Award-winning PowerVR GR6500 ray tracing GPU tapes out

[Valentyn]

Ray tracing is finally getting more traction and PowerVR's new GR6500 is coming. It's promising to be x100 times more efficient than using current GPU compute or other software-only approaches,

This could finally help push Ray Tracing into something that someone outside of a massive render farm can work with, without sacrificing hours upon hours of waiting to get a scenes done.

 

Sadly we won't be getting our hands on these new GPUs just yet, and we could still be waiting a few years. 

 

 

 

Firstly, I’d also like to take this opportunity to congratulate everyone in the engineering team; they have worked really hard to take the Wizard architecture from a theoretical concept to a mature, real-world implementation that can now be found in silicon.

For those who haven’t kept track of our ray tracing solution, the PowerVR GR6500 GPU is the culmination of eight years of effort from a global team of designers. The technology was originally conceived at Caustic Graphics, a startup with a vision for interactive ray tracing, going against twenty years of industry momentum.
 

One of the many advantages of PowerVR Wizard ray tracing GPUs lies in the power efficiency gains delivered by the architecture. Thanks to the innovative nature of the design, the majority of data can remain in a subset of the local cache, eliminating the need for any high bandwidth random access path between processing elements and internal cache memory.

This coherence gathering results in huge savings in memory bandwidth and power consumption, making Wizard ray tracing GPUs 100x more efficient compared to using GPU compute or other software-only ray tracing approaches on traditional graphics architectures. The approach has the added benefit of recollecting SIMD coherence, enabling graphics processors to tackle previously unapproachable challenges.

 

 

 

 

 

                            

 

 

PowerVR GR6500: ray tracing, graphics and compute coming together in one revolutionary GPU

The first member of the Wizard family is PowerVR GR6500. This ray tracing processor is based on a latest generation, quad-cluster PowerVR Rogue design, with all the bells and whistles of our recently-launched PowerVR Series6XT GPU processors, including:

 

Full-blown graphics and compute performance: four Unified Shading Clusters (USCs), with 128 ALU cores delivering more than 150 GFLOPS (FP32) or 300 GFLOPS (FP16) at 600 MHz

 

Unmatched real-world ray tracing performance: Up to 300 MRPS (million rays per second),  24 billion node tests per second and 100 million dynamic triangles per second at 600 MHz

 

PowerGearing G6XT for advanced power management and dynamic resource allocation

 

PVR3C™ triple compression technologies (PVRTC and ASTC for texture compression, PVRIC for frame buffer compression, PVRGC for geometry compression)

 

Deep Color support for very high image quality at Ultra HD resolutions and beyond

PowerVR GR6500 is designed to provide leading support for a range of APIs such as OpenGL ES 3.1/2.0/1.1, OpenGL 3.x, Direct3D 11 Level 10_0, OpenCL 1.2, and OpenRL 1.x.

 

There is no  currently listed support for Vulkan or DX12, but seeing as neither of those have official launched and the GPUs are still some time away we could see these added later.

 

Sources: 

 

http://blog.imgtec.com/multimedia/award-winning-powervr-gr6500-ray-tracing-gpu-tapes-out

 

http://blog.imgtec.com/powervr-developers/powervr-gr6500-ray-tracing

[LucidMew]

So I take it this GPU is mostly for large companies like Pixar and Disney and then in 5 years it'll be for medium sized companies, and 5 years later pro-sumer/content developers will get their hands on it, and then in another 10 years intel's iGPU will have it.

[patrickjp93]

Let's see how they do. So far CPUs have been king at it because of the branch-heavy structure of it. It's a bit difficult to imagine dedicated hardware for it because of the dynamic scales it could be used on.

[Valentyn]

Let's see how they do. So far CPUs have been king at it because of the branch-heavy structure of it. It's a bit difficult to imagine dedicated hardware for it because of the dynamic scales it could be used on.

 

Oh yes. It reminds me a little bit when GPUs started doing compute intensive tasks that CPUs were always great at, and they've been rapidly gaining traction.

If the new PowerVR GPU really can manage x100 improvement, and if it doesn't get delayed to launch we could be looking at something rather interesting. It'll certainly be very interesting to see how the first cards perform in actual use.