Samsung's Galaxy S20 series phones have been available since last Friday in markets such as the US. And earlier this week we also finally received a unit, in the form of a North American, Snapdragon 865-based Galaxy S20 Ultra. While our review is already underway, we’re also still waiting for public availability in Europe in order to get our hands on our Rest of World, Exynos 990 variant, so that we can take a comprehensive look at both variants of the S20 series. As we've seen in previous years, there have been some pretty significant differences between the Snapdragon and Exynos models at times, thanks to the SoC selection impacting everything from performance to image processing.

But first things first: since we have a bit of a lead time with the Snapdragon unit, we wanted to at least publish the performance figures for this model ahead of the full review, to temporarily satisfy everyone’s curiosity on at least this aspect of the phone.

Read More:

System Performance

System performance of the new Galaxy S20 Ultra is an interesting topic, as there are several new aspects to this year’s flagship phone. The one big difference that trumps every other addition is the fact that Samsung has been able to integrate a new 120Hz refresh rate display. This change alone puts the new S20 series far ahead of other mainstream phones in the market, and the new experience is fantastic.

Besides the new higher refresh rate screen, we’re also naturally seeing the upgrade to newer generation SoCs. In this first instance, we’re testing the Snapdragon 865 variant of the Galaxy S20 Ultra. For their latest flagship SoC, Qualcomm adopted the new Cortex-A77 CPU cores, promising to bring 20-25% higher performance over its predecessor.

Finally, we do have to remember that Samsung has a “performance” feature in its battery settings, which increases the aggressiveness of the scheduler to fully unlock the performance of the phone. Usually we test Samsung phones with this option enabled, both in our performance as well as battery life testing.

Starting off with our usual system performance tests, these evaluations are highly sensitive to the responsiveness of the phone, which is tied to the aggressiveness of the DVFS and scheduler of the CPUs. For the Galaxy S20 Ultra, we have four score combinations, showcasing the 60 and 120Hz modes, as well as the “High Performance” mode on or off.

PCMark Work 2.0 - Web Browsing 2.0

The web browsing test in PCMark is quite sensitive to performance responsiveness, and in this regard, the new Snapdragon 865 doesn’t disappoint. Switching between the 60 and 120Hz modes, we see a notable increase in fluidity, and this is picked up by the benchmark.

At the highest performing settings, the new Galaxy S20 Ultra even outperforms the QRD865 platform that we tested back in December. This was quite surprising, as I wasn't expecting commercial devices to ship with as aggressive settings as that phone’s “Performance Mode”, which did seem tad aggressive.

PCMark Work 2.0 - Video Editing

The video editing test has largely lost its performance scaling usefulness, but still is able to pick up the new 120Hz mode of the S20U, representing a jump ahead of any other phone in the market.

PCMark Work 2.0 - Writing 2.0

The writing sub-test is very important in terms of being a representation of every-day snappiness of a phone, and the Galaxy S20 Ultra here tops the charts, falling in line with the best scores from the QRD865 as well as now slightly leaping ahead of Huawei’s Mate 30 Pro.

PCMark Work 2.0 - Photo Editing 2.0

The photo editing test similarly is scaling up in performance across the different setting configurations, showcasing fantastic performance.

PCMark Work 2.0 - Data Manipulation

The data manipulation score also seemingly is tied to the framerate achieved during the test, and the 120Hz mode of the S20U leads all other devices.

PCMark Work 2.0 - Performance

In the overall results, no matter what setting you’re using, the Galaxy S20 Ultra with the Snapdragon 865 tops all other commercially available Android phones out there, and at its peak settings, it even outperforms the QRD865 in its aggressive performance mode.

I did some quick testing of the DVFS aggressiveness between the optimized and performance modes, and was surprised to see no difference in the resulting scaling behavior. This means that the performance differences must arrive from the overall more aggressive scheduling, rather than scaling up to higher frequencies sooner. We’ll go over this aspect in more detail in the full review.

