The Snapdragon 865 is the reference chip on board the terminals with Android that we see throughout 2020. It will probably even be in more devices than we are accustomed if it finally lands on board the international versions of the Galaxy S11 .
The next Qualcomm chip arrived with some modest, although solid, improvements in the performance section of the terminal. These fit the figures that the manufacturer had promised throughout the different sections: improved processing latency which has an impact on performance, as well as the adjustment of the effective CPU and GPU power.
Or so say the first analyzes in relative depth published on Qualcomm's own development unit, the QRD865 platform. This has the next standard of LPDDR5 memory, and of course the Snapdragon 865.
Snapdragon 865 performance: critical improvement in AI, not so much in the rest
According to the data of the specialized hardware site AnandTech, the area where we see higher profits is in the section of machine learning or machine learning. In it we see a performance improvement combined to the entire system that doubles or triples the one seen in Snapdragon 855.
This fits again with the figures provided by the manufacturer itself, which stood at 15 TOPS billions of operations per second for this generation, compared to the 7 TOPS of the last generation. In both cases, the inference capacity of both the CPU and the GPU and the cores dedicated to this type of tasks in the DSP and NPU is taken into account.
While the gain in the CPU section is around 25-30% depending on whether we talk about performance in the integer calculation or in floating point, which are somewhat more modest in the graphic section, around 20-25% improvement Depending on the specific test.
On the other hand, when we consider a measuring tool or benchmark that takes into account all the specific implementation of the chipset of each device that is different if we talk about the Kirin 990 of Huawei, the Apple A13 or these Snapdragon 865, we see that Qualcomm is located as the great reference in terms of performance of machine learning. Beat the rest of the alternatives by more than 100%, taking into account the Snapdragon 855 or the Apple A13 Bionic.
Which does not give him the lead immediately. And it is that all these available APIs that developers can implement in their applications, and that will be more decisive especially in some graphic type such as the camera, they will be in specific situations. Therefore, all this performance is not always used and there may be occasions where other manufacturers perform better than this.
In general, the Snapdragon 865 seems to be in these first analyzes a robust evolution over the past generation, although due to a relative stagnation in the 7-nanometer process and higher performance, we do not see a great evolution in energy efficiency. More good, performance seems to partially improve due to higher energy consumption, especially in the graphic section.
MS in Hypertext