TI's OMAP 4430: CPU And GPU Performance
CPU Horsepower: Closer to A5 Than Tegra 2
As we’ve mentioned in the past, mobile devices like smartphones and tablets use highly integrated logic referred to as SoCs to minimize physical footprints and power consumption, all the while optimizing for performance and functionality. By putting execution resource, graphics processing, system memory, and several other subsystems in a single ASIC, data transfers can be achieved more efficiently without soldering a bunch of separate chips onto a PCB. Incidentally, the same SoC term applies to modern desktop processors like Intel's Sandy Bridge- and AMD's Bulldozer-based chips, which combine cores, cache, memory controllers, and other capabilities.
The Kindle Fire uses TI's OMAP (Open Multimedia Application Platform) 4430 SoC. We haven't had as much experience with this specific piece of hardware, but the underlying technology is similar to the more familiar Tegra 2 and A5 in that all three employ dual-core Cortex-A9s clocked at 1 GHz.
| SoC | Apple A4 | Apple A5 | Nvidia Tegra 2 | TI OMAP 4430 |
|---|---|---|---|---|
| Processor | 1 GHz ARM Cortex-A8 (single-core) | 1 GHz ARM Cortex-A9 (dual-core) | 1 GHz ARM Cortex-A9 (dual-core) | 1 GHz ARM Cortex-A9 (dual-core) |
| Memory | 256 MB 333 MHz LP-DDR (single-channel) | 512 MB 1066 MHz LP-DDR2 (dual-channel) | 1 GB 667 MHz LP-DDR2 (single-channel) | 512 MB 400 MHz LP-DDR (dual-channel) |
| Graphics | PowerVR SGX535 (single-core) | PowerVR SGX545MP2 (dual-core) | ULP GeForce (single-core) | PowerVR SGX540(single-core) |
| L1 Cache(Instruction/Data) | 32 KB / 32 KB | 32 KB / 32 KB | 32 KB / 32 KB | 32 KB / 32 KB |
| L2 Cache | 640 KB | 1 MB | 1 MB | 1 MB |
Given the common CPU architecture, it shouldn't be a surprise that Linpack generates the same score on Amazon's Kindle Fire as it does on our Tegra 2-based Xoom.
Even so, the OMAP 4430 has more in common with Apple's A5 than it does with the Tegra 2. Why is this? Apple and TI both use A9s with an ARM SIMD engine, better known as NEON. It's capable of decoding MP3s on an ARM CPU running as slow as 10 MHz, which provides added benefits in power savings. Why don't all SoCs feature the Advanced SIMD extension? Architecturally, it imposes a physically larger and more costly SoC.
GPU: Better Than A4, Worse Than A5
On the graphics side, the OMAP 4430 employs Imagination Technology's PowerVR SGX 540. This should be familiar because it's derived from the same architecture as the GPUs found in Apple's A4 and A5. The key difference is in the shaders.
| GPU Subsystem(System-on-Chip) | PowerVR SGX 535(Apple A4) | PowerVR SGX 540(OMAP 4430) | PowerVR SGX 543(Apple A5) |
|---|---|---|---|
| SIMD | USSE | USSE | USSE2 |
| Pipelines | 2 | 4 | 4 |
| TMUs | 2 | 2 | 2 |
| Bus Width (in bits) | 64 | 64 | 64 |
| Triangle rate @ 200 MHz | 14 MTriangles/s | 28 MTriangles/s | 35 MTriangles/s |
The SGX 543 in the A5 includes four USSE2 (Universal Scalable Shader Engine 2.0) pipes. In comparison, the SGX 540 has the same number of pipes but uses the older USSE design. The SGX 535 hails from the same GPU generation as the SGX 540, but it only has two USSE pipes.
While the architecture is interesting, performance data is what we're after, which is why we're turning to GLBenchmark.
The OMAP 4430 falls expectedly between the A4 and A5. Though it's probably shocking to see the SGX 540 outperform Nvidia's ULP GeForce by such a large margin, remember that GLBenchmark tests at a device's native resolution. On our Tegra 2-based tablets, that's always 1280x800. Meanwhile, the Kindle Fire only has to do its work at 1024x600.
We can get normalize for that resolution issue by looking at the off-screen tests set to 720p. Based on those results, we see the SGX 540 core performing similarly as Tegra 2's ULP GeForce.
