The Sony Ericsson Xperia Play: Where Do You Want To Take Your Gaming Today?

At Brian Klug's recommendation, I ran the Xperia Play through a battery of common benchmarking tests. Below you'll find the alphabetically ordered results, in comparison to similar-featured and similar-age Android handsets from other suppliers:

Basemark (aka 3DMarkMobile) ES 2.0 V1

BrowserMark

GLBenchmark 2.0 ('Standard' results in both cases)

kwaak3 aka Quake 3 (note that this particular test was run with lightmaps off, per longstanding AnandTech practice)

Linpack

And SunSpider 0.9

The results are largely predictable, but some encouraging surprises also emerged. Performance differences between platforms based on the first-generation 65 nm-based QSD8x50 (aka S1-class SoC) and second-generation 45 nm-based MSM8x50 SoCs predominantly derive from two factors, since the processor cores are fundamentally the same save for the process they're built on:

·       Clock speed differences (800 MHz versus 1 GHz, for example), and

·       Graphics core differences (Adreno 200 with the QSD8x50, Adreno 205 with the MSM8x55)

Slight variances between similarly clocked systems in CPU-centric benchmarks are likely the result of benchmark run-to-run differences, coupled with minor evolutionary Dalvik virtual machine enhancements in successive Android versions. Perhaps the most significant Dalvik improvement came with the transition to Android 2.2 'Froyo', which added a JIT (just-in-time) compiler to the mix with a resultant claimed 2x–5x performance boost with CPU-heavy code. The Xperia Play still notably outperforms its fellow MSM8x55- and Adreno 205-based peers on graphics-heavy benchmarks. This result surprised me, until Brian Klug explained that the Xperia Play represented one of the first production systems containing improved graphics drivers from Qualcomm.

Unlike with a PC, where you can regularly install updated drivers from AMD, Intel or Nvidia, with a handset you need to rely on Google and its hardware and carrier partners to bundle the updated drivers with an operating system upgrade for your hardware...or in this particular case, for an OEM like Sony Ericsson to include them from the get-go. Note, for example, the frame rate improvements of the Xperia Play on Basemark ES 2.0's 'Taiji' and Hoverjet tests, versus the seemingly identical (at least from CPU and GPU standpoints) HTC Thunderbolt. Similarly, the Xperia Play had notably higher frame rates in the Quake 3-based kwaak3 test, versus the Thunderbolt.

One other result was also intriguing, until I thought about it for a bit. Linpack recently added a multi-threaded benchmark option, so I decided to run both it and the legacy single-threaded test. The multi-threaded MFLOPs (millions of floating-point operations per second) was lower than that for the single-threaded test variant; consider, however, that the MSM8x55 is a single-core CPU. I believe that the decrease is the result of multi-thread management thread-switching latency overhead, in spite of the Cortex-A8 superscalar dual-issue microarchecture foundation upon which Qualcomm built its Scorpion approach. Conversely, if you look at the Linpack data for a system such as the HTC Sensation 4G, based on a true dual-core CPU such as Qualcomm's MSM8x60 (S3-class SoC), you'll see dramatically higher multi-threaded Linpack results in comparison to those for the traditional single-threaded test.

FYI I also tested the Xperia Play using Qualcomm's recently introduced Vellamo benchmarking suite, but AnandTech isn't quite ready yet to start publishing Vellamo results. Stay tuned for more in this regard. Similarly, I ran the Xperia Play through the newer SunSpider 0.9.1, whose results we'll start publishing once we have a sufficient-sized system sample set.