Harish Jonnalagadda / Android Authority
Add Android Authority on Google:
I like benchmarks a little too much and use them on every new phone I test. Synthetic benchmarks like Geekbench and 3DMark do a good job of simulating real-world usage and gaming scenarios, and they predict a device’s performance potential reasonably well. As someone who gets over 75 phones a year, a synthetic benchmark lets me easily measure how a particular device holds up against its predecessor.
There’s a new benchmark that aims to go beyond just delivering an overall score. It’s called Evolve, and it provides useful insights across five metrics, allowing you to then pinpoint which part of the GPU is handling modern workloads well, and what’s causing a potential bottleneck. I’ve been testing it with a few phones, including the Find X9 Ultra, Find X9 Pro, Find X8 Pro, and the Xiaomi 15 Ultra, and it did a good job showcasing just how much (and where) mobile GPUs have evolved in the last two years.
Did you run any mobile benchmarks recently?
8 votes
Yes, I benchmark every time I get a new phone.
50%
No, just don’t see the point these days.
50%
Evolve shows what a mobile GPU is actually doing
Harish Jonnalagadda / Android Authority
We’re starting to see ray tracing on mobile, and Arm and Qualcomm are going all-in on neural rendering, leveraging techniques like frame generation to bring higher framerates to mobile titles. It’s in this area where Evolve comes into its own; it has a dedicated score that takes into account all the hardware-accelerated ray tracing workloads — lighting, shadows, and reflections — done via Vulkan Raytracing and DXR APIs, and it does a good job indicating how well a phone handles ray tracing in general. There’s also an Acceleration Structure score that evaluates how quickly the GPU builds all the data structures that are necessary for ray tracing.
Similarly, Evolve has a rasterization score that tests the hardware rasterizer (which is dedicated to 3D graphics), a test that measures animation, graphics effects, and post-processing scenarios, and a driver score that measures how efficiently the CPU is being utilized. The idea with all of these tests is to provide a realistic representation of how a particular mobile GPU holds up during gaming. Evolve itself is free, but like 3DMark, there’s a professional version which lets you export charts and all testing data, and this is what I used.
Putting Evolve to work
Harish Jonnalagadda / Android Authority
I tested the last two generations of Qualcomm and MediaTek chipsets with Evolve. That includes the Snapdragon 8 Elite Gen 5 (Find X9 Ultra), Dimensity 9500 (Find X9 Pro), Snapdragon 8 Elite (Xiaomi 15 Ultra), and the Dimensity 9400 (Find X8 Pro).
As Evolve’s charts only mention the GPU (which is what it is testing anyway), a quick overview: the Snapdragon 8 Elite Gen 5 has the Adreno 840, the Dimensity 9500 has the Mali G1, the Snapdragon 8 Elite uses the Adreno 830, and the Dimensity 9400 has the Mali-G925. With that out of the way, let’s take a look at the scores.
Harish Jonnalagadda / Android Authority
It shouldn’t come as any surprise that Qualcomm absolutely dominates the charts; the last two generations of Adreno GPUs have delivered a sizeable boost in performance, and that reflects in the score. The Adreno 840 is 27% faster at the ray tracing workload than the Mali G1, 40% faster at rasterization, and it has a similarly-outsized 46% lead in GPU compute. Of course, it doesn’t mean that the Adreno 840 is 40% better at gaming than the G1; in real-world gaming situations, the difference is much closer than that. It’s just that the Adreno 840 has much better headroom, and that makes itself evident in sustained gaming scenarios.
What’s notable to me is just how much of a year-on-year difference there is between the Adreno 830 and Adreno 840; the latter does 1.47 times better in the rasterization workload, and 85% better at ray tracing. Also interesting to see is how the Mali G1 fares against the last-gen G925-Immortalis; Arm touted a 2x increase in ray tracing with the G1, and that is evident in the scores. That said, the G1 doesn’t have a drastic uptick in other areas — certainly not on the same scale as what Qualcomm achieved with the Adreno 840.
Another thing Evolve does well is show just how much GPU time has been utilized per frame in each workload; this is useful as it shows you just what is taxing the GPU. In the case of the Adreno 840, 27.7% was spent towards ray tracing, just over 17% towards rasterization, and a bulk of the time (45.7%) was dedicated to GPU compute. With the Mali-G925, over 50% was dedicated towards ray tracing as it doesn’t have the requisite cores to handle this workload. These charts are interesting as they detail just where a particular GPU is struggling, and it’s great to see this level of granularity.
Evolve also has an energy score that measures how much power the GPU is using when running a workload. This is useful to test on mobile chipsets, as they need to be energy-efficient by their inherent design. That said, the feature wasn’t available on the phones I used, so I talked to Jasper Bekkers, founder of Evolve, to get a better understanding of what’s going on. He said, “To calculate the energy score, Evolve reads actual power data from the device through Android’s Power Stats HAL. Some manufacturers don’t expose this, and rather than guess based on the battery API, we leave the score out on those devices.”
Harish Jonnalagadda / Android Authority
Looking at these charts, I can’t help but be excited about what’s coming with the Snapdragon 8 Elite Extreme Gen 6 and Dimensity 9600 Pro. Both platforms switched to the 2nm node, with Qualcomm and MediaTek touting generational GPU gains. I’m using the Dimensity 9600 Pro (it’s a beast) and will talk more on that in the coming weeks, and Qualcomm’s 2027 flagship chipset should get a global release relatively soon, so I’m looking forward to testing phones running these designs.