / V100. The new 2080 Ti have Tensor Cores. I will be posting numbers when I get my hands on one (should be very soon).