Yesterday at their event, Amazon’s CEO, Jeff Bezos, made what seemed like a surprising comment about the OMAP 4470 chip, which is powering the new Kindle Fire HD 8.9″ (the new Kindle Fire and Kindle Fire HD 7″ have OMAP 4430 and OMAP 4460 respectively). Amazon said that OMAP 4470 can do 50% more floating-point operations per second than the Tegra 3, while the dual-channel LPDDR2-466 RAM on the Fire HD has 40% more bandwidth than what Tegra 3 comes paired with. So is it true?
First, we need to understand where the OMAP 4470 is coming from. Initially, OMAP 4470 was supposed to be a dual-core 1.8 GHz processor, while the OMAP 4460 was supposed to hit 1.5 GHz. It seems TI failed in both missions — the OMAP 4460 arrived at 1.2 GHz per core, on the planned timeline; the OMAP 4470 arrived significantly later than expected (spring 2012) and clocked at only 1.5 GHz per core.
The OMAP 4470 uses a PowerVR SGX MP1 GPU, although it’s important to note that it’s running at a 384 MHz, compared to the PowerVR SGX MP2 in Apple’s A5 chip, which has 2 cores, both running at just 250 MHz. This is usually TI’s strategy with their chips — they use an older core, but they raise the clock speeds. I don’t like this approach, because it means TI will always be behind others when it comes to new GPU features, like OpenGL ES 3.0 or OpenCL. Moreover, they won’t even be able to get these new features on the upcoming OMAP 5 chips, because they will use an overclocked PowerVR SGX543 MP2, at a time when everyone would have moved already to a the next-gen architecture.
However, in terms of raw performance, TI’s chips should be pretty competitive, and the OMAP 4470’s GPU should be almost as fast as Apple’s A5 GPU. As you can see in these benchmarks from Anandtech, the GPU in the Apple A5 is 30%-80% faster than the Tegra 3, depending on the workload.
It’s possible Texas Instruments also made further optimizations to the SoC to get even closer to that PowerVR SGX5 43MP2 performance, considering they’ve made their overclocked PowerVR SGX543 MP2 exceed the performance of the PowerVR SGX543 MP4 by 5% in their upcoming OMAP 5. This means the above charts should give an approximate GPU performance difference between OMAP 4470 and Tegra 3.
As for the memory bandwidth, one of the biggest drawbacks of the Tegra 3 is that it still doesn’t support dual-channel memory, which means memory bandwidth is a bottleneck in some cases. OMAP 4470 reduces this problem by using dual-channel along their processors.
So, while we don’t have any benchmarks of these chips and devices yet, Jeff Bezos’ comments on Tegra 3 vs. OMAP 4470 are likely true. Still, it’s important to remember that, although the GPU has become very important lately, especially as it’s needed to run these high resolution displays, the CPU is still the most important part of the SoC, as it helps you run not just graphics related elements, but pretty much everything. With that said, Tegra 3 is still a quad-core CPU vs the OMAP 4470 which is dual-core.
Also, Tegra 3 is quite old already, and Tegra 3+ is about to ship soon.