Bad news for NVIDIA? Google TPU

[5150Joker]

Looks like NVIDIAs dominance in machine learning may come to an abrupt end if this new ASIC dubbed a TPU by Google really delivers on what they claim: http://www.pcworld.com/article/3187...ips-smoke-cpus-gpus-in-performance-tests.html

So how would NVIDIA react to something like this? Do they go the route of creating a dedicated ASIC themselves or let a start up and/or Google steal their thunder?

 

[razor1]

they are testing against a K80, maxwell was like what 10 times faster than a keplar and pascal is 10 times more than that

 

[ChosenUno]

Looks like NVIDIAs dominance in machine learning may come to an abrupt end oi this new ASIC dubbed a TPU by Google really delivers on what they claim: http://www.pcworld.com/article/3187...ips-smoke-cpus-gpus-in-performance-tests.html

So how would NVIDIA react to something like this? Do they go the route of creating a dedicated ASIC themselves or let a start up and/or Google steal their thunder?

That Google chip serves a completely different market segment. Nvidia Tesla GPUs are mainly used for training neural nets, while their Tegra units are used for inference.

There is no need for Nvidia to create the ASIC, they already have the ASIC in Tegra and their Drive PX ecosystem.

Read more over at https://blogs.nvidia.com/blog/2016/08/22/difference-deep-learning-training-inference-ai/