my account

The best graphics cards for rendering in FurryBall

Important rendering rules!
 

  • Always use the latest NVIDIA drivers! (some older versions can cause crashing of FurryBall)
  • Remember, if you use FurryBall with OLD, or SLOW GPU, it NEVER is powerful as it is!!! 
    GPU rendering could be MUCH slower than CPU render when you render on old GPU. The meaningful ratio is above 0,5 for serious work - Compare your card with others.
  • Multi-GPU rendering is supported, but we recommend combine only SAME GPUs. Combining LOW and HIGH card in one computer can be paradoxically slower, than only one single GPU.  You can't combine MAXWELL, KEPLER, FERMI and TESLA cards together. (for example: Titan core is KEPLER, but TITAN X is MAXWELL) - More info.
  • Number of GPU in a single computer is unlimited, but we recommend 3-4 at maximum because more GPU is not so effective and overhead is too big.
    In single GPU there is 2000 - 3000 CUDA cores. When your render has to be effective, you have to use ALL of them fully.
    Imagine your image. You have for example top of your image the sky (almost no rendering for GPU) and in the middle of the scene there is a glass object (very slow for rendering). You have 4 GPUs and your computer splits your image into 4 parts. 1 GPU gets the top of the image and render time will be just 5 sec (the rest of the render time will GPU do nothing)… other 2 parts will take you 2 min and the worst part with glass object will take you 10 min… (so your 3 GPU will wait almost 8 minutes for 1 GPU all the time….)

  • Processor (CPU) speed is NOT important for GPU rendering an expensive multi-core CPU is not required, but a fast multi-core CPU will improve the scene loading speed and working in 3D application.
  • The most important for rendering is GPU memory - Don't skimp on GPU Memory!  You can't swap your GPU RAM to HDD - without free GPU memory your scene will ALWAYS crash!
  • System memory is NOT using for GPU rendering - just GPU memory.
  • GPU memory is NEVER multiplied in systems with more GPUs. (2x Titan with 6GB has "just" 6GB total)
  • SLI mode has no effect on FurryBall - multiple GPU can be selected in FurryBall app. (SLI can cause lower GPU memory - turn it off!)
  • Available memory in FurryBall means TOTAL FREE memory in the system (Windows and other application also use GPU memory..)

How to build render farm with FurryBall RT

Tips for building render farm for FurryBall RT.

 

Recommended GPUs

Some of recommended GPUs
 

Nvidia GTX 1070 8GB, 1080 Ti 16 GB
Nvidia  TITAN, TITAN X and Z , - 6 - 12 GB
Nvidia Quadro P6000, K6000, M6000 / Tesla K40 - 12 GB 
Nvidia GTX 980, 980 Ti - 4 GB, 1060, 1080
Nvidia GTX 780, 780 Ti, - 3 GB (fast as Titan, but just 3 GB)
Nvidia GTX 590, 690, - 1.5 - 3 GB

Nvidia Quadro K5000 / Tesla K20 - 4 GB 

Nvidia GTX 680, 770 - 2 - 4 GB
Nvidia Quadro 6000 - 6 GB
Nvidia GTX 480, 580, 660 - 1.5 - 3 GB
Nvidia Quadro 2000, 4000, 5000 - 4 GB 
Nvidia GTX 650, 460, 560, - 2 GB

 

Benchmarks page FurryBall RT
Please visit this FurryBall benchmarks page and you can compare your computer with others

 

Memory consumption examples 

Rendered on Nvidia 680M - 2GB memory

 

Car scene 

  • 1.007.000 polygons  -  280 MB in GPU memory
  • 2 textures on disk (PNG) - 5MB on disk, but uncompressed 50 MB GPU memory
  • Total scene in GPU memory 330 MB

 

Interior scene 

  • 270.000 polygons  -  107 MB in GPU memory

  • 176 textures on disk (PNG) - 208 MB on disk, but uncompressed 1.4 GB in GPU memory

  • Total scene in GPU memory - 1.6 GB