GTC 2020: Baidu's CTR in HGX-2
After clicking “Watch Now” you will be prompted to login or join.
Click “Watch Now” to login or join the NVIDIA Developer Program.
Baidu's CTR in HGX-2
Deping Xie, Baidu | Ruiquan Ding, Baidu
Share Baidu's experiences in migrating CTR model training from CPU nodes to HGX-2, which can either reduce the hardware cost by an order of magnitude at the same training speed, or accelerate the training speed by several times for the same hardware cost. All the work depends on the cutting-edge hardware, the highly integrated AI supercomputer, and the hardware/software tightly co-work. If you're using deep learning to improve the ads or recommendation system, or you're overcoming the challenge of high-dimensional sparse data for deep learning, welcome to our session!