Qualcomm is set to launch its next flagship mobile chipset, likely to be called the Snapdragon 8150, next month. The mobile chipset will bring new capabilities in terms of CPU and GPU performance but one area where it could outshine rival mobile SoCs could be in AI performance.
A new benchmark test reveals the upcoming Snapdragon 8150 from Qualcomm might claim the top spot among all Android smartphones. A test of the developer platform of the Snapdragon 8150, i.e, a prototype board, shows close to double the improvement in performance in AI computation. It shows that the chip will be much faster than Huawei Kirin 980, which has dual NPUs and has a score that is only equivalent to Snapdragon 845.
With Kirin 980, Huawei is claiming double the performance improvement in AI tasks over its predecessor, the Kirin 970. However, this benchmark listing shows the improvement is only marginal and equivalent to only 15 percent. The Kirin 980 has a score that is equivalent to that of the Snapdragon 845, which seems like a disappointment considering the company claimed twice the improvement in terms of AI performance.
This benchmark listing states that the prototype board of Snapdragon 8150 got a total AI score of 22,082 running on a device with 6GB RAM and Android 9 Pie. It beat MediaTek Helio P80 Dev Platform, which scored 19,453 with 6GB of RAM and Android 9 Pie. The interesting thing, however, is how Snapdragon 845 performed on two different kind of devices.
The Snapdragon 845 score from the OnePlus 6 is much higher than the one from Google Pixel 3 XL. The listing shows that OnePlus 6 with Snapdragon 845 scored 12,274 in AI benchmark, while the Pixel 3 XL scored just 6,895. The Huawei P20 Pro with Kirin 970 managed to beat the Pixel 3 XL in this testing an AI score of 8,921. MediaTek Helio P60 Dev Platform rounds up the list with an AI score of 5,767.
Watch: OnePlus 6T Thunder Purple First Look
With most computing switching to machines and AI for basic functions, it is important for these chips to be much faster. The test for AI benchmark involves object recognition with neural network MobileNet – V2 and Inception – V3, face recognition, image deblurring, image super-resolution with neural network VGG-19 and SRGAN, semantic image segmentation, photo enhancement and memory limits.