Nvidia Tegra K1 Futuremark Icestorm Unlimited Benchmarks Smoke Apple’s A8X SoC
Just a few hours ago, the initial benchmark comparison of the Nvidia Tegra K1 SoC and Apple's new flagship, the A8X appeared. The benchmarks put the Tegra K1 as the on-screen winner while the the A8X took home the crown when it came to actual GPU Power. Interestingly however, benchmarks of the Apple iPad Air 2 have been uploaded on Futuremark's database and it would appear that it took quite a thrashing from the K1.
Nvidia's Tegra K1 still holds the crown in gaming benchmarks
It goes without saying that it is pretty much impossible to fake benchmarks on Futuremarks website. You might remember the multiple instances on which it has weeded out and even banned entire devices because of allegations of tampering. You can wait for more benchmarks to appear before making a conclusion, but these appear to be pretty conclusive. The K1 still holds the crown as far as gaming performance is concerned. So why then did the Tegra K1 loose in the GFXbench score posted some while back. Was that fake? No, it doesn't have to be.
The answer lies I think in the K1's unique design. One explanation would be that the GFXbench score was not able to correctly utilize the 192 cores of the K1 while Futuremark's Ice Storm has been bred and engineered towards utilizing GPU cores. Where the Nvidia Shield Tablet scores an impressive 30, 759 aggregate score the Apple iPad Air 2 only manages 21, 604 points. Notice however, that this is actually the aggregate or average score. The actual graphic score is 31, 396 for the A8X PowerVR GPU and 36, 109 for the Tegra K1 Kepler GPU. Ofcourse you have to keep in mind that we are comparing a 20nm SoC with a 28nm SoC and considering that this is pretty impressive. Also, this is the Cortex A15, 4+1 (32 bit) variant, which is weaker on the CPU side.
The actual area in which the iPad Air 2 takes a beating is in the Physics department where it scores 10, 337 points compared to the Tegra K1's 20, 308. That is almost double the performance on the K1. Here I believe, the benchmark is utilizing the GPGPU capabilities of the K1, which should be pretty much non-existent on the A8X. I speculate this must be the case because the variant in question is not the powerful Denver variant and should actually score lower than the A8X unless the 192 cores are being leveraged in some way. Ofcourse with UE4 and AEP on the scene we already know the Tegra K1 is heavily optimized for gaming but this should shed light on just how much difference there is between the two. The final verdict will have to have to wait till the Google Nexus 9 arrives and the Denver variant is benchmarked, but I have a feeling we already know which one will go home with the win in terms of gaming performance.