18 Apr Qualcomm Fills Out The Rest Of Its AI Portfolio
Qualcomm is widely known as a powerhouse of wireless communications. Lesser known is the fact that it has been working on AI for longer than most of its competitors. Qualcomm announced its Zeroth processors back in 2013, which were far ahead of their time. Much of the industry was not quite ready yet for dedicated AI processors. Qualcomm then pivoted its AI strategy towards software-defined, utilizing the best possible cores inside of its smartphone SoC (which can range anywhere from the CPU, GPU, or DSP). With its launch of the Snapdragon 855 in December, Qualcomm re-introduced dedicated AI processing capabilities into its chips and vastly boosted their AI performance. The Snapdragon 855 is already shipping in phones like the Samsung Galaxy S10 and LG G8. Qualcomm followed up the 855 launch with its recent AI Day in San Francisco, where it announced several new AI-powered products.
I attended the event with my colleague, Karl Freund, who focused his coverage on the new Cloud AI 100 processor for inference (you can read his analysis here, if interested). For me, the most interesting announcements Qualcomm made at this event were its three new SoCs, which it refers to as “platforms.” These SoCs were not announced for the high-end tier of smartphones, but rather the mid-range. This is important, because it means they will ship in the highest volumes and make the biggest impact on AI performance on a global scale. Let’s take a closer look.
Three new SoCs
These new SoCs that Qualcomm announced are the Snapdragon 730, 730G, and 665. The Snapdragon 665 integrates the last generation of Qualcomm’s AI engine (3rd Gen) into Qualcomm’s mid-range SoCs. The Snapdragon 665 also adds triple camera and 48MP sensor support, bringing these features from the Snapdragon 855 down to the mid-range. It improves the camera and AI performance of the 600-series, but doesn’t quite bring it up to what’s offered in the 800 and 700 series. The Snapdragon 665 is designed to enable high-end features at mid-tier device prices.
The Snapdragon 730 and Snapdragon 730G are effectively the same SoC. They include the new computer-vision-enabled ISP (CV-ISP), which is designed to accelerate certain machine learning computer vision functions. These capabilities enable smartphone features such as 4K HDR video, portrait mode video, super resolution, and super night shot. The 730 chips feature the 4th generation of Qualcomm’s AI Engine, which has all of the major performance improvements and new cores for AI that are in the Snapdragon 855. This includes the newly added Hexagon Tensor accelerator for the dedicated processing of all kinds of workloads, including voice, photography, gaming, and security.
The 730G’s “G” designation means it is catered specifically to the gaming market within the mid-tier. Gaming phones are growing as a market segment, and the Snapdragon 730G will further enable this by supporting the full suite of Snapdragon Elite Gaming features that were introduced with the Snapdragon 855. The GPU in the Snapdragon 730G is 15% faster than the standard 730 and 25% faster than the current Snapdragon 710. I believe this is a good move for Qualcomm and the industry; there is absolutely a desire from smartphone OEMs and their customers for an affordable, gaming-centric SoC.
Qualcomm has clearly spent a lot of time thinking about its AI capabilities and what the market will need in the future. It was already in a leadership position with the Snapdragon 855, but the Snapdragon 730, 730G, and 655 broaden the market for its mobile AI capabilities even further. I believe that in the future most AI inference will be done on smartphones and other edge devices. Qualcomm’s deep involvement in the smartphone industry gives them a low power edge in mobile and cloud inference that its competitors may not have. While the company doesn’t seem to be as concerned with training as others, I believe that ultimately most AI processing will be inference anyways. AI will ultimately be a balance between cloud inference and local device inference. I believe that Qualcomm is one of the few companies that can strike that balance, with its intensive experience in devices, AI, and connectivity.