Hardware
- CPU: AMD Ryzen Threadripper 39950x Pro 64-Core.
- Motherboard: Asus Pro WS WRX80E SAGE.
- System Memory: 256 GB DDR4 3600 CL18.
- GPUs: 8x RTX 3090Ti Founders Edition.
- Bandwidth: PCIe Gen 4.0, x8.
- OS: Windows 11 Pro.
- Nvidia Studio drivers 536.40.
- No monitors attached to the GPU(s).
Testing
We have tested 3 scenes on this hardware using Bucket mode, the results are double-checked. Scenes are set up carefully to push the GPUs to 100% usage, to ensure accurate results. All tests run for at least one full minute.
Note: although V-Ray GPU officially supports a maximum of 4 GPUs for rendering, we added the results for rendering with 8 GPUs.
Test 1:
We noticed CUDA scaling of 99.61% scaling for 2 GPUs, 96.08% for 4 GPUs and 88.47% for 8 GPUs
The noticed RTX scaling of 101.13% for 2 GPUs, 96.26% for 4 GPUs and 92.35% for 8 GPUs:
Test 2:
We noticed CUDA scaling of 97.41% scaling for 2 GPUs, 99.51% for 4 GPUs and 88.72% for 8 GPUs
The noticed RTX scaling of 97.98% for 2 GPUs, 97.71% for 4 GPUs and 95.03% for 8 GPUs:
Test 3:
Note: this scene includes thousands of leaves and grass objects, and has all shading baked to textures, hence showing a big boost for RTX mode. In most cases, CUDA mode should be close to RTX mode in performance.
We noticed CUDA scaling of 99.11% scaling for 2 GPUs, 98.89% for 4 GPUs and 89.79% for 8 GPUs
The noticed RTX scaling of 99.79% for 2 GPUs, 97.69% for 4 GPUs and 93.75% for 8 GPUs:
Conclusion
V-Ray GPU scales near linearly across multiple GPUs. We have recorded an average of 97% with 4 GPUs and 91% with 8 GPUs, the highest efficiency is usually found when combining GPUs of the same generation and of similar performance. Furthermore, we recommend having 6 physical cores at least per GPU when setting up machines with multiple GPUs, new CPUs and motherboards with PCIe 4.0 could be helpful.