in Techspeak

Does not GPU compute.

The semiconductor industry has seen its fair share of debates. The latest one revolves around the internal architecture of Google’s Tensor Processing Units (TPUs) for machine learning; it even sparked accusations of click bait on social media and generated plenty of speculation among analysts.

The discussion reminded me of one of my personal favorite (non-)topics: the relevance of FP64 for mobile compute. Let’s rewind to four years ago and analyze what happened: the year is 2012 and a number of mobile silicon vendors start to add FP64 support in GPU ALUs. Industry commentators are quick to label the move as the best thing in embedded graphics since the arrival of OpenGL ES.

Around the same time, 64-bit computing becomes a hot topic in mobile. Marketing folks take the opportunity to employ some (deliberately) confusing terminology and start linking 64-bit CPUs and FP64-capable GPUs. Makes sense, right?

The cold, hard reality is that 64-bit CPUs had little (i.e. absolutely nothing) to do with FP64 support. Somehow they got lumped together because everyone scoffed at the mere sight of anything 32-related.

computer-vision-machine-learning-HDRFP64 for HDR photography? No.

Anyway, once the initial dust settled and some of the confusion disappeared, the second wave of marketing hit. This time FP64 was positioned as a vital requirement for next-generation compute applications which led to a lot of huffing and puffing about how the mobile industry is going all out on GPU compute. Articles went on forever about how we will soon see data center workloads running on mobile devices.

Fast forward to 2016 and every single use case for GPU compute is still revolved around maximum power efficiency. Can you guess what that implies? Mainly finding the optimal balance between FP16 and FP32 support. Apart from a few developers still stuck in desktop-only mode, nobody is crying out for FP64 ALUs in GPUs. Instead, the industry is moving in the opposite direction by increasing the number of FP16 cores. For example, PowerVR Series7 GPUs have doubled FP16 performance over the previous generation while NVIDIA made a huge deal about the introduction of half-precision support in Tegra X1.

computer-vision-machine-learning-ARFP64 for AR? Not really, no.

Everyone now agrees that FP64 is too power hungry for mobile while delivering absolutely no performance benefit in typical applications (image convolution, augmented reality, software-based video decoding, etc.). However, FP64 precision is indeed a requirement for a very limited set of scenarios. For example, the microserver market defines a set of high-precision workloads that can be offloaded from the main CPU across on-chip accelerators (including GPUs) for improved efficiency.

Since I don’t plan to run any DNA analysis or fintech simulations on my smartphone anytime soon, I am very satisfied having FP32/FP16 precision in mobile right now. And so should you.

  • LDM

    Hi Alex,

    What you are saying makes a lot of sense. But unfortunately the problem of many companies nowadays it’s the implementation of their products. The competition became so stiff that some of the competitors you have mentioned disappeared from mobile market (Nvidia) a part some tablet implementation.

    At this stage I am not even sure debating about FP32/16 in the devices is relevant when the company you are working with, and promoting, is struggling big time to get winning products around (at least as much as in the past).

    • I think it’s important to look back and understand how various trends influenced the market and, implicitly, the devices we use on a daily basis.

      We are at a point where several new computing paradigms are appearing (e.g. VR) and it’s critical that the underlying technologies are well understood. I feel this was not the case with FP64 and 64-bit CPUs.

      I also believe the topic I wrote about (precision math for computing algorithms) is actually very relevant today, particularly in vision applications that extend beyond mobile (e.g. automotive or robotics).

      Finally, when it comes to Imagination, I wouldn’t be too quick to count them out.

      Regards,
      Alex.

      • LDM

        Sure Alex, mine is just a modest opinion from a customer that looks at current mobile situation. And he notes that previously Imagination was a real leader in this sector, but for various reason other companies took over (ARM – Mali for example). However it would be interesting having a discussion about your thoughts thru some other channels and seeing if mine are only unfounded views.

        Talking about VR, 4K etc, one of your main competitor has just announced a new GPU solution Mali G71; in paper it seems very interesting. Do you guys have or are you going to announce something similar? I recall, in one of your articles, you mentioned PowerVr GR6500 for this sort of market. Is that still the future solution for VR, 4K etc?

        L

        • Thanks, I never said that your opinion did not matter nor did I pretend I was somehow the absolute judge of computing. Your comments are absolutely fair but I still believe PowerVR and MIPS will continue to be successful in the long run.

          When it comes to VR/AR, we’re currently promoting our Rogue Series7 and Wizard GPUs. We are of course working on next-gen GPUs (e.g. Series8XT) and will announce them in due time.

  • Hi Alex,

    Thanks for a good article.

    Moreover in the 64bit GPU fight Apple did it because they wanted a different dimension in the market and notably they used the 64bit support to manage memory more optimally which proved successful to them as a result a much faster IOS compared to former v5.0 give there are more complex process running in background more efficiently than before.

    • Hi,

      I think you are talking about 64-bit CPUs? I am not disputing the merits of 64-bit CPU architectures in mobile devices (or elsewhere).

      Instead, I am talking about FP64 precision for mobile GPUs and how I believe it is a waste of silicon and power – at least for the types of algorithms that a mobile GPU is running today and in the near future.

      Regards,
      Alex.

  • Joseph Hall

    Hi, just a quick comment to thank you for providing an interesting read on a break from work. Cheers 🙂

    • Thank you for the feedback!

      Are your overlords aware that you’re reading such questionable material during work hours?

      • Joseph Hall

        I am the sole sysadmin 🙂