Preliminary Tegra K1 Benchmarks Emerge, K1 Packing a KO Punch!
Posted January 13, 2014 at 09:00 pm by Will Verduzco
About a week ago, we talked about Nvidia’s Tegra K1 announcement and what it could mean for the future of mobile gaming. Then, our own XDA Developer TV Producer Jordan got a hands-on look at the chipset and its reference platform, and he took a look at some of what it can do. The chip, which merges Nvidia’s GeForce architecture with their mobile line, is based on the same Kepler architecture that powers their desktop GPUs.
Despite of all of this, one question still remained: Just how fast will this thing actually be? We all knew it was poised to be near or at the top of the pack. After all, it’s the first in a new generation of high-end mobile SoCs. And on top of that, Nvidia has made some spectacular claims by billing the K1 as having 192 cores.
Now, we all know that these 192 “cores” aren’t cores in the same respect as how your quad-core Snapdragon 800 has 4 CPU cores. In fact, they aren’t even really discrete GPU cores like the 2 GPU cores in the PowerVR SGX 543MP2. Rather, a better way of thinking of these Cuda “cores” is to consider them unified shaders, much in the way that desktop CPUs are categorized. The Tegra K1 happens to have a single Kepler SMX (Next Generation Streaming Multiprocessor) unit, and each Kepler SMX is comprised of 192 Cuda Cores. Thus, a more accurate comparison would be not to the 4 CPU cores in the Tegra 4, but rather its 72 unified shaders—or the 32 Cuda Cores per desktop Fermi SM unit. But even after dispelling some of the marketing buzzword-induced misconceptions surrounding the K1 and its 192 cores, we are still left wondering about this SoC’s performance.
Now, thanks to some benchmarks run by the folks over at TomsHardware, we have a bit of a better idea of how the K1 will perform. And as demonstrated on the Lenovo ThinkVision 28, the Tegra K1 makes the GPUs in our current generation of SoCs look like yesterday’s news.
When controlling for resolution by using an offscreen 1080p buffer in the industry standard GFXBench, the Tegra K1 manages 48 frames per second in the T-Rex HD test. This is vastly superior to what’s seen on the Nexus 5’s Snapdragon 800 (23 fps), the Tegra Note 7’s Tegra 4 (20 fps), and the iPhone 5s’s Apple A7 (27). A similar spread (but with a smaller percent difference) can be seen in Futuremark’s 3DMark benchmark.
On the CPU front, the quad-core ARM Cortex A15 in the initial version of the Tegra K1 looks to be fast, but not revolutionary. This is to be expected, as there is no reason for the four 2.0 GHz A15 cores in the K1 to be much faster than the four 2.3 GHz Krait-400 cores in the Snapdragon 800. After all, the Krait-400 CPU is essentially a highly modified ARM Cortex A9 with many performance-enhancing features found in the A15. As such, the K1-powered ThinkVision 28 bested the quad-core Nexus 5 and Tegra Note 7 by less than 10%. And strangely, the AnTuTu results for the ThinkVision 28 actually proved worse than the Tegra Note 7.
Now there are still a few caveats with these results. These devices all feature different OS versions. And as such, any direct comparisons are impossible to make. The ThinkVision is running Android 4.2, whereas the Tegra Note 7 is on 4.3 and the Nexus 5 is naturally on Google’s latest and greatest. Furthermore, the 64-bit ARMv8-based Denver version of the Tegra K1 is still yet to be seen, and that could potentially bring some major increases in CPU performance—all while being pin-compatible with the A15 version.
There’s still a lot to be seen regarding how the Tegra K1 will perform in the devices of tomorrow, but at least we know that its GPU performance does not fail to impress. What are your thoughts on the K1? Will it be at the heart of your next mobile device? Let us know in the comments below!
Want something on the XDA Portal? Send us a tip!