LINPACK
As we saw in the previous benchmarks, the iPad 2 has a significant increase in floating point performance vs. its predecessor. This was really pounded home by the LINPACK tests which showed the iPad 2 besting its predecessor with a 292% (~3x) improvement. We ran this rest on the Motorola Xoom running Android 3.0 and the Samsung Galaxy Tab running 2.2. The iPad 2 shows a surprising 336% improvement over the Motorola Xoom on its NIVIDIA Tegra 2 platform and a 724% increase over the Samsung Galaxy Tab’s 1GHz “Hummingbird ARM” processor.
Most surprising was that the iPad 2 was able to perform so significantly better than the Xoom in raw FP performance. The Xoom runs on the Tegra 2 SoC which is essentially a dual-core Cortex A9 platform like the iPad 2. Like NVIDIA, Apple has put its chip design team to work on improving performance on the lower-power version of the A9 configurations.
The doubling of memory bandwidth in the new iPad contributes to some of its FP advantage vs. the original iPad. But the only factor we can think of that would explain such a boost in LINPACK performance vs. both the original iPad and rival Tegra 2-based devices is that A5 must have full support for ARM's NEON vector extensions (the ARM equivalent of Intel's SSE instructions). The NEON extensions, along with the larger vector FPU (VFPU) to support them, are included by default in the A8 family, but are optional in the A9. NVIDIA's Tegra 2 opted for the smaller, lower-power, but weaker ARM FPU, while Apple seems to have included the beefier NEON VFPU in their implementation of the A9. Clearly this is the way to go, and we can expect other ARM SoC vendors to follow suit.
JavaScript Benchmarks
SunSpider is a JavaScript benchmark that shows off both the performance of a device’s JavaScript engine, and (to a lesser extent) the raw processing power of the underlying hardware. A poor JavaScript implementation on an insanely fast device can mask the true performance of the platform. Alternatively, a great JavaScript implementation can make up for a slightly slower system. More and more these days, the major JavaScript implementations are converging and borrowing ideas from each other. JavaScript tests are inherently single-threaded due to the nature of browsers. The SunSpider avoids many micro-benchmarks like we see in applications like GeekBench, and it attempts to benchmark processor-intensive tasks that might actually be found in a real Web application.
We’ve run this test on both the iPad and iPad 2 under iOS 4.3, as well as the Motorola Xoom under Android 3.0 and the Samsung Galaxy Tab under 2.2. The Android devices are both using Google’s
V8 JavaScript engine while the iPad and iPad 2 are using Apple’s
Nitro JavaScript Engine.
First, the new iPad’s hardware has resulted in a 35% improvement over the previous generation. We see that the Motorola Xoom and the iPad 2 are roughly evenly paired when it comes to this JavaScript benchmark. Despite the iPad 2’s advantage in raw processing power, the Xoom makes up for that with Google’s
superior JavaScript engine. Yet the Samsung Galaxy Tab, with its older 1GHz “Hummingbird ARM” processor, can’t keep up and is considerably slower than the competition. As Apple continues to evolve their Nitro engine, however, they should be able to get additional JavaScript performance out of the same hardware.