Phill Smith, Demo Manager at ARM, has filmed and uploaded four very interesting demos of what new features will be possible thanks to new generation ARM Mali-450 and Mali-T6xx GPUs including 4K 3D user interfaces and games, ASTC texture compression, and OpenCL accelerated gesture recognition and HEVC / H.265 video decoding.
4K Resolution 3D User Interface and Game Demo
The first demo showcases a Geniatech box (ATV1800?) powered by AMLogic AML8726-M8 featuring an ARM Mali-450MP6 GPU running Android with a user 4K 3D interface designed by Autodesk using Scaleform UI. The rest of the video shows Timbuku 3D gaming demo running at 3840×2160 (4K2K) @ 24 fps. The frame rate appears to be low, but that’s because the box is using HDMI 1.4, which limits UHD output to 24fps. 2160p60 is only available via HDMI 2.0.
ASTC Compression Demo on Samsung Galaxy Note 3
3D Textures are getting bigger and increased size requires extra bandwidth, as well as more energy. In order to reduce the size of textures and power consumption, textures are normally compressed by the GPU. The demo below compares compression and energy consumption of raw textures, ETC2 compressed textures and the new ASTC compressions. The hardware is a Samsung Galaxy Note 3 with Exynos 5 Octa 5420 using a Mali T-628 GPU.
- Raw textures – Size: 263 MB. Power consumption – DDR3: 2.732 mJ/frame, DDR2: 3.32 mJ/frame
- ETC2 compression – Size: 93 MB. Power consumption – DDR3: 1.387 mJ/frame, DDR2: 1.687 mJ/frame
- ATSC compression – Size: 49 MB. Power consumption – DDR3: 1.101 mJ/frame, DDR2: 1.330 mJ/frame
In this demo, ATSC almost doubles the compression ratio, and consumes about 20% less energy. There’s also less differences from the raw texture with ATSC compared to ETC2.
Gesture Recognition with XBMC
The next demo show OpenCL accelerated Eyesight gesture recognition in Arndale board powered by Exynos 5250 SoC with ARM Mali-T604. The board runs Linux and is connected to a standard (and crappy) Logitech webcam. It can follow hand gestures with OpenGL acceleration, something that is not possible with the dual core Cortex A15 CPU only. They’ve also integrated the demo with XBMC, and showed how to navigate XBMC user interface with your hand only, no remote needed.
HEVC / H.265 Video Decoding
The last demo is also running Linux on Arndale board. They show H.265 decode with the GPU thanks to Ittiam implementation.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress
Oh I want gesture recognition with XBMC, will they submit that code upstream?
Is it known for sure if HEVC / H.265 video decoding is done on VPU or on CPU and GPU?
@Harley
The demo above is done with CPU + Mali-T604 on Arndale board, but some upcoming SoCs will have support in the VPU. In the video the guy explains the standard is not finalized (I thought it was now), and it may be possible some videos won’t play with hardware decoder if a change is made to the specs.