Speedometer 2.0 - OS WebView WebXPRT 3 - OS WebView JetStream 2 - OS Webview

In all our three web tests, we see the S20 Ultra in line with what the QRD865 was able to achieve, with some slight leads in WebXPRT 3. The Cortex-A77 seemingly doesn’t improve its instruction throughput very much in high instruction pressure heavy workloads such as the web-based JS benchmarks, so that’s possibly why we aren’t seeing that large increases here.

CPU Performance & Power

Our power testing back in December on the QRD865 platform wasn't quite as accurate as I would have hoped for, as the phone's power management funcitonality as well PMIC calibration weren’t quite polished. I had opted to publish numbers on the more pessimistic side of the scale, and I’m glad I did as it does turn out that the new chipset performs quite a bit better in actual commercial devices, including of course the S20 Ultra.

These new figures mean that the Snapdragon 865 actually does not behave as expected – we had been anticipating it requiring more power to achieve its higher performance points – and instead Qualcomm has managed to reduce absolute power all while increasing the performance. As a result, we’re seeing a much larger generational improvement in the energy efficiency of the chip. The new “middle” cores of the Snapdragon 865 also outperform the performance cores of the Snapdragon 855, and that does signify for quite a large multi-threaded performance boost.

We’ll be going over the detailed results as well as include an analysis of the Exynos 990 chip in our full review, but for now, it seems that the Snapdragon 865 is an excellent chipset, and will serve as a great base for 2020 devices.

GPU Performance & Power
POST A COMMENT

116 Comments

View All Comments

  • masimilianzo - Friday, March 13, 2020 - link

    Hi Andrei,
    Just a small request..going forward could you include a disclaimer about your methodology for power/energy estimation?
    Reply
  • Andrei Frumusanu - Friday, March 13, 2020 - link

    I'm outright measuring the power via different methods, it's not an estimation. Reply
  • masimilianzo - Friday, March 13, 2020 - link

    Good, see your point.
    Still could you share the different methods you are using?
    Reply
  • dotjaz - Friday, March 13, 2020 - link

    The more I think about it, the more I feel the numbers don't add up. ARM claims no efficiency gain at the same frequency, and TSMC only promised 10% power reduction, yet perf/W increased by 40%, perf/J a whopping 66%, fp even more. Reply
  • masimilianzo - Friday, March 13, 2020 - link

    In this generation they have double L3, lower memory latency and higher memory bandwidth.
    So it is not a apples-to-apples comparison :)
    S865 will obviously be better
    Reply
  • dotjaz - Friday, March 13, 2020 - link

    That doesn't explain a whopping 30% increase in efficiency. Doubling L3 has diminishing benefit, I doubt it can improve performance by 1% considering 2MB is already very big for single core.
    And single core is obviously not bandwidth bound.
    You can put 16MB L3 and LPDDR5-3200 on it, the single core performance won't increase even 5%.
    Reply
  • Andrei Frumusanu - Friday, March 13, 2020 - link

    You're just wrong there, both those aspects will have huge impacts on ST perf and efficiency. Reply
  • dotjaz - Friday, March 13, 2020 - link

    That's impossible. No amount of cache and memory bandwidth can improve A77's efficiency by 30 to 80 percent given the baseline is 2MB L3$ + 3MB Sys$ for a single 2-watt CPU core.

    Otherwise Ryzen 2600X would have the same advantage over 2500X, yet it definitely doesn't.
    Reply
  • Wilco1 - Friday, March 13, 2020 - link

    Increasing caches has a huge impact on efficiency - on memory intensive benchmarks you will now be able to run everything from L3 rather than waiting forever on DRAM. A quadrupling of caches would significantly reduce the gap with Apple, gaining 20-30% is easy *if* you are willing to spend the area. Reply
  • dotjaz - Friday, March 13, 2020 - link

    So then, you can explain why Thunder didn't have that with doubled L2$. Not even close to 66% per J improvement. Reply

Log in

Don't have an account? Sign up now