Graphics Performance

While the SHIELD Android TV and the underlying Tegra X1 SoC are by no means slouches on the CPU side of matters, ultimately NVIDIA is still just another ARM Cortex implementer. On the other hand when it comes to GPUs, this is where NVIDIA truly shines. As NVIDIA is first and foremost a GPU company, and as a design philosophy always invests more time and die space into GPUs than any other SoC builder (save perhaps Apple X-class SoCs), they have always led the pack on GPU performance. And with the Tegra X1 packing a 256 CUDA core implementation of the very potent Maxwell GPU architecture, an estimated 1GHz clockspeed, and all the power and cooling it needs to keep from throttling, all of the ingredients are in place for a very strong showing from NVIDIA for GPU performance.

If there’s any real downside for NVIDIA and the SHIELD Android TV here, it’s that the Android GPU benchmarking situation hasn’t really caught up with what their GPUs are capable of. The standard benchmarks work, but game benchmarks are virtually non-existent, even in the AA and AAA games NVIDIA has helped bring over to the Android/SHIELD ecosystem. So we can’t for example look at the performance of The Talos Principle in the same way as we can the PC today.

Anyhow, we’ll start off with 3DMark’s Ice Storm Unlimited benchmark.

3DMark 1.2 Unlimited - Overall

3DMark 1.2 Unlimited - Graphics

3DMark 1.2 Unlimited - Physics

The results here pretty much speak for themselves, especially in the all-GPU graphics sub-score. SHIELD Android TV is 50% faster than the Surface Pro 3 and almost twice as fast as other devices like the iPad Air 2 and the last-generation SHIELD Tablet. Graphics workloads tend to be the most punishing from a heat and power standpoint, and hence are the most likely to get throttled. This really lets the SHIELD and its SoC open up here. At the same time, even in the physics score and the composite overall score, the SHIELD is well in the lead thanks to this combination of factors.

Up next we have BaseMark X 1.1.

BaseMark X 1.1 - Overall (High Quality)

BaseMark X 1.1 - Dunes (High Quality, Offscreen)

BaseMark X 1.1 - Hangar (High Quality, Offscreen)

BaseMark X 1.1 - Dunes (High Quality, Onscreen)

BaseMark X 1.1 - Hangar (High Quality, Onscreen)

Once more it’s a clean sweep for the SHIELD. Apple’s iPad Air 2 tends to be the second-place finisher, thanks to Apple’s own significant investment in GPU resources, the SHIELD gets yet more resources and the power to fully exploit them. With all of that said, the one thing that does surprise me a bit here is that even the SHIELD can’t crack 60fps on Dunes and Hangar; BaseMark X will likely take one more generation before its tests can be completely beaten.

Our final GPU benchmark is GFXBench.

GFXBench 3.0 Manhattan (Onscreen)

GFXBench 3.0 Manhattan (Offscreen)

GFXBench 3.0 T-Rex HD (Onscreen)

GFXBench 3.0 T-Rex HD (Offscreen)

Starting first with the benchmark’s high-level tests, we once again find the SHIELD well in the lead. Of particular note here, the SHIELD Android TV becomes the first Android device to break 60fps on Manhattan, something that as recently as a generation ago seemed impossibly far away.

GFXBench 3.0 ALU Test (Onscreen)

GFXBench 3.0 ALU Test (Offscreen)

GFXBench 3.0 Alpha Blending Test (Onscreen)

GFXBench 3.0 Alpha Blending Test (Offscreen)

The low-level tests meanwhile offer us an interesting look at SHIELD and Tegra X1 that the high-level tests don’t provide. The ALU test for example drives home the point of just how much in the way of shading resources Tegra X1 has at its disposal (at least when unrestricted), and is a big reason why the SHIELD is doing so well here. On the other hand while NVIDIA still takes the top spot in alpha blending, the lead over the iPad Air 2 isn’t nearly as great, thanks in large part to the iPad’s relatively large 128-bit memory bus. Ultimately SHIELD doesn’t struggle here, but it’s an interesting point of comparison since it shows one of the only cases where SHIELD isn’t a run-away winner, and what kind of graphics workload may eat into its otherwise ridiculous advantage.

