Mobile and IoT spot elephantine Qualcomm on Wednesday announced an enlargement of its artificial quality developer runtime offering, called Qualcomm AI Stack portfolio, which the institution claims volition code the "Unique needs of AI for each and each concern line" that Qualcomm's chips serve, including mobile phones, connected infrastructure successful Internet of Things installations, and connected vehicles.
The bundle incorporates existing tools, specified arsenic the company's Neural Processing SDK, portion allowing developers to "leverage that aforesaid enactment crossed each product."
"We are extending 1 exertion roadmap crossed each our antithetic concern lines," said Qualcomm vice president of merchandise absorption Ziad Asghar successful a media briefing. "The measurement relation present is to beryllium capable to instrumentality your enactment and larboard it crossed those antithetic concern lines."
The bundle stack has a absorption connected powerfulness efficiency, specified arsenic however to quantize workloads, and portability of codification crossed the galore hardware targets of Qualcomm and its customers.
On the ratio side, "the show per watt that we are showing is conscionable outstanding successful presumption of however overmuch enactment we tin bash for a fixed magnitude of power," said Asghar. "Even for the aforesaid hardware, arsenic we optimize the software, we are capable to get 30% to 40% amended show successful immoderate cases."
Asghar said the quality to grip galore contexts is the main virtue.
"Some of the environments are precise focused connected their extremity markets — connected car oregon cloud," noted Asghar.
"This is the champion of people of acquisition for what you could get retired of our hardware, that's the unsocial advantage," helium said.
"This volition let you spell get the astir optimized and champion acquisition from an ratio position and an accuracy perspective, so, we deliberation this is simply a immense differentiation for america and our partners."
A extremity of the exertion is for customers to beryllium capable to set show of instrumentality learning for antithetic criteria, said Asghar.
"For this application, latency is precise critical, you person to person the quality to optimize for that, whereas for that application, latency is astir important, oregon accuracy — you tin marque those commercialized offs with the stack we are offering, and past leverage that enactment everywhere."