V-Ray Tip: Optimizing Multi-GPU V-Ray Performance

Multi-GPU rendering in V-Ray can deliver dramatic speedups on large scenes—if you configure devices, memory, and scheduling for balance rather than raw count.

• Enable the right engine:
  • Switch to V-Ray GPU (RTX) in your Render Settings. RTX mode leverages hardware RT cores for faster ray tracing on supported NVIDIA cards.
  • Open GPU Device Selection and explicitly choose which GPUs to render on. Leave the display/GUI GPU unchecked to keep the viewport responsive.
• Balance heterogeneous GPUs:
  • The slowest GPU can gate overall progress. If you mix old and new cards, consider disabling the weakest device—often faster overall than waiting on it.
  • VRAM is mirrored across GPUs; the smallest VRAM device sets the memory ceiling. Prefer similarly specced cards for best scaling.
  • Bucket rendering: try 256–512 px buckets. Smaller buckets improve load-balancing among GPUs; overly tiny buckets add scheduling overhead.
  • Progressive rendering is great for lookdev (excellent natural balancing); use Bucket for predictable finals and stable memory.
  • Advanced: pin devices per machine with the VRAY_GPU_DEVICES environment variable when automating farm jobs.
• Make large scenes fit:
  • Enable On-demand mip-mapping to stream texture resolution by need, cutting VRAM pressure with minimal visual cost.
  • Turn on out-of-core textures for safety on massive assets; expect some slowdown when spilling to system RAM.
  • Use V-Ray Proxy geometry for heavy models and instance wherever possible; avoid unique copies of repeating assets.
  • Prefer bump/normal to displacement for mid/far assets; keep displacement to close-ups and enable adaptive displacement only where needed.
  • Compress and tile textures; use UDIMs wisely; prune 8–16K maps that never hit screen as such.
• Sampling and denoising:
  • Let the Adaptive sampler drive quality; target a realistic Noise threshold (e.g., 0.01–0.03 for previews, tighter for finals).
  • Use the NVIDIA OptiX denoiser for instant IPR feedback; switch to the V-Ray Denoiser render element for finals and better AOV consistency.
  • Keep glare/bloom in the VFB post stack so they don’t inflate sampling costs across GPUs.
• Scale out with Distributed Rendering (DR):
  • V-Ray GPU supports multi-GPU across multiple machines. Match V-Ray versions and NVIDIA drivers across nodes for stability.
  • Serve assets via UNC/absolute paths and shared caches; a 10GbE (or faster) network avoids I/O stalls when compiling kernels or streaming textures.
  • Use the Light Select and Cryptomatte elements to minimize re-renders when rebalancing lights in comp.
• Diagnostics and reliability:
  • Watch the V-Ray Log for “device out of memory” and fallbacks; adjust texture/mip settings or deselect the limiting GPU.
  • Profile with short region or lower-res tests; confirm scaling by adding one GPU at a time, not all at once.
  • Keep GPUs cool and power-stable; thermal throttling ruins load balance. Clean cases and ensure adequate PSU headroom.

Need guidance on building or expanding a balanced multi-GPU setup for V-Ray? Talk to the experts at NOVEDGE for hardware recommendations, licensing, and upgrades, or browse current V-Ray offerings directly on novedge.com.