GFXBench 3.0 Driver Overhead Test (Onscreen)

GFXBench 3.0 Driver Overhead Test (Offscreen)

GFXBench 3.0 Fill Rate Test (Onscreen)

GFXBench 3.0 Fill Rate Test (Offscreen)

On the last batch of subtests, I’m going to stop again with the fill rate test just to point out another interesting design decision for Tegra X1 and the SHIELD. With NVIDIA banking so much on the set top box’s ability to drive 4Kp60 TVs, NVIDIA needs to push a lot of pixels to get there, as 4K is four-times as many pixels as 1080p. For this reason Tegra X1 is paired with 16 ROPs, giving it low-end PC desktop-like pixel throughput, and of course a huge advantage in the GFXBench fill rate test. And though this fill rate will help with games as well, it’s keeping up with the Android UI at 4Kp60 that is the most important reason for all of this fill rate performance.

Wrapping things up, what else is there to say about the SHIELD Android TV’s GPU performance? What we’re looking at here is twice as fast (or more) than the best phones and tablets today, a combination of the Tegra X1’s very strong GPU design and the set top’s practically unlimited energy and cooling capabilities. Overall NVIDIA is following the path they started long ago on Tegra, always favoring strong GPU performance, leading to the incredible performance we’re seeing here.

With that said, given NVIDIA’s decision to focus on the Android TV aspects of the SHIELD Android TV first and foremost – and not gaming – it’s also fair to say that the SHIELD is overpowered for Android TV work. Other than keeping up with the basic pixel fill needs of 4K, the real power of the box’s GPU is going to be left untapped by Android TV (especially on the ALU side). Which makes all of this humorously absurd in a way, but it’s also why SHIELD Android TV isn’t just an Android TV box. The device’s GPU performance will be put to good use with gaming, though stepping outside the world of benchmarks for a second, the bigger challenge NVIDIA faces is not delivering high performance, it’s delivering games that make full use of that performance.

System Performance The Android TV Experience
Comments Locked


View All Comments

  • Ryan Smith - Thursday, May 28, 2015 - link

    It's a matter of time in covering everything, along with the fact that GameStream is not a new technology in the ecosystem. Once the commercial GRID service launches, we'll be taking a more focused look at gaming on the SHIELD.
  • Sonicadvance1 - Thursday, May 28, 2015 - link

    Hello. Dolphin emulator developer here.
    I've had an Nvidia SHIELD Android TV in my hands for a while now.
    I'd like to clarify in my comment the CPU and GPU speeds on this device
    The Cortex-A57s max clock speed is 2.01Ghz, and the Maxwell GPU's max clock speed is 998Mhz.
    I haven't checked the Cortex-A53's max clock speeds because being a Dolphin developer I effectively don't care about them.
  • jjj - Thursday, May 28, 2015 - link

    Got to wonder if it wouldn't be more appealing as a Linux PC or even a Chromebox. In those segments the pricing wouldn't be as unreasonable.
  • funtasticguy - Friday, May 29, 2015 - link

    Sonicadvance1, how does the Dolphin emulator work on this? What kind of speed improvement does this offer compared to the Tegra K1?
  • Sonicadvance1 - Friday, May 29, 2015 - link

    It runs the new AArch64(ARMv8) JIT recompiler for the CPU core, which lets it run a few games full speed. This new core will take a long while to become fully optimized, but I'm taking steps to make it easier to profile the code's performance from Android which will help in the future.
    This new CPU core is going to be more maintainable than the old 32bit ARMv7 CPU core.
  • funtasticguy - Friday, May 29, 2015 - link

    Thanks for your reply!
  • Wardrive2015 - Friday, May 29, 2015 - link

    Thank you Sonicadvance1 for the clarification on the clocks. Would you say single threaded performance is roughly equal to that of Denver?
  • Sonicadvance1 - Friday, May 29, 2015 - link

    It's about on par, slightly better in some cases.
    Due to the Denver's quirky architecture the A57s will be quicker than the Denver in code that only executes a few times. This is due to the Denver only recompiling code to native VLIW if it has been called a certain number of times(This number is undisclosed).

    For code that executes over and over the two CPUs are about on par with each ones winning depending on the situation.
  • Haydon987 - Thursday, May 28, 2015 - link

    The big question is, can it do HDMI 2.0 with HDCP 2.2 at 4:4:4 all at the same time? Last I heard, no HDCP decoding chip currently in existence does this.

    They all downgrade to 4:2:0 when playing protected content and can only do 4:4:4 when playing unprotected content, which is about nothing other than gaming.
  • Haydon987 - Thursday, May 28, 2015 - link

    At 4K, I should add.

Log in

Don't have an account? Sign up now