GPU Performance

Much like the System Performance section, I’m not expecting any big surprises in the GPU performance section. The Kirin 980 uses a Mali G76MP10 at 720MHz and while the chip isn’t blazing as much as the competition, it still performs adequately well and is a significant upgrade to last year’s Kirin 970.

3DMark Sling Shot 3.1 Extreme Unlimited - Physics

In the 3DMark Sling Shot physics test, the P30s are taking the top spots in terms of performance, both in peak as well as sustained figures. The limitation here lies mainly on the part of the CPU as well as its thermal throttling characteristics. Both the P30 and P30 Pro barely throttle in this regard, at least not in GPU power constrained scenarios.

3DMark Sling Shot 3.1 Extreme Unlimited - Graphics

In the graphics test, we see expected results on the P30 Pro, however on the smaller P30 there’s essentially no thermal throttling at all, which is extremely peculiar.


P30 Overheating Warning & App shutdown

Shockingly enough, I didn’t manage to make the P30 throttle at all in any of the tests, as before it could even get to a point of thermal equilibrium, the OS would shut down the application and raise a thermal overheating warning. I don’t know what’s going on with devices nowadays that this keeps happening as I’ve encountered the issue in last year’s Qualcomm Galaxy S9+ with release firmware as well. The last time this happened, it was due to disabling of the thermal throttling when the OS was detecting benchmarking applications, however in our case we’re using altered application IDs. Still even with this the smaller P30 overheated repeatedly. The fact that this is an OS warning means it’s triggered by a different driver than the usual SoC thermal drivers, so something must be off on the current firmware.

GFXBench Aztec Ruins - High - Vulkan/Metal - Off-screen

In the GFXBench Vulkan High benchmark we see both P30 and P30 Pro neck-in-neck with quite excellent performance. Again what is interesting here is that both devices perform significantly better than the Mate 20s and the View20 with the same chipset. I explain this through the fact that the P30s come with newer GPU drivers, and Arm must have made more significant improvements in their Vulkan drivers.

GFXBench Aztec Ruins - Normal - Vulkan/Metal - Off-screen

In the Normal variant of the Aztec benchmark, we see the P30 Pro throttle a little more, yet it still manages to showcase much better performance figures than the Mate 20, and also higher peak figures than the Mate 20 Pro & View20. The smaller P30 here posts the best figures, however its sustained performance is so high simply because the device is getting extremely hot. I’ve argued if I should be posting the figures for the P30 at all since if you continue to load the device in this manner it’ll simply crash the application.

GFXBench Manhattan 3.1 Off-screen GFXBench T-Rex 2.7 Off-screen

Overall, GPU performance of the P30s is in line with that of last year’s Snapdragon 845 phones, which is still great. Huawei and HiSilicon still trail behind Samsung’s Exynos Mali GPU implementations, although the difference isn’t all that big this generation.

I hope that Huawei figures out the thermal issues on the smaller P30 and issues a firmware update, I’ll be updating the article with the relevant data once this is all sorted out.

System Performance Display Measurement
Comments Locked

85 Comments

View All Comments

  • Andrei Frumusanu - Thursday, April 18, 2019 - link

    The sensor should be in 4:1 binning mode, otherwise the dynamic range would be much worse.

    No 10-bit HDR or HDR recording.

    4K30: 40Mbps AVC High@5.1 // 25Mbps HEVC High@L5
    FHD60: 25Mbps AVC High@4.2 // 16Mbps HEVC Main@4.1

    The bit-rates are quite low.
  • eastcoast_pete - Thursday, April 18, 2019 - link

    Thanks Andrei! The low bit rates are unfortunate, and likely hurt the quality of the video. If you still have the P30 Pro, could you try the Android app "Open Camera" and see if that allows for higher bit rate recording?
    In general, I don't get why so many phone manufacturers are hamstringing even their flagships like that. At least give users the option of recording at higher bit rate. These days, even many affordable mainstream microSD cards support write speeds of 60 Mb/s and up.
  • Andrei Frumusanu - Friday, April 19, 2019 - link

