Tibor Nyers writes:
With the release of Turing NVIDIA created a new class of GPU devices specifically tailored to target ray tracing workloads. Unfortunately, Blender Cycles does not support HW accelerated rendering just yet, nevertheless, the new RTX GPUs provide a significant leap in GPU performance. In this article we compare the performance speedup with Blender v2.79b and Blender v2.80 beta using three top-end GPUs - RTX 2080 Ti, Titan V and GTX 1080 Ti - Turing, Volta and Pascal. Additionally, varying tile size measurements give further insight to extract maximum speed from Cycles.
8 Comments
Another proof of a total scam from Nvidia. Twice the price 2 years later : same performances ….
Are you kidding us expecting no benchmarks MrNV ? I'll keep my cheap n powerfull 1070Ti for a while…..
I will say it's worth looking at the breakdown of those results, the 2080 and Titan V actually outperforms the 1080 in every single test scene except for one. (Often by a substantial margin) The chart on this page only shows the difference of how each card is effected by upgrading from 2.79 to 2.80 not how they perform relative to each other.
Don't forget that RT cores aren't enabled
It's not a chart of absolute performance, but a relative change from 2.79 to 2.80. It's not that hard to understand. Also Cycles cannot use RTcores and for some time it will stay like that. Octane will have the support soon.
I think there's actually a great case to be made for buying XX60 cards from nvidia and running an SLI setup if you're looking for rendering value. As for this test the 2080 Ti is substantially faster than the 1080 Ti per frame. More than a minute per frame faster in 2.8 w/ the classroom scene at 120x90 render blocks. Especially over long render projects this would make a huge difference.
Can you test with E-Cycles (https://blenderartists.org/t/e-cycles-faster-cuda-rendering/1139717) ? Would be interesting to see how it performs with the Titan V.
I will say it's worth looking at the breakdown of those results, the 2080 and Titan V actually outperforms the 1080 in every single test scene except for one. (Often by a substantial margin) The chart on this page only shows the difference of how each card is effected by upgrading from 2.79 to 2.80 not how they perform relative to each other.
Not sure if you use any Threadripper CPU's in your testing. This might help if you do:
https://bitsum.com/portfolio/coreprio/
Coreprio now also offers an experimental function labelled ‘NUMA Dissociater‘.
Dynamic Local Mode (DLM) was conceived for the Threadripper 2990wx and 2970wx CPUs, addressing their asymmetric die performance. There are potentially use cases for other CPUs as well.
DLM works by dynamically migrating the most active software threads to the prioritized CPU cores. In the case of the TR, that means the first half of logical CPUs. The Windows CPU scheduler is then free to choose specifically where within that set of CPUs to assign threads. Since no hard CPU affinity is set, applications are still free to expand across the entire CPU. For this reason we call it a prioritized soft CPU affinity.
The NUMA Dissociater is confirmed to work on EPYC 7551 and TR 2970/2990, but may work on other HCC NUMA platforms.