Tegra 3+ review: Not good enough

NVIDIA Tegra 3 is finally starting to show its age, with competitors catching up with them as much on the CPU side as on the GPU side of things, most notably, with Qualcomm's Quad-core Krait in the Snapdragon S4 Pro, and the Exynos 4412. Tegra now doesn't look as good as it did in early 2012. And since the next-gen Tegra 'Wayne' is scheduled for release in Q1 2013, NVIDIA needed to refresh its Tegra lineup before Wayne comes up. That is the purpose of Tegra 3+, or T37. T37 supposedly has a CPU that can run at up to 1.7GHz and a faster GPU. The thing is, the architecture of the CPU/GPU remains unchanged. The CPU should be good enough too keep ahead of the competition, but the GPU still has many issues. It still uses an Immediate Mode Renderer with Early-z rejection, which is not a very good renderer for a mobile platform, and still uses the eight-year-old core architecture, as opposed to the unified architecture that all other mobile GPU vendors, most recently ARM's Mali GPUs, have adopted.

In my opinion, Tegra 3+ didn't even get a very significant bump. It would be much better if NVIDIA had forgotten about T37 and concentrated on getting Tegra Wayne ready faster. The first device shipping with T37 will be the HTC One X+, which is a refresh of HTC's successful One series. The One X+ will have a 5"  1280 x 720 display (294 pixel density) and Corning Gorilla Glass 2, and will ship with Android 4.1 Jelly Bean. The only benchmark we have of it so far is NenaMark2:

Tegra 3+'s graphics performance there is...ok at most.

Of course, most of the game-changing upgrades to Tegra will be made in the next-gen Wayne. Rumors are that it'll be built on 28nm process, and will thus eliminate the heat problems found in Tegra 3, and it will have four Cortex-A15 cores @ up to 2GHz, plus a battery saver core, which will put it's performance ahead of the upcoming Qualcomm Snapdragon S4 Pro's Quad-core Krait. The GPU core is also getting quite a bit of an upgrade: Rumors are the graphics core of Wayne will be based upon NVIDIA's Kepler GPU architecture, which offers the best performance with the least power consumption, which is perfect for mobile devices, which have to keep substantially long battery lives. Rumors are that there'll be 24 CUDA (GPGPU) enabled cores, which will support DirectX 11.1, OpenGL 4.x, and PhysX, and it'll probably support OpenGL ES 3.0. The jump from eight-year-old architecture cores to CUDA-enabled cores will be a huge jump. According to NVIDIA's roadmap for Tegra, Wayne should be 10x faster than Tegra 2, and 3x faster than Tegra 3. Given that Tegra 2 has the raw processing power of 4.8 GFLOPS, Wayne should have, theoretically speaking, 48 GFLOPS of performance, which will put it far ahead of the Adreno 320 and Apple's most powerful A5X's PowerVR SGX543MP4. The SGX543MP4 can do 32 GFLOPS, therefore, Wayne will be expected to give 25% more performance than the Apple A5X. If these rumors are all true, then this means that NVIDIA has finally began to take Tegra GPUs seriously, a good news for Google and Microsoft.