    In Open Camera:

    AVC tops out at 25Mbps at 4K30 for some reason.
    HEVC can be set to 80Mbps in the settings and results in a 62Mbps file at 4K30.
  • eastcoast_pete - Friday, April 19, 2019 - link

    Thanks Andrei! Appreciate it! Any comments on how the 80 Mbps video looked like?
    I suggested to try this to a. learn what the phone hardware can support and b. hopefully motivate/shame Huawei (and others) with your results to improve their camera software. If Open Camera can go up to 80 Mbps in 4K HEVC recording, the phone is clearly capable of higher bitrates than Huawei uses. As someone interested in digital video, I always want to acquire at the highest bit rate possible, and then compress afterwards (thank you, Handbrake team!). Throwing out data is easy, re-creating details is hard or impossible. Also, higher bit rates generally reduce compression artifacts.
  • s.yu - Friday, April 19, 2019 - link

    Hey Andrei, I've been looking at the 40MP shots of the P20P, 30P, and Mate20P, and a very peculiar thing I noticed is that both the P20P and 30P have false color interpolation artifacts, while the Mate20P avoids that; but in terms of chroma noise, Mate20P and P30P are on the same level, while in terms of texture aggregation, the P20P and Mate20P behave similarly.

    So the suppression of the interpolation artifacts doesn't seem to be due to chroma NR, nor is it in any way correlated to sharpening, *nor is it CFA-bound*, as the P20P shares a CFA with Mate20P while P30P uses RYYB. So do you have any idea why this might be?

    I've been examining the area below the left half of the bridge for chroma noise and the pale-looking tree there exhibits a lot of interpolation artifacts in the P20P/30P shots:
    https://images.anandtech.com/galleries/7027/P30Pro...
    https://images.anandtech.com/galleries/7027/Mate20...
    https://images.anandtech.com/galleries/7027/P20Pro...
  • Andrei Frumusanu - Friday, April 19, 2019 - link

    I think all of it is just results of the lens; if you check the centre of other images they all look fine, but it gets worse and worse off-centre for the P20P and P30P.
  • s.yu - Friday, April 19, 2019 - link

    Hmmm, I couldn't think of an optical artifact with that kind of behavior...I don't think CA would be especially evident on foliage especially when the lighting of the scene renders it very low contrast. Basically in that really typical scene you're seeing blobs of other colors (which are not actually present) in a cloud of yellowish green.

    I chose another set and this time there's orange artifacts in the green bush in the center of the scene possibly with magenta artifacts in white or very faintly pink flowers slightly above it(the generally uniform color of the petals seem artificially "split" by the P30P between magenta and white, but I'm not really sure), again neither are visible on bayer or Mate20P.
    https://images.anandtech.com/galleries/7025/P30Pro...
    https://images.anandtech.com/galleries/7025/Mate20...
    https://images.anandtech.com/galleries/7025/Pixel3...
    It isn't so obvious in the P20P shot, but the severe texture aggregation rendered it impossible to tell. Usually for these artifacts to show, the pattern's frequency has to be high enough.
  • Andrei Frumusanu - Friday, April 19, 2019 - link

    It's hard to separate sensor effects from post-processing effects. Another thing to take into account is that the P30P sends RYYB data directly to the ISP where things get recomputed. It can be either a DTI issues on the sensor, an ISP recombination issue, or simply a senseless noise reduction filter.
  • neothe0ne - Thursday, April 18, 2019 - link

    It looks like the speaker evaluation is missing from the review?
  • Andrei Frumusanu - Thursday, April 18, 2019 - link

    I'll be re-adding that soon / tomorrow, I wanted to add a part regarding stereo separation.

    In general the speaker is good, however it's still a mono bottom firing speaker.

Log in

Don't have an account? Sign up now