GPU Rendering – NVIDIA GeForce RTX 3060 Ti Performance

Introduction

NVIDIA launched the GeForce RTX 30 Series a few months ago, but new models in this family continue to trickle in. Today we are looking at the RTX 3060 Ti 8GB model and how it performs with regard to GPU-based rendering in OctaneRender, Redshift, and V-Ray.

The NVIDIA GeForce RTX 3060 Ti, which is the focus of this article, is effectively a cut-down version of the RTX 3070. It uses the same design but with some of the CUDA cores disabled – yet retains the full 8GB of video memory. This should result in slightly lower performance than the 3070, and from NVIDIA's Founders Edition pricing it looks like this comes with ~$100 savings in cost.

Here is a table showing the various specifications of the NVIDIA GeForce RTX 30 Series cards that have been released so far:

Test Setup

Listed below are the specifications of the system we used for this round of testing:

As you can see, we are comparing the RTX 3060 Ti to the rest of the RTX 30 Series as well as the RTX 2080 Ti, which was the most popular video card we sold for rendering in the last generation. To test these five video cards we used three different GPU rendering benchmarks: OctaneBench 2020, Redshift 3, and V-Ray 5. Each benchmark was run twice, and the fastest score was used here.

All of these charts show results in order of performance, from top to bottom, and the new GeForce RTX 3060 Ti is highlighted in green.

OctaneBench Results

Redshift Demo Results

V-Ray Benchmark Results

For V-Ray we have two sets of results, from running the benchmark in both CUDA and RTX modes, in the gallery below:

Analysis

Across these three benchmarks, the GeForce RTX 3060 Ti comes in only 6 to 14% behind the RTX 3070. That is quite a good showing, considering that the MSRP is 20% lower than that card. The exact performance difference does vary from one rendering engine to another, though, with Redshift seeming to have the smallest difference (~6%) while V-Ray 5's CUDA mode has the largest (~14%). This level of performance also means that the 3060 Ti matches or beats the RTX 2080 Ti, which was our most popular GPU for rendering workstations during its heyday and cost three times as much money!

We didn't test dual 3060 Ti cards in this roundup, but based on recent multi-GPU scaling results for other 30 Series cards a pair of these should do very nicely as well – roughly doubling render performance. Just be aware of the cooling issues surrounding multiple cards! With the Founders Edition models from NVIDIA, we have found that two cards can be used in tandem without much trouble – as long as there is plenty of airflow from the chassis fans. The bulkier, triple-fan cards that many other manufacturers are fond of producing are more difficult, both because of their sheer size and the fact that those designs tend to vent all their heat back into the case.

It is also worth noting the other big limitation that the RTX 3060 Ti shares with the 3070 when it comes to rendering: VRAM. Since the RTX 3060 Ti and 3070 have only 8GB of onboard memory, compared to the RTX 3080's 10GB and the 3090's 24GB, they will be much more limited in terms of scene complexity and texture resolution. Some rendering engines can use system memory for portions of that data, known as "out of core memory", but doing so will usually have a negative impact on performance.

Conclusion

NVIDIA's GeForce RTX 3060 Ti offers solid performance in GPU rendering engines, coming very close to the RTX 3070 for $100 less!

For an entry-level rendering system, then, the RTX 3060 Ti is a great choice. With the Founders Edition cards that NVIDIA sells directly, two of these cards should be doable in many cases – as long as there is some space between the cards and plenty of airflow from chassis fans. For more than two GPUs, though, blower-style cards are really needed… and it seems unlikely that those will be coming any time soon, if the lack of RTX 3080 and 3070 blower models is any indication. Users who want more than two GPUs, NVLink support, or more than 8GB of video memory for scene data will probably want to look at the RTX 3090 24GB or the upcoming A6000 48GB instead.