Imagination Technologies is a leader in mobile graphics for a reason. Though NVIDIA is most often noticed and talked about, it’s the PowerVR architecture that powers some of the best selling mobile devices. For the past seven years since the release of their Rogue architecture they’ve been focusing on expanding into many different markets, bringing about improvements in efficiency to allow them to break into new markets. Furian is nearly a new GPU entirely that’ll focus purely, at first, on the high-end spectrum.
Furian is a refocusing of Imagination Technologies on the top-end, something that has been attractive to Apple and has led to them licensing Imaginations designs in the past, using their own technologies to provide great gaming performance within a very tight and low power window. Furian builds upon that with a drastic increase in performance over Rogue. The top-end Furian design will use less power and much greater gaming performance than the 7XT, already a fine performer and featured in the iPad Pro.
The huge leap in performance is due to a transition to a 7nm process and some architectural design decisions that help to speed up the graphics and compute pipeline. The ALU is more dense and capable of asynchronous and parallel execution to help efficiency.
Inside each ALU, there are now 32 pipelines, widening it from the 16 featured in Rogue. The ALU’s have also become a bit more simplified to help with efficiency. Instead of having two multiply + add components, they’ve dropped the second for a simple multiply ALU, which are more simple but less flexible and technically capable than a MAD. Essentially, this means that it should be quick for parallel workloads and the added texture unit performance increase, from 4 samples-per-clock to 8 samples-per-clock should mean a sizable the amount of work that can be done in one cycle.
What’s more, the Furian can handle memory,, from both system and dedicated to it, in a more efficient fashion with multithreaded access.
One of the hallmarks of Rogue was how it could scale to more clusters put together on a coherent fabric. Furian, too, will have that capability and will theoretically be able to go beyond the 16 clusters that are currently the theoretical limit of Rogue.
Being a higher performing part, Imagination sees Furian entering many different markets and even competing against NVIDIA in terms of embedded DNN’s and AI markets. That means we could potentially see partnerships with automobile manufacturers and perhaps even large-scale IoT manufacturers to include PowerVR Furian as a means to provide smart things.
Of course, Furian is OpenGL ES 3.0 capable, compatible and capable of running the latest iteration of Vulkan and OpenCL 2.0 as well as OpenVX. This is a GPU that Imagination will likely use to tackle the mobile and embedded GPGPU market, which is growing surprisingly well.
So when will we see the Furian inside any devices? One given is that Apple will probably adopt the design for one of their devices in the near future, though we’re likely only start appearing around the end of 2018.