ARM ‘All-In’ With Mobile GPU Compute

Unless you have been living under a rock the last few years, you know that smartphones and tablets have been blowing the doors off of the tech market compared to the personal computers.  Inside of every one of these devices is what’s called an “SoC”, or system on a single chip. Each SoC has distinct blocks of intellectual property and functionality that do different tasks depending on the type of software.  These are blocks like the processor, graphics, video, camera, audio, DSP, connectivity, GPS, etc.  For years, ARM has had the dominant instruction set for processors, but companies like Qualcomm and Imagination Technologies have dominated in mobility graphics.  I had the pleasure of researching a graphics white paper (you can find here) with ARM, Samsung, Aptina Imaging and Codeplay, and wanted to share a few of the highlights with you.  Let me start with some background first.

Graphics chips, or “GPUs”, are best known as componentry that help makes games and user interfaces run better.  Since Nvidia kicked off their CUDA effort 8 years ago, the GPU is slowly becoming synonymous with “GPU compute”.  ”GPU compute” occurs when the GPU is used as a general purpose compute engine, like the CPU.  For some compute tasks, particularly parallel, the GPU is better suited than the CPU.  (DSPs and fixed function controllers are even better for some workloads, but I’ll save that for a future column.) Net-net, GPUs are getting more and more important to the overall smartphone and tablet experience.

While GPU compute is being done today on the smartphones you are using today, it only gets better in the future.  Today, GPU compute is used today primarily to make games play better and to make pictures look better by assisting with color correction, lens aberration correction, distortion compensation, and noise filtering.  In the next few years, GPU computing will get even better as standards gel, programming becomes more standard and simple, and the performance rises.   This will enable even more powerful and quicker image processing, 4K video editing and effects, super-resolution, improved voice control, security and thoroughly enhanced augmented reality, all without decreasing battery life.

My firm had the chance to do a deep-dives with ARM, Samsung, Aptina Imaging, and Codeplay in the context of researching a GPU compute paper for ARM.  What is crystal clear is that phone OEMs and ISVs are excited about GPU compute, and directionally, ARM is “all-in” on GPU compute to take advantage of the next generation of apps.  By this, I mean that their hardware and software is optimized to do GPU compute.  For example, the ARM Mali-T600 series of GPUs have Khronos OpenCL 1.1 Full Profile conformance and supportsGoogle Android’s Renderscript GPU Compute, and Microsoft DirectCompute DX9-DX11.  If you wonder where Apple fits in, well, they developed the first version of OpenCL. This pretty much covers all the mobile bases so far for GPU compute.

What does this mean to ARM and the mobile market?  It’s simple.  With ARM “leaning into” GPU compute the way that they are with their designs, this could give them a foothold in the mobile GPU market where they they had been playing “catch-up” for a few years.