Hardware recommendations for Cinema4D - HackMD

<style> h1, h2, h3, h4, h5, h6, h7, p, a { font-family: Arial, Helvetica, sans-serif; } </style> <center> <br/> <br/> <br/> <img src="https://raw.githubusercontent.com/Elemento-Modular-Cloud/graphic-assets/main/logos/horizontal/Logo%20horizontal%20lightbg%20transp.svg" width=50%/> <br/> <h4> ELEMENTO technical docs </h4> </center> <center> <img src="https://i.imgur.com/T5Rvdaj.png" width=50%/> <h1> Hardware recommendations for Cinema4D </h1> </center> ## CPU recommendations Cinema4D by Maxon is known to have two CPU spec dependencies: * **Number of cores:** important when using the CPU for rendering * **Core frequency:** crucial while developing the scene and composition The two aspects are treated below. The reference benchmark of the Cinema4D CPU-renderer is perfectly represented by the Cinebench benchmark by Maxon. <center> <iframe width="80%" height="600" seamless frameborder="0" scrolling="no" src="https://docs.google.com/spreadsheets/d/e/2PACX-1vTfHe3pNJtaudHTNBI_PThpO8MnBBekLycIsULBTgQ4GSoUQxXl5Jry8XQYwrHTJTErWTqtDu63jW8W/pubchart?oid=134864087&format=interactive"></iframe> </center> The graph above by Puget Systems displays the strong dependence on the number of cores available for the rendering process. The performance scaling in not linear (double the number of cores is not going to be twice as fast). The graph below shows such absence of perfect scaling: <center> <iframe width="600" height="603" seamless frameborder="0" scrolling="no" src="https://docs.google.com/spreadsheets/d/e/2PACX-1vTfHe3pNJtaudHTNBI_PThpO8MnBBekLycIsULBTgQ4GSoUQxXl5Jry8XQYwrHTJTErWTqtDu63jW8W/pubchart?oid=228137754&format=interactive"></iframe> </center> To display the fact the real scaling happens against the product of the `core count` and the `frequency` the same graph can be divided by the frequency of each processor: <center> <iframe width="600" height="603" seamless frameborder="0" scrolling="no" src="https://docs.google.com/spreadsheets/d/e/2PACX-1vTfHe3pNJtaudHTNBI_PThpO8MnBBekLycIsULBTgQ4GSoUQxXl5Jry8XQYwrHTJTErWTqtDu63jW8W/pubchart?oid=1495946238&format=interactive"></iframe> </center> While higher CPU cores count processors represent an efficient choice, the performances in Cinebench and Cinema4D show three specific products which lay out from the `(core count) x frequency` graph. The products are the following: * **Intel Core i9 12900k**: the low core count (8+8) affects the peak performances. This CPU is not workstation or server graded. * **AMD TR Pro 5965WX**: the 24 cores are well balanced by an high frequency (3.8GHz) which makes this AMD product the most balanced one. * **AMD TR Pro 5975WX**: very much equal to the previous product but with 32 cores and 10% more performances. Honorable mention to the AMD TR Pro 5995WX thanks to the 64 cores with much lower frequency (2.7GHz). To complete the picture we display the performance per Watt and per € plots below: <center> <table> <tbody> <tr> <td> <iframe width="600" height="603" seamless frameborder="0" scrolling="no" src="https://docs.google.com/spreadsheets/d/e/2PACX-1vTfHe3pNJtaudHTNBI_PThpO8MnBBekLycIsULBTgQ4GSoUQxXl5Jry8XQYwrHTJTErWTqtDu63jW8W/pubchart?oid=154217791&format=interactive"></iframe> </td> <td> <iframe width="600" height="603" seamless frameborder="0" scrolling="no" src="https://docs.google.com/spreadsheets/d/e/2PACX-1vTfHe3pNJtaudHTNBI_PThpO8MnBBekLycIsULBTgQ4GSoUQxXl5Jry8XQYwrHTJTErWTqtDu63jW8W/pubchart?oid=1688695663&format=interactive"></iframe> </td> </tr> </tbody> </table> </center> While the Intel consumer chip breaks the performance per € plot, thanks to the lack of professional features and the resulting lower price, the performance per Watt plot is ruled by AMD offerings. ### Conclusions :::warning Our top pick for a smart top performance platform would be the **AMD TR Pro 5965WX** SKU with an option to upgrade to larger models (5975WX and 5995WX). The Intel consumer grade CPUs remain a great choice for casual non-professional use. ::: --- ## RAM recommendations The RAM configuration needed by Cinema4D is much less complex than what we saw for CPU power. Typically Cinema4D uses less than 16GB of RAM, with some peaks around 32GB. Only rather specific uses with extreme rendering resolutions (e.g. 8K) require 64GB of system memory. --- ## GPU recommendations Cinema4D requires a great distinction regarding GPUs. In case the user uses plugins to perform GPU-rendering (e.g. Redshift, Octane, V-ray) the GPU computing power is crucial. On the contrary, if the GPU is used just for visualization, an entry level GPU is more than enough. Both setups are treated below. ### Best choice for visualization To handle the Cinema4D viewport during modeling with no issues an entry level GPU with sufficient memory to handle 4K display resolution is more than enough. The best options we selected among the currently available SKUs are: * **Nvidia 3060 Ti 8GB:** consumer grade GPU good for great price/performance rating. * **Nvidia A2000 6GB:** entry level prosumer GPU with low power consumption (75W). * **Nvidia A2000 12GB:** same GPU with twice the memory for improved support for texturing and higher resolution. ### Best choice for GPU-rendering GPU rendering depends on huge GPUs with large memory endowment. Our best picks are the following: * **Nvidia 3090 Ti 24GB:** reference high-end GPU expressed in a prosumer package. * **Nvidia A5500 24GB:** professional expression of the Nvidia 3090 with ECC memory. * **Nvidia A6000 48GB:** recommended product for huge texture sets and large resolutions thanks to the larger ECC memory. * **Nvidia A6000 Ada Lovelace 48GB:** pinnacle of current technology, even faster and with double the RAM with respect to the consumer flagship Nvidia 4090. In addition, the performances scale almost linearly with the number of GPU, as shown in the below graphs. <center> <img width=49% src="https://i.imgur.com/9RLO9HU.png"/> <img width=49% src="https://i.imgur.com/BhXERlr.png"/> <img width=49% src="https://i.imgur.com/ZydFpBt.png"/> </center> This observation induces ourselves to suggest to the customers a setup made with the following pattern: 1. Maximize first the power of each single GPU to gain maximum vertical scaling * for improved future proofness * to provide each single-GPU virtual machine the right amount of performance 3. Add more GPUs to further scale the rendering power * rendering tasks can be performed merging together all the GPU in a single huge virtual machine * horizontal scaling is well supported by the software ### Conclusions ::: warning We recommend the **Nvidia A2000 6GB** GPU for visualization tasks since the power consumption is incredibly low and the density of hardware achievable with such package is disruptive. ::: ::: warning The **Nvidia A6000 48GB** GPU is a great package to work with huge textures. In addition the non-Ada Lovelace version is getting discounted while keeping the possibility to get partitioned to host multiple users concurrently for visualization tasks. The **Nvidia A5500 24GB** GPU is a possible slight downgrade. **Both GPUs can be installed in multiple copies to scale up performances.** :::