Consider the gauntlet thrown down by NVIDIA. The company has provided additional information about its upcoming Tegra 4 and Tegra 4i mobile platforms at the Mobile World Congress conference currently taking place in Barcelona, including new performance benchmarks and architectural details. The well-timed platform upgrade comes at a time when Tegra 3, with its older GPU technology, is starting to lag behind the best the competition has to offer, which includes SoCs from Apple and Qualcomm. Tegra 4 is an attempt to wrestle back the performance crown in a big way.
Before we examine some of NVIDIA's performance claims and in-house benchmarks, let's have a look at the architecture. There are two versions of NVIDIA's next generation Tegra 4. One of those is Tegra 4i (codenamed "Grey"), which is NVIDIA's first Tegra SoC with integrated LTE processor. The Tegra 4i derivative integrates an optimized version of NVIDIA's software defined i500 radio modem technology. It not only supports HSPA+ and LTE, but is programmable to support next generation interfaces and standards as well. It has a smaller die size, consumes less power, and is more flexible in terms of implementation compared to competing fixed LTE modems, and since it's software-configurable, it's possible for carriers and OEMs to update the modem features as needed. Tegra 4i is the chip that will show up in entry-level and mid-range smarphontes.
For the so-called "superphone" and tablet categories, Tegra 4 (codenamed "Wayne) is a powerhouse platform based on ARM's Cortex A-15 architecture (Tegra 4i is based on Cortex A9 r4).
Like its predecessor, Tegra 4 sports a 4+1 CPU core design, consisting of four CPU cores for intensive tasks like gaming and video decoding, plus a fifth battery saver core. If all you're doing is surfing the web or fielding phone calls, the battery saver core takes over, allowing NVIDIA to offer the best of both worlds (performance when needed and optimal battery life).
Tegra 4 is also outfitted with 72 custom GPU cores (Tegra 4i boasts 60 GPU cores) and can be paired to up to 4GB of DDR3L or LPDDR3 memory (versus 2GB in Tegra 4i). It's not just the number of cores that gives Tegra 4 its pixel-pushing power, however. Tegra 4 features on-chip vertex, texture, and pixels caches to reduce off-chip memory accesses, resulting in better performance and, to an extent, improved power efficiency.
There are 24 vertex cores in Tegra 4, which is six times more than Tegra 3. What's more, NVIDIA claims that architectural improvements have resulted in a performance increase of up to 1.5x over Tegra 3 per vertex pipe when running at the same clockspeed. And here's the icing on the cake -- not only is Tegra 4 significantly faster, but images should look better too, as both Tegra 4 and 4i now support 2x and 4x Multisample Antialiasing (MSAA), NVIDIA says.
Outside of raw performance, NVIDIA is also promising much longer battery life with Tegra 4 compared to the competition. The slide above highlights how a Tegra 4-based phone would compare to the Droid DNA. A Tegra 4 device would be able to play videos hours longer than the Droid DNA, as well as play music and other low-level tasks.
This is in large part due to the 4-Plus-1 design touched on above. Not only does Tegra 4 benefit from having a fifth battery saver core to handle low-level tasks, but the fifth core is built using transistors that require less power to operate. It also has its own low-power 512KB L2 cache, allowing the device to flush the 2MB L2 cache associated with the other four cores, resulting in additional power savings.
You should start to see Tegra 4 devices enter the marketplace in the coming months. It's unlikely NVIDIA will build its own phone, but it is providing third-party partners with a Phoenix reference phone platform. Phoenix is a development device for OEMs built around NVIDIA's Tegra 4i platform, and on paper, it's a killer handset. The device is thin (just 8mm), has a 5-inch Full HD 1080p display, and sports the integrated 4G LTE radio.
Tegra 4 looks impressive, no doubt, we just have to wait and see what OEMs do with it.