PowerVR Series6 Wizard GPU Is The First To Combine 3D And Ray Tracing Hardware In A Single Chip |
Imagination Technologies has unveiled a new version of its Rogue family of Series6 GPUs — and this upcoming flavor may be the first consumer hardware to integrate rasterization and ray tracing into a single die. The conventional GPU core is based on the same Series6 XT hardware that PowerVR unveiled earlier this year — these new chips are tweaked versions of the Series6 architecture that Apple uses for the iPhone 5S, with significantly higher processing capability and a wider architecture.
The new GPU, the Wizard GR6500, will combine a 128-core conventional GPU with a ray tracing unit (RTU) that appears to be nearly identical to the dedicated hardware units we reviewed last summer. It’s difficult to say for certain because PowerVR was notoriously tight-lipped with details on how the PowerVR RTU functioned, but certainly the quoted specifications and capabilities appear to be identical.
According to Anandtech’s writeup, Imagination Technologies claims the RTU can deliver 300 million rays per second when clocked at 600MHz. That’s compared to a measly 50 million rays for the current single-chip R2100 and 100 million rays for the dual-core R2500. This squares with what we know of the first-generation rendering cards, which reportedly ran at roughly 100MHz on a 90nm process.
A performance jump this massive might raise eyebrows in other contexts, but it’s not surprising here. With a normal 2-3 year gap between product discussions and shipping hardware, any Wizard-class GPU that eventually debuts will launch on 16nm FinFET or 14nm-XM nodes at TSMC or GlobalFoundries. The dual-RTU R2500′s power consumption was impressive even with 16GB of 90nm DDR2 attached; we expect the technology to shine when given the benefit of four full process nodes and SoC integration.
What kind of game experiences could a hybrid rendering model drive?
Meanwhile, Imagination Technologies’ decision to integrate an RTU directly into the conventional graphics pipeline is a smart thing to do. One of the things we’ve discussed in our previous ray tracing coverage is the difficulty of creating an entire ecosystem for ray-traced titles. Rasterization (i.e. the current, conventional method of on-screen drawing) has often been derided as a “trick” of graphics, but these are tricks that artists, developers, and programmers have gotten exceptionally good at. While ray tracing offers some significant benefits when it comes to certain visual effects, rasterization has repeatedly proven it can rise to the challenge of creating “good enough” game experiences.
With Imagination Technologies’ next-gen designs, game developers won’t have to choose.
According to IT, a hybrid rendering approach would combine rasterization and ray tracing to allow a developer to choose which light sources cast rays or created other various effects. In the example below, the reflections off the vehicle are ray traced while the rest of the image is rendered using a conventional scanline algorithm.
This hybrid approach would allow developers to leverage the superior performance and power consumption of the hardware ray tracing blocks in scenes where it could complete work more effectively than conventional rasterization allows, then rely on rasterization for the areas where it offers superior performance per watt.
Imagination Technologies reports that it’s already signed Unity Technologies and the Unity game engine to support this capability, with Unity 5.x adding real-time lighting preview support when running in editor mode. More generally, I’d say IT has decided to bet on hardware ray tracing as a major distinguishing characteristic for its mobile GPUs — a functional capability that other companies, like Nvidia and Qualcomm, have yet to implement. Given Imagination Technologies’ reach and market share it’s possible that we’ll see such blocks from multiple companies in the future.
Post a Comment