Since the term cloud isn’t getting much attention anymore it is time to move to the Internet of Things and Vivante is right there for it. In fact they have a solution called IoTGC 400, a full GPU in less than 1mm^2 of silicon.
The idea for the Internet of Things (IoT) is simple, put sensors and a bit of compute power everywhere and tie it in to the Internet. In theory this will allow for all sorts of wonderful things and incalculable efficiencies. In practice it is a good way to fill a fab with promise and test curious vendor battery life claims. Like cloud before it, IoT has some very solid use cases and a lot more that seem intertwined with what a vendor is trying to sell you.
All of the real IoT plays have two things in common, they are cheap and they have extremely long battery lives. ARM is the reigning champ in the is world but for many markets they are too big and expensive leaving lots of profitable room for microcontroller players under them. In general think about silicon costing a dollar or two with battery lives that can be years if necessary.
Intel with its >$5 Quark line is a few times too big and power-hungry to be a real player in IoT world. For Arduino sure, to stick on the leg of a deep-sea oil rig with a strain gauge, not so much. This market is about price, battery life, price, and battery life, not necessarily in that order. Every tenth of a mm^2 counts as does every milliwatt.
This is where Vivante comes in with their IoTGC 400 GPU. It is a full OpenGL ES 2.0 that fits in less than a square mm of silicon on TSMC’s 28HPM process. SemiAccurate went over the tech of the Vivante GPU architecture a while ago, (Parts 1, 2, 3, 4) if you aren’t familiar with it you might want to look those over. What the IoTGC 400 GPU brings to the table in that sub-1mm^2 area is the 3D GPU, the Vector unit and Composting Engine are not part of this total. Should you want the Vector Engine, that will add between .1 and .2mm^2 to the total, the Vector unit adds a whopping .2 to .47mm^2 to the tally but you could also run that as a complete GPU as well.
What do you get on the <1mm^2 GPU? As mentioned earlier a full Open GL ES 1.1/2.0 GPU with a somewhat slimmed down Shader Unit Quantum, it has a full Vec-4 shader slice with the full 32-bit precision but is limited to 32-bit math. Then again for IoT applications a 32-bit OpenGL ES 2.0 GPU is akin to sever overkill unless you need to do the occasional GPU compute number crunching session on gathered data, then it is a welcome way to save power. In this space GPUs are less about showing nifty 3D graphics as they are a way to do a job with less power than a CPU.
One thing we keep harping on about with the IoT is cost, it is two of the four most important things for an IoT device. In the $1-ish segment, storage is also a key concern. If you are talking about a $1 SoC, how much leeway do you think you have to spend on storage space? If you are thinking roughly zero, you are right on, the SoCs usually have a bit of RAM and flash for code and data storage on board and that is about it. Go look at the latest round of 100MB+ GPU drivers and think about how much that would cost to store. Sure most of that is BS and fluff but even if you cut it down, modern GPU drivers are still a tad bloated.
That is where things get interesting with the IoTGC 400 GPU, the driver footprint is really tiny. Vivante claims that the full OpenGL ES 2.0 driver stack for it is less than 100KB, more than small enough to fit in the storage available in this class of SoC. Better yet you can run it and the frame buffer for a 720p screen with than 2MB of storage. In short even if you want the full Compositing Engine and Vector Unit along with the IoTGC400, you can probably do it on an ~$1 class device. While it won’t put an R9 290X to shame, you probably can’t fit that GPU in a mm^2 or two much less run it on milliwatts.S|A
Have you signed up for our newsletter yet?
Did you know that you can access all our past subscription-only articles with a simple Student Membership for 100 USD per year? If you want in-depth analysis and exclusive exclusives, we don’t make the news, we just report it so there is no guarantee when exclusives are added to the Professional level but that’s where you’ll find the deep dive analysis.
Latest posts by Charlie Demerjian (see all)
- More on Intel’s 10nm process problems - Sep 17, 2018
- Intel puts out another 14nm 2020 server platform - Sep 11, 2018
- Why Can’t Intel Supply Enough 14nm Xeons? - Sep 10, 2018
- Intel can’t supply 14nm Xeons, HPE directly recommends AMD Epyc - Sep 7, 2018
- AMD reintroduces the Athlon name with two CPUs - Sep 6, 2018