GPU Performance

3D and GPU performance of the Pixel 4, much like all other devices this year with the same Snapdragon 855 chipset, will only be able to differentiate itself from the pack if it has any kind of special heat dissipation or extremely lax thermal throttling designs. We’re not expecting any big surprises here, and do hope the Pixel 4 XL is able to fare competitively.

3DMark Sling Shot 3.1 Extreme Unlimited - Physics

Starting off with the 3DMark Physics test, which is actually a CPU benchmark within a temperature constrained test scenario, we see the Pixel 4 XL fall in line with the middle of the pack of Snapdragon 855 devices in terms of the sustained performance scores. It’s interesting to see the peak performance standing out and being ahead by a measurable margin against other S855 devices. I’m not too sure why this would be other than maybe Google having extra optimisations in the scheduling of the workload, or maybe even DVFS behaviour of the CPUs, as the actual workload performance shouldn’t change based on any other external factors such as drivers or software.

3DMark Sling Shot 3.1 Extreme Unlimited - Graphics

In the graphics workload, things are GPU bound and that’s the main limiting factor for the performance scores. Here the Pixel 4 XL again falls around the middle of the pack amongst other S855 devices.

GFXBench Aztec Ruins - Normal - Vulkan/Metal - Off-screen GFXBench Aztec Ruins - High - Vulkan/Metal - Off-screen GFXBench Manhattan 3.1 Off-screen GFXBench T-Rex 2.7 Off-screen

This ranking is continued on over all the GFXBench tests as the Pixel 4 XL does adequately but still remains below medium amongst our Snapdragon 855 devices. A peculiarity we’re seeing in the benchmarks is that the peak performance of the Pixel 4 XL is a few percentages lower than that on other S855 phones. Again, I have no proper explanation for this other that it may be some regression in Qualcomm’s GPU drivers, or that maybe Google is being more relaxed on other DVFS behaviour such as on the memory controllers.

Again, whilst this performance isn’t outright bad, we have to keep in mind the pricing of the phone and its very late release date in the year. The contrast to Apple’s iPhone 11s here in the charts is pretty absurd, as it’s able to showcase scores essentially twice as fast as what the Pixel 4 XL can achieve.

System Performance Display Measurement
Comments Locked

159 Comments

View All Comments

  • dudedud - Friday, November 8, 2019 - link

    > rarely see utilization go beyond 900MB on my current phone

    Are you using just one app or do you still have a 1GB phone?
  • PeachNCream - Friday, November 8, 2019 - link

    The phone has 2GB of RAM. I use a lot more than one app at a time.
  • Oliseo - Sunday, November 10, 2019 - link

    You can't possibly if you're only using less than 1Gb.
  • PeachNCream - Sunday, November 10, 2019 - link

    I don't understand why that's so impossible. My prior phone had 1GB and hovered around 600MB utilization with occasional need for up to 750MB. Both my current phone and my prior one saw/see heavy use as a stand-in for my laptop at times when I can't be bothered to turn it on so everything from working on a few documents to gaming have fallen to my phones. On the other hand, I find it odd that people feel RAM-limited at 4GB or would be worried that the next couple of years using a Pixel 4 would push 6GB to its limits.
  • s.yu - Sunday, November 10, 2019 - link

    It's perplexing what phone you're using.
    My S6E eats more than 1GB on a clean boot, and easily surpassed 2GB even in the past when it was still supported. Now it's about a year since it's last update and it could only juggle 2 apps maximum, with lag, and system animations already off. Any additional apps and the oldest will be terminated, not to mention certain popular and well-maintained apps that are just a glorified browser (I'm talking about Taobao and the sort) will only run half-smoothly(it would sometimes get stuck for 3-5 seconds, which was very rare on my Note8) with all background apps terminated, I don't know about hidden processes but in my experience Samsung's app manager puts those to sleep as much as possible, since it puts even active apps to sleep resulting in me missing notifications from almost all apps, only SMS, call, alarm, and occasionally the calendar break through.
    I could only tolerate S6E's performance because it's a backup device I only use for occasional communications and shopping, yet it still can't smoothly handle these light loads.
  • Spunjji - Monday, November 11, 2019 - link

    Your RAM usage looks low specifically because you don't have much RAM. With only 2GB of RAM to go around, your apps will be being suspended and thus not running simultaneously.

    For comparison - my OnePlus 6 has been used for the browser, Signal and WhatsApp so far today and is sitting at 3.7GB of RAM usage.
  • StormyParis - Friday, November 8, 2019 - link

    Very informative thank you; especially the real-life pictures from almost all current flagships.

    I've jumped off the flagship bandwagon several years back. I'm curious how more reasonnably-priced handset (Redmi Note 10 especialy) would fare.
  • id4andrei - Friday, November 8, 2019 - link

    Great review Andrei but I have some questions regarding the benchmarks in general.

    First, on the GPU section, how do you corroborate the massive discrepancy between 3dmark GRAPHICS and the GFXBench series of benchmarks? I mean, on the former, the A series GPU seems on par with the competition, but on the latter the A series GPU is orders of magnitude better. I would interpret it that the difference is actually not between the GPUs but between the API implementations, GFX runs in Metal on ios and in Vulkan on android. Am I correct to read it this way?

    Secondly, on the CPU side, referencing 3dmark PHYSICS, the A-series actually does not keep up with the competition. Now don't get me wrong, I understand that A13 is the most complex mobile CPU on the market, biggest and most expensive die. I also understand your testing with SPEC in an active cooled scenario. It clearly shows the quality of the design, desktop class arch. On the bursty side, the A13 is also on top, referencing Geekbench and other short benches. Why is the 3d mark PHYSICS the Achilles' heel? Talking "phones in hand", past the burst time interval, is Snapdragon able to sustain its performance better and higher than A13?
  • Andrei Frumusanu - Friday, November 8, 2019 - link

    Adrenos performing well in the 3DMark GPU test might be related to some corner case and their microarchitecture does well there.

    Similarly, the physics test might have some prefetcher or inter-core dependency limitation that doesn't fare well on Apple's CPUs.
  • id4andrei - Friday, November 8, 2019 - link

    Thank you for the prompt response. On the GPU, if I may, similarly with how javascript tests are also about browser engine optimizations in addition to hardware, isn't the difference between Vulkan and Metal as implemented in their respective OS, an academic asterisk on GFX numbers?

Log in

Don't have an account? Sign up now