Maximizing Performance Per Millimeter Squared
Nvidia clearly needed to make difficult decisions in order to enable Tegra 4’s GPU in just 10.5 square millimeters of die space—less than Qualcomm’s Adreno 320, ARM’s Mali-T604, or Imagination Technologies’ PowerVR SGX554MP4. Holding off on a unified shader architecture, forgoing features that would have met OpenGL ES 3.0/Direct3D 9_3 minimum specifications, and paring back Tegra 4i’s features show that the company kept its eye on a number of tough targets.
Ultimately, though, we’re not looking at PC-equivalent content on mobile platforms yet. Nvidia’s leadership says the company approached Tegra 4 with rendering efficiency in mind, and the in-house benchmarks we’ve seen tend to back this up. Instead of focusing on frame rates or expensive features, the numbers compare GPU area and power consumption.
| Header Cell - Column 0 | Tegra 3 | Tegra 4 | Tegra 4/Tegra 3 |
|---|---|---|---|
| GLBenchmark 2.5 Frame Rate720p, On-Screen | 57 FPS | 60 FPS | Row 0 - Cell 3 |
| Power (mW) | 2,294 mW | 1,521 mW | Row 1 - Cell 3 |
| Performance/Watt | 24.9 | 39.4 | 1.6x |
| GLBenchmark 2.5 Frame Rate1080p, Off-Screen | 12 FPS | 54 FPS | Row 3 - Cell 3 |
| Power (mW) | 1,629 mW | 3,674 mW | Row 4 - Cell 3 |
| Performance/Watt | 7.4 | 14.7 | 2x |
Source: Nvidia
Here’s one example: in GLBenchmark 2.5’s on-screen 720p test, Tegra 3 must go all-out to achieve 57 FPS, using 2.3 W in the process. Tegra 4 pegs the display’s refresh at 60 Hz. But because it doesn’t get taxed by that workload, it only draws 1.5 W. The result is 1.6x Tegra 3’s performance (in FPS) per watt. Conversely, if you let Tegra 4 run unconstrained by rendering out to an external 1920x1080 display, it’s able to do 4.5x Tegra 3’s frame rate using 2.25x as much power. That neatly works out to 2x the performance per watt.
Granted, those are Nvidia’s internal numbers, and they ultimately only cover one of the SoC’s subsystems. However, we already know from ARM Vs. x86: The Secret Behind Intel Atom's Efficiency that power is the front on which other vendors are most eager to do battle with Nvidia. A Tegra 4 able to deliver better battery life in the next generation of Android-based games would certainly put Qualcomm and Samsung on notice. We’ll have to see how much success Nvidia has convincing Microsoft to allow some of Tegra 4’s more advanced features under Windows RT, though.
