Results: CPU Benchmarks
AnTuTu
AnTuTu is a benchmark designed to test the performance capabilities of four major aspects of mobile devices: Graphics (encompassing 2D, UI and basic 3D), CPU (fixed, floating-point, and threading), RAM (read and write), and I/O (read and write).
Shield comfortably leads in terms of Graphics and CPU performance, which makes sense since it’s up to 100 MHz faster than the Tegra Note and is the only mobile device enjoying active cooling. Tegra Note does come in a very close second, edging out the far more expensive Galaxy Note 10.1, beating it in terms of I/O performance, and finishing almost on par in RAM performance. Tegra Note is priced to compete with Google’s Nexus 7, but manages to keep up and beat out both Sony and Samsung’s flagship devices. Of particular note is the performance of premium devices in terms of RAM throughput: Shield and Xperia Z1 seem to indicate lower latency memory performance, while Samsung’s Note 10.1 does not.
CF-Bench
Chainfire’s (developer of SuperSU, among many other recognised Android tools) CF-Bench Pro is designed to characterize the performance of multi-core systems by simulating loads in Java both in terms of natively compiled and managed code. It tends to be a decent indicator of the tweaks vendors make to Android to eke out extra performance when they optimise with specific hardware features in mind. Nvidia’s Shield wouldn’t run this benchmark, and is thus absent.
Sony has been optimising for Snapdragon-based devices since the Xperia range took on the Krait core, and its experience shows as the Xperia Z1 comfortably leads the Tegra Note and Galaxy Note 10.1 in both Managed and Native. Still, Nvidia’s Tegra Note 7 easily takes second place overall, despite being slower in terms of native performance than Samsung’s Exynos 5-based 10” tablet. Critically, Exynos seems to have to a lot of raw performance, as demonstrated by its substantial Native Code score. But Samsung doesn’t seem to be optimising at the Java side as much. Tegra Note, again, performs extremely well considering it’s priced to compete with Google’s Nexus 7.
Geekbench
Primate Labs’ Geekbench is somewhat of an industry standard due to its comprehensive database and cross-platform compatibility, supporting x86 Windows, PPC and x86 Apple OS X, iOS, and Android. While Geekbench is more of CPU performance benchmark, it also covers some general memory performance, too. Geekbench’s approach is based on real-world applications and simulations of regularly used tasks (file/data manipulation, compression, encryption, and image processing).
When we isolate single-core performance, Apple’s iPhone 5s sweeps ahead with a substantial lead in every category, which may indicate that Geekbench favors Apple’s A7 ARMv8 64-bit double-wide register approach. Comparing Android devices, it seems that the Exynos 5-powered Galaxy Note 10.1 has considerably more grunt than either Tegra 4 or Snapdragon 800, but it’s necessary to weigh this data critically in light of recent revelations. Tegra Note runs almost neck and neck with Shield, taking the lead in memory throughput and floating-point math, but losing in the integer testing. Tegra Note once again proves to be an extremely good value for you spend on it.
In multi-core testing, the finishing order gets reshuffled quite extensively. Sony’s SD800-based Xperia Z1 takes a reasonable lead, proving that Apple’s reliance on ARMv8’s 64-bit double-wide registers isn’t the whole story. The use case for single-core in mobile computing is continually shrinking, and these results may be more indicative of real-world performance. Shield's active cooling might help explain the complete turnabout from the previous single-core results, as more resources remain available due to less heat. Tegra Note, which was ahead of Shield in single-core performance, now sits well behind by at least a 10%.
MobileXPRT
Principled Technologies’ MobileXPRT is not dissimilar to certain aspects of Geekbench in that it simulates real world applications and use cases. Where it differs is twofold: first, it’s Android-only, and second, it simulates some more modern aspects that Geekbench doesn’t, namely Biometrics (Facial Recognition), UX (List Scroll, Grid Scroll, Gallery Scroll, Browser Scroll, and Zoom and Pinch to Zoom), and Photo Management (Apply Photo Effects, Create Photo Collages, Create Slideshow and Organize Photos).
Anyone who has used Shield will tell you that it yields a very smooth experience, and that’s likely attributable to specific software optimisations implemented by Nvidia (such optimisations also might indicate why CF-Bench would not run at all on the device). The platform is in full effect here; Shield blows this test away in the performance tests and maxes out the User Experience suite. Tegra Note’s theoretical clock rate deficit might be reflected in the slower Performance Tests result, but could indicate fewer optimisations or tweaks applied to the Tegra Note (which, again, would go some way to explaining why it could run CF-Bench, while Shield could not).