The Mate 20 & Mate 20 Pro Review: Kirin 980 Powering Two Contrasting Devices

GPU Performance & Power

The Kirin 980 is the first SoC to sport Arm’s newest generation Mali G76 GPU. The new IP differs significantly to previous generations, in more or less simplified terms, in that the GPU cores are essentially twice as big and capable as the previous generation Mali G72 cores. So while the G76MP10 configuration of the Kirin 980 might sound small, it’s not small at all in terms of theoretical performance.

GPU performance and efficiency has been a big thorn in the side of both the Kirin 960 and 970, as both SoCs showcased less than stellar power figures, which in turn also resulted in forced limited clocks and performance of the GPUs. It’s here that Huawei made the biggest promises in terms of improvements: a claimed 46% increase in performance while showcasing a staggering 178% increase in power efficiency. The latter figure especially caught some attention, as you just don’t see such increases in the industry.

Starting off with 3DMark Sling Shot Extreme Unlimited and the Physics sub-test, we see the Mate 20’s showcase some leading peak performance figures. This test is mainly a CPU test with just some more minor GPU load. The performance jump here undoubtedly comes from the new Cortex A76 microarchitecture.

In terms of sustained performance, we see some diverging figures between the Mate 20 and Mate 20 Pro, as the Pro is able to reach much higher sustained scores. Before getting into any conclusions, it’s worth to also look at the GPU results.

On the Graphics sub-test, we see both new Mate 20’s reach respectable peak performance figures, however they are both still throttling quite a lot until they reach thermal equilibrium. Comparing the results to the stock firmware Kirin 970’s, such as the P20 Pro, the performance increase is nevertheless quite significant.

In the new Aztec Ruins Vulkan benchmarks, both in High and Normal quality modes, we see some really odd performance behaviour. While the peak performance isn’t all that great, the sustained performance is pretty much almost identical. On the Normal run the Mate 20 Pro actually was able to maintain a higher performance than the Mate 20, something that we also saw on the 3DMark Physics run. It would be definitely interesting if the benchmark is in some way CPU bound, or if the devices have different thermal limits between Vulkan and OpenGLES workloads.

In Manhattan 3.1, we see again respectable performance gains both in peak and sustained figures. Compared to the Vulkan runs, these scores showcase a more expected delta between peak and sustained. The Kirin 980 here generally matches most Snapdragon 845 devices – short of the OnePlus 6 and G7 which seem to allow much higher sustained power limits.

Looking at the power efficiency during Manhattan 3.1, we unfortunately see that the phone and chipset didn’t quite meet my projections in efficiency. Performance is exactly where it should be, however the power is off by about 1W as I had hoped to see about 3.5W peak power. At peak performance of both chipsets, the Kirin 980 showcases a 100% efficiency gain over the Kirin 970, which is still a pretty massive generational improvement, even if the previous generation didn’t exactly set the bar all that high.

In regards to Huawei’s 178% power efficiency claim during the chipset’s announcement: I still think this number is correct, however evidently this was a traditional case of somewhat misleading presentation or a mixup between “or” and “and” in the relationship between the performance and power efficiency improvements. Now in hindsight, the 178% efficiency figure likely refers to the efficiency advantage of the Kirin 980 at the same performance of the Kirin 970, which given the measured power figures here, is something that’s definitely plausible.

In T-Rex, the peak performance improvements over the Kirin 970 are far less, and I do wonder exactly what the bottleneck here is. Nevertheless, the sustained performance jumps 50%, but yet again this is just for the Mate 20 Pro as the regular Mate 20 sees far more severe throttling. T-Rex would be in many ways CPU bound as it’s hitting very high frame-rates on modern SoCs.

Again, the power efficiency as measured on T-Rex sees a significant jump over the Kirin 970, however most of this improvement is simply going towards reducing the actual power usage from the ridiculously high values of its predecessor, with only a little gained peak performance.

I wouldn’t take this as a definitive verdict on the Mali G76 as of yet, as over the last 3 generations Samsung has been able to extract much better results out of their GPU implementations inside the Exynos SoCs than what HiSilicon was able to achieve in the Kirins. The next generation Exynos 9820 should be able to do better than this, so maybe that’s where the Mali G76 will hit its projected targets.

Overall, the Kirin 980 definitely is posting substantial improvements over its predecessor, however Arm’s Mali GPU still seems to lag a tad behind the higher end competition from Apple and Qualcomm. What is definitely positive for Huawei is that the new SoC finally is able to shed off the atrocious performance showcased in the previous generation chipsets, and is now actually competitive with most recent devices.