When most folk consider Qualcomm, the primary applied sciences that seemingly come to thoughts are the corporate’s industry-leading cellular platform system-on-chips for smartphones, in addition to the corporate’s end-to-end 5G connectivity options. Nevertheless, whether or not you think about purposes like picture recognition, speech enter, pure language translation or suggestion engines, trendy smartphone platforms usually require quite a lot of synthetic intelligence (AI) processing horsepower as effectively. As such, after years of growing silicon and software program platform options for cellular AI purposes, it stands to motive that Qualcomm has a chance to convey its AI accelerator know-how to different clever edge units and the cloud. And that’s simply what’s occurring with Qualcomm’s Cloud AI 100 inference accelerator portfolio, as evidenced by the corporate’s latest joint announcement with Foxconn, one of many largest electronics contract producers and ODMs on the earth.
Business Deployment Of Qualcomm Cloud AI 100
Foxconn’s Industrial Web division has launched a brand new AI-enabled machine imaginative and prescient platform known as Gloria. Powered by Qualcomm’s Cloud AI 100 answer, the system is design to assist as much as 24 high-definition cameras for video analytics purposes like visitors evaluation, safety monitoring and good retail areas. Additional, Vietnamese-based AI safety powerhouse BKAV has introduced the primary buyer deployment of the Qualcomm Cloud AI 100-powered Foxconn Gloria in its AI View good safety digital camera system portfolio.
Qualcomm has been working with BKAV since late final 12 months, and to get sense for what the corporate’s digital camera methods can do, right here’s a have a look at a previous-gen product based mostly on Qualcomm’s legacy Imaginative and prescient Intelligence Platform…
As you’ll be able to see, picture recognition and video analytics are highly effective instruments for a myriad of purposes, from good cities to safety, to retail and healthcare purposes and extra.
Below The Hood Of Qualcomm AI 100, Efficiency Claims And Foxconn Gloria
Calling upon related silicon engine know-how taken from Qualcomm’s Snapdragon Cell Platform Hexagon and Tensor Accelerator architectures, AI 100 itself employs as much as 16 AI Cores (AICs) and scales all the best way from 15 Watts on as much as 75 Watt energy envelopes. Qualcomm additionally notes that AIC chiplets on board AI 100 accelerators are linked by way of a proprietary high-speed interconnect. Bodily, the product is obtainable in each M.2 gumstick type plugin playing cards which might be utilized in AI edge purposes, and PCI Categorical Gen3/4 accelerator playing cards for cloud knowledge heart servers. AI 100 additionally has as much as 144MB of quick on-die SRAM, relying on the variety of cores, and may interface with as much as 32GB of on-card or in-system LPDDR4 DRAM.
When it comes to efficiency, Qualcomm makes some daring claims versus aggressive options for purposes like picture recognition (ResNet-50 mannequin efficiency), for instance, as you’ll be able to see within the graph above the place an AI 100 20 Watt configuration affords considerably higher inferencing throughput than even an NVIDIA T4 GPU at 70 Watts.
Foxconn’s Gloria AI Edge Field is a completely custom-made turnkey industrial system powered by Qualcomm Cloud AI 100 that BKAV will deploy with its AI View digital camera methods, however once more the applying alternatives are broad and various. Supporting each Sub-6 and mmWave (Millimeter Wave) 5G connectivity as effectively, “The Gloria on-premises AI Edge field manufactured by Foxconn Industrial Web will enable us to display unbelievable efficiency with low energy at a aggressive value level,” notes Nguyen Tu Quang, Chairman & CEO, BKAV. “Gloria will assist us to speed up AI purposes in companies and authorities, and revolutionize our work with good cities, good building, and agriculture, amongst different verticals.”
In fact, all purposes, from edge units like Foxconn Gloria to knowledge heart cloud AI processing necessities, that can be serviced by Qualcomm’s PCI Categorical accelerator playing cards, require deep software program instruments and library assist, which is a power for Qualcomm, once more due in-part to the corporate’s a few years of AI-enabled growth, deployment and assist in cellular purposes. Assist for native AI frameworks like TensorFlow, PyTorch and Caffe will enable AI-focused builders to code on their platform of selection after which goal for optimization on Qualcomm Cloud AI 100.
Clearly, the AI house is sizzling proper now and performance-per-watt optimized options like Foxconn’s Gloria are well-suited for the ever-expanding good edge system ecosystem, so the alternatives are sturdy for each firms. Engineering samples of Foxconn’s Gloria AI Edge Field can be accessible later this 12 months, and industrial availability is anticipated within the second quarter of 2022. Qualcomm’s AI 100 devkit (seen above) that can be powered by a Snapdragon 865 platform host processor, with 5G connectivity supported by Qualcomm’s Snapdragon X55 Modem-RF System, are transport to prospects for growth efforts now.