This is news to me. Will it likely affect Nvidia in any significant way?
From p.18 of the report:
"What is new in the hardware driver is that Chinese tech giants and unicorn startups are competitive with some
of the world’s leading companies in designing AI chips. For instance, Chinese company Cambricon, a statebacked
startup valued at $1 billion, has developed chips that are six times faster than the standard GPUs for deep
learning applications and use a fraction of the power consumption.65 Moreover, equipped with a new “neural
processing unit,” Huawei has arguably overtaken Apple in mobile AI chips.66"
Nvidia will face more competition in the future, but it has technological edge.
So far Nvidia has build microarchitectures that are generic. They just put tensor cores, F64, F32, F16 units and 8-bit inference support into the same GPU. They are used in HPC, graphics, DL inference and training. That's four different domains.
I suspect that in the future Nvidia builds 2-3 different microarchitectures.
Right, I wonder just how specialized these chips are. There are still changes in modeling and architecture coming hot off the research presses. It's easy-ish to hard code a specific architecture in your Verilog and claim massive efficiency improvements. While AI is still very much in the research stage, general-purpose is what really matters.
From p.18 of the report:
"What is new in the hardware driver is that Chinese tech giants and unicorn startups are competitive with some of the world’s leading companies in designing AI chips. For instance, Chinese company Cambricon, a statebacked startup valued at $1 billion, has developed chips that are six times faster than the standard GPUs for deep learning applications and use a fraction of the power consumption.65 Moreover, equipped with a new “neural processing unit,” Huawei has arguably overtaken Apple in mobile AI chips.66"