OTOY Forums

We’re still optimizing iray for optimal Kepler performance here at NVIDIA, so it’s not Autodesk dragging their feet here. Our team is working hard to get out something you will like this summer. But to set expectations, you should not expect the initial Kepler products (out now) to deliver a dramatic speed increase for iray over their Fermi generation predecessors. While Kepler has many more cores than Fermi, they run at much lower power, which means they have less performance per core. The gain you are guaranteed to see is superior performance per watt. This also makes it much easier to fit larger or more GPUs into power-constrained systems.

As for viewport/raster performance, it’s quite possible that many high end GPUs will not report high usage unless your scene is really taxing the GPU, most likely with many programmable shaders. High face counts and texture usage impact memory far more than they do workload. The graphics pipeline itself can also have a bottleneck. Here NVIDIA is working with Autodesk (and many other companies) to eliminate unnecessary data transfers that hold back the GPU. This not a reflection on Autodesk’s ability to design but rather how much more rapidly GPUs have evolved than CPUs, as these practices were often negligible to performance a GPU generation or two ago. The good news is that once these “speed bumps” get removed, all modern GPUs should benefit.

- Phil
Author: Phil Miller