According to the specs, the OMAP4470 is manufactured in 45-nm technology at TI's own fabs. Its maximum rated frequency is 1.8 GHz, and the LPDDR2 memory works at 466 MHz on a 32-bit wide BUS in dual-channel mode. The chip's GPU is based on Imagination's PowerVR SGX544 architecture, and is clocked at 384 MHz and packs a dedicated 2D core.
Despite being a quad-core SoC, Nvidia's Tegra 3 reportedly can only do 8 billion floating point operations per second whereas the OMAP4470 can do 12 billion. Even more, the Nvidia chip reportedly only has a 5.3 GB/s memory bandwidth whereas the TI solution has a wider 7.5 GB/s bandwidth thanks to the dual-channel LPDDR2.
Yes I know the specifications. The question is, does it achieve it in practice? Memory controllers rarely reach their peak, at least with CPUs. I have more experience on the server side, where we typically use benchmarks like STREAM to verify how much of the peak is being reached. I don't really believe anything the vendors say until I can verify it for myself
Charbax, if you have time, can you quickly run the test above on the G10 and report results?