Every time Qualcomm launched a new high-end mobile processor, the next question on everyone’s lips is: “what’s the performance like”?
Some performance metrics were disclosed during the Snapdragon Summit keynote, but now Qualcomm has released actual benchmarking numbers, most likely run on development devices called MDP.
Keep in mind that the purpose of the MDP device is not to optimize performance. It is to allow developers to get a head start with pre-production hardware (and drivers). Therefore, we expect these numbers to be slightly conservative compared to a commercial device with fully optimized drivers.
We have been following Qualcomm for many years, and the numbers the company provides have always been pretty accurate. Typically, reviewers and experts would also be given a chance to test the MDPs for themselves and check their speed before eventually experiencing the same performance on commercial devices.
Starting with CPUs, the Geekbench CPU benchmark stresses the CPU cores, along with the memory sub-system. If we compare it with Snapdragon 865 (in the Galaxy S20 Ultra), the Snapdragon 888’s speed increase is 25.6% and 19.2% for single-core and multi-core tests, respectively.
Apple’s latest processor performs even better in that specific Geekbench benchmark. It is essential to recognize it, because Apple has consistently won that test for a number of years.
For graphics performance, we look at the GFXBench Manhattan 3.1 (offscreen), which is not the most recent benchmark, but it is generally representative. Snapdragon 888 is 31% faster than the Galaxy S20 Ultra (Snapdragon 865), which is quite close to Qualcomm’s claim. Some applications may show even more significant gains.
For Artificial Intelligence (AI) performance, Qualcomm mentions Ludashi AIMark, we ran the numbers on the Galaxy S20 Ultra, and Snapdragon 888 was about 2X faster (104,967 vs. 217,223 according to Qualcomm). A lot of it has to do with Snapdragon 888’s new Hexagon unit that has seen a massive upgrade since last year.
AI is typically not used for sustained periods, but the amount of computation is insanely massive when needed. The multi-layer processing for photo capture is an excellent example of that “spiky” compute requirement.
We’re confident that commercial devices, probably starting with the Samsung Galaxy S21, will reflect this performance’s generational leap, and countless more phones will follow. We will check again in our review cycle of 2021.
It’s also important to remember that benchmarks don’t measure 100% of the performance and that some of it remain hidden from the view of such tools. After all, Snapdragon and many other SoCs are platforms (including security and all), and not merely a compute unit.