Tegra 4i: A Chip Off The Old Block
Tegra 4 is being groomed for the tablet and superphone space, wielding 4+1 Cortex-A15 CPU cores, the full 72-shader GPU, and an optional i500 soft modem. Tegra 4i, on the other hand, is Nvidia’s smartphone-oriented solution, armed with 4+1 Cortex-A9 r4 cores, a 60-shader GPU, and the i500, integrated.
Although Tegra 4i’s CPU component is clearly different, its GPU is based on the same architecture as Tegra 4. Instead of six vertex units, you get three, totaling 12 MADs. Meanwhile, four pixel pipes narrow to two. Notice that the same number of MADs exist in the fragment pipe, though. Nvidia made a very conscious decision, based on today’s content, to emphasize complex shading rather than fill rate. Some of this has to do with the effects developers can achieve by leveraging shader horsepower.
However, as a phone-oriented device, Tegra 4i needs to make efficiency its priority, and Nvidia picked its ratios with a single-channel 32-bit memory interface in mind. Why just one channel, rather than two? First, a dual-channel design would have been hard to fit into Tegra 4i’s 12x12 mm package. Two channels would have adversely affected baseline power consumption as well. So, implementing one channel and then populating it with DDR3 running at up to 2,133 MT/s as much as doubles Tegra 3’s bandwidth without hurting idle efficiency.
It also helps to consider the devices these SoCs are designed for. Tablets like the Nexus 10 are pushing 2560x1600 resolutions, while some of the latest super-phones sport 1080p 5” displays. At two pixels per clock, a 660 MHz Tegra 4i does more than 1.3 Gpix/s. A 1080p phone with just over two million pixels, refreshing at 60 Hz, needs fewer than 125 million pixels per second. Tegra 4i gives you enough fill horsepower for more than 10x overdraw, filling every pixel on the screen. Hence, Nvidia thinks it effectively balanced resources given the platform it’s targeting with Tegra 4i.