Intel and Nvidia Fighting for Deep Learning Dominance

Related previous discussion: https://news.ycombinator.com/item?id=12300868

The real fight probably won't start until something like Knight's Mill (http://www.anandtech.com/show/10575/intel-announces-knights-...) is launched. Even then, it seems that unless Intel provides a drop-in cuDNN-like library for deep learning the big frameworks will have a tough time supporting Intel hardware quickly/efficiently.

Couple more data points:

Intel recently bought Nervana Systems, a startup focused on creating a neuromorphic ASIC for deep learning, which had been financed by DFJ. Nervana wrote a fast open-source library called neon for GPUs. I expect the Nervana team will be involved in this fight.

Right now, Intel largely owns the data centers while NVIDIA owns academic hardware acceleration. It's faster. Obviously those are generalizations. NVIDIA's DGX-1 is a way to leap into enterprise, while the Xeon Phi is Intel trying to compete on speed and distributed neural net training.

Interesting times.

A more apt headline would be "Intel trying to dislodge Nvidia from its current dominance in deep learning."

It would be great for Nvidia to get real competition.

Meanwhile, google already has an ASIC.

Why isn't AMD able to compete here, given their GPU history (acquisition of ATI)? It seems like they would be a natural fit for the market.