BoostClock
GPU rendering - Maxwell vs Pascal vs Volta performance scaling - V-Ray, Redshift, Indigo Renderer, LuxMark, Blender Cycles

How do Nvidia's different GPU microarchitectures perform in GPU rendering workloads - Volta (NVIDIA Titan V) goes up against Maxwell and Pascal in V-Ray, Redshift, Indigo Renderer, LuxMark, Blender Cycles.

Everything points toward an August-September launch window for the next-generation GeForce graphics cards as NVIDIA will host the "GeForce Celebration Event" on 20th August at Gamescom 2018. For a long time, Nvidia’s Volta was tipped to succeed Pascal as the architecture of next-generation consumer graphics cards, now, it is unknown how Nvidia will segment their line-up and what are the actual differences between different codenames like Turing and Ampere. Nevertheless, I crunched the data of our recent NVIDIA GPU rendering benchmarks and plotted the results in a diagram where the peak theoretical performance goes on the x-axis and the actual rendering performance relative to the top performers goes on the y-axis. With this kind of data presenation we can speculate how the next-gen cards could potentially perform in these GPU rendering scenarios.

Test methodology

The card in question was used to drive the display - some applications gain performance when a card is a compute only device. All tests were executed three times (average of these runs are presented in all cases), the test was repeated in case of outlier results or anomalies.

V-Ray provide a free stand-alone application to evaluate how fast your hardware renders.

Redhsift offer a demo version of their fully fledged application, it is functionally identical to the commercial version, but renders with a watermark. For the data points in this video I used the command line version of the tool, instead of the benchmark variant - strangely enough, they don't produce the same results. The command line version provides a more in-depth log file about the various stages of the rendering process - the time of the "Rendering blocks..." stage is used as the result.

The Cycles render engine in Blender was assessed with help of the Blender Institute-prepared benchmark pack + the recent Barbershop Interior scene from Agent 327 animated feature film. Render time is extracted so that it only covers pure path tracing time - no OpenCL kernel compilation, scene loading, CPU-side BVH construction, final composition - and Million samples per second is calculated from the resolution / resolution percentage / sample count values. Blender is launched headless (no GUI) with a python script responsible to get everything sorted and start the rendering process.

Indigo and LuxMark provide neat benchmark tools with multiple scenes to test. Results are provided in Million samples per second.

Hardware setup

PSU: Cooler Master 1000W VANGUARD

MOTHERBOARD: Gigabyte GA-AB350-Gaming 3

CPU: AMD Ryzen 5 1600

GPU Maxwell: MSI GTX 960 GAMING 4G / MSI GTX 970 GAMING 4G / MSI GTX 980 GAMING 4G / MSI GTX 980Ti GAMING 6G

GPU Pascal: MSI GTX 1050Ti GAMING X 4G / MSI GTX 1060 GAMING X 6G / MSI GTX 1080 GAMING X+ 8G / MSI GTX 1080Ti GAMING X 11G

GPU Volta: NVIDIA TITAN V

OS: Ubuntu 16.04.4 LTS x86_64 - Linux 4.13.0-45-generic

DRIVER: CUDA 9.2.106 - NV 396.26

RAM: G.Skill FlareX 16GB (2X8GB) DDR4 3200MHz

STORAGE: Samsung m.2 SATA 500GB SSD 850 EVO

COOLER: AMD Wraith Spire Cooler