The Xiaomi Mi MIX 2S Review: Fantastic Overall Value

GPU Performance

On the GPU performance side, I’m expecting again very good results as we’ve investigated the performance of the Snapdragon 845 and the Adreno 630 GPU in depth in our Galaxy S9 review. The Mi MIX 2S should not differ too much from the S9 – apart from maybe slight differences in thermal dissipation capabilities or maybe different thermal throttling configurations.

In Futuremark’s Sling Shot 3.1 Extreme Unlimited Physics test (I’m hoping for shorter benchmark names in the future), the Mi MIX 2S actually performed quite a bit less than the Galaxy S9+. This test is mostly CPU bound, and the only explanation I have here for the results is that Xiaomi must have had made changes to the thermal balance between CPU and GPU and limited the CPU performance more in 3D workloads.

In the graphics score results of the test, the MIX 2S performed a lot closer to the Galaxy S9, posting identical peak performance figures while having only slightly slower sustained performance. Unfortunately we’ve verified both with a second Galaxy S9+ as well as now with the Mi MIX 2S that the sustained performance degradation of the Snapdragon 845 was not an isolated case. This makes sense given that the increased peak performance of the Adreno 630 seems to have come from increased power draw, exceeding the 5W figure in ALU heavy workloads. As naturally phone thermal envelopes can only sustain 3-3.5W SoC TDPs, performance has to go down over time to keep temperatures in check.

Manhattan 3.1 Off-screen performs again nearly identically to the Galaxy S9+, topping the benchmark scores.

T-Rex also matches the Galaxy S9+. The performance of the S845 devices here seems to be lower as the benchmark is running at high fps figures and demanding a lot of memory bandwidth – it’s possible that the increased memory latency on the S845 chips is hindering the performance and efficiency of the chip compared to the S835 in this case.

Overall, in terms of 3D performance the Mi MIX 2S leaves little complain about. While the increased peak power is of a concern, I’ve only found few games and real scenarios where the higher GPU frequencies are actually currently needed. Over the coming months I’m also planning to expand the GPU testing methodology and among other things try to introduce a “real game” battery or power draw test that doesn’t actually require peak performance, but rather maximum efficiency for a given workload. We’ll also hoping to introduce Vulkan GPU tests from Kishonti as well as evaluating the recently announced Basemark GPU.