GPU engine

What is a GPU engine and why should we use it?

A little bit of background.

Until version 3, Maxwell has only rendered using the power of the CPU of the computer and RAM memory of the system. Now with Maxwell 4, we added a brand new render engine that uses the computing power of the processor of the graphics card (GPU) and its memory, so now you can choose to render as before with traditional CPU engine or use the new GPU engine.

Traditionally, GPUs main tasks are showing things on the screen and moving triangles in 3D viewports; because of this, GPUs manufacturers were focused making the graphic chips able of making a big amount of simple calculations very fast; that's why modern GPUs have lots of cores but usually slower than CPU ones. At the moment, the power of GPUs has increased and we can count the number of cores in thousands instead of units or tens like CPU cores.

So, as we have all those cores available, could we use them for rendering too? Yes! That's the purpose of the GPU engine. When using it, Maxwell will be able to calculate many more rays at the same time (as there are many more cores available), although complex light paths could take more time to be calculated than with CPU (as GPU cores are usually slower). In general, you will experience a very noticeable speed increase (between 4 and 15 times faster than CPU), especially in scenes where light doesn't have to bounce a lot or the light path is not very complex.



GPU engine hardware requirements

  • Graphics cards have to be based on CUDA. The GPU engine is built upon CUDA computing platform, which is developed by Nvidia, so only Nvidia graphics cards will work. AMD or Intel cards won't work.

  • Maxwell and Pascal micro-architectures are supported (compute capability 5.0 and up). Kepler architecture is not supported but cards based on it could work.

  • Graphics card drivers should be up to date.

 

Here is a table of supported graphics cards:

GPU architecture 

 GeForce

Quadro and NVS

Tesla

GPU architecture 

 GeForce

Quadro and NVS

Tesla

Maxwell

GeForce GTX 750 Ti, GeForce GTX 750, GeForce GTX 960M, GeForce GTX 950M, GeForce 940M, GeForce 930M, GeForce GTX 860M, GeForce GTX 850M, GeForce 845M, GeForce 840M, GeForce 830M, GeForce GTX Titan X, GeForce GTX 980 Ti, GeForce GTX 980, GeForce GTX 970, GeForce GTX 960, GeForce GTX 950, GeForce GTX 750 SE, GeForce GTX 980M, GeForce GTX 970M, GeForce GTX 965M

Quadro K1200, Quadro K620, Quadro M2000M, Quadro M1000M, Quadro M600M, Quadro K620M, NVS 810, Quadro M6000 24GB, Quadro M6000, Quadro M5000, Quadro M4000, Quadro M2000, Quadro M5500, Quadro M5000M, Quadro M4000M, Quadro M3000M.

Tesla M10, Tesla M4, Tesla M40, Tesla M6, Tesla M60

Pascal

Nvidia Titan Xp, Nvidia Titan X, GeForce GTX 1080 Ti, GTX 1080, GTX 1070, GTX 1060, GTX 1050 Ti, GTX 1050, GT 1030

Quadro GP100, Quadro P6000, Quadro P5000, Quadro P5000(Mobile), Quadro P4000, Quadro P4000(Mobile), Quadro P3000(Mobile), Quadro P2000, Quadro P1000, Quadro P600, Quadro P400

Tesla P100, Tesla P40, Tesla P4



Multi-GPU (coming soon)

With Maxwell you could now get render processes up to 8 times faster, as it brings a lot of new capabilities such as a Denoiser feature, optimized GPU support, Multilight™ standalone tool, and more. Soon Maxwell will include MultiGPU support, which will multiply the processing power of the workstations depending on the number of GPUs installed. Maxwell users will be able to connect as many GPUs as your system support. Stay tuned.  See all the details here



Which card should I buy to use with the GPU engine?

Only Nvidia, no AMD or Intel, sorry.

As usual, the bigger the numbers, the better, but which numbers do really affect the render?

  • The memory of the graphics is crucial as the whole scene has to fit in it to be able to render, so the higher the better. The amount of memory is not relevant for the speed of the render though; this will affect the size of the image you can render, the amount of geometry you can load and the number of textures that can be loaded to render.

  • The higher the number of CUDA cores and their speed, the better. This will determine the rendering speed for that particular graphics card.

  • The size of the memory interface and its bandwidth will also affect speed as it affects how fast the information travels between the memory and the cores. The higher the better.