Choosing a Computer for GPU Rendering
If we compare Redshift, Octane, VRay GPU (formerly Vray RT) with traditional CPU renderers, both types utilize the graphics card for processing. However, GPU renderers have made significant advancements in terms of image quality and functionality. While there are still some challenges in GPU rendering, the advantages outweigh the disadvantages. In this article, you will learn about choosing the best computer for GPU rendering.
Choosing a Computer for GPU Rendering
GPU (Graphics Processing Unit) is a separate processor located on the graphics card that handles the processing of 2D or 3D graphics. With a dedicated processor on the graphics card, the computer's CPU is freed from additional work and can perform other important tasks faster. The GPU is designed to maximize the speed of graphics calculations, such as textures and objects. Due to its architecture, GPU processors are much more efficient at processing graphics information compared to typical central processing units (CPUs).
When selecting a processor, the focus is on the quality of the cores since GPU engines utilize the functionality of the graphics card during rendering. Therefore, the maximum core frequency is crucial. Suitable processors for this purpose include Intel i9 9900K (3.6 GHz base frequency and up to 5 GHz in turbo mode) or i7 8700K (3.7 GHz base frequency and up to 4.7 GHz in turbo mode).
When choosing a processor, it is important to pay attention to the supported number of PCIe lanes because graphics cards communicate with the CPU through PCIe lanes on the motherboard. The number of these lanes depends on how many the CPU supports. In the case of typical graphics cards, to achieve their full performance, you will need a resource of 16x PCIe lanes 3.0. This means that one graphics card paired with an i9 9900K or i7 8700K will work at its maximum potential since each of them supports 16 lanes. If you want to install multiple graphics cards for GPU rendering and extract their maximum performance, you will need a processor with a higher number of PCIe lanes, such as the AMD Threadripper 2950X with 64 lanes or the i9 7800X (28 lanes) or i9 7900X series processors (44 lanes).
It is also worth noting that a graphics card can operate at a slower speed by utilizing fewer than the full 16x PCIe lanes, such as 8 lanes. In terms of the cost-effectiveness of this matter, installing multiple graphics cards in a computer does not make much sense because, in practice, the difference between 16x and 8x modes is only a few percentage points. However, when comparing 8x and 4x modes, 4x will significantly reduce performance. Therefore, an optimal choice in this case would be a graphics card that supports 8x lanes.
The image shows the performance difference of the Titan X graphics card in Octane Render in x16 and x8 modes. The difference is very small, so if you plan to install 2+ graphics cards in your system, it is not worth chasing after this. During rendering, when the entire scene fits in the video memory (VRAM), the speed at which the rendering completes becomes a question of the graphics card's performance itself. However, there are preceding processes that can significantly burden the CPU and hard disk before or during rendering.
In very complex scenes, processes such as unloading and preparing meshes for processing on the graphics card, loading textures from the disk, and preparing scene data can significantly increase the rendering process if you have insufficient RAM, a slow processor, or a non-performant hard disk. If there is not enough memory, the rendering process will utilize the regular RAM, and if memory still runs out, the data will be swapped to the swap file, which will significantly slow down the process.
RAM (Random Access Memory)
It is important for the RAM not to hinder the entire rendering process, and the amount of RAM should always remain sufficient, regardless of the type of RAM used, as it does not provide a significant time advantage for GPU rendering.
Here are a few nuances to consider when choosing RAM for GPU rendering:
Remember that you need twice as much RAM as the system's video memory (VRAM).
Paying extra for top-of-the-line RAM will always be expensive and not always justified because memory frequency is important but not critical.
If your budget is limited, buying multi-channel memory does not make sense; although it is better, it is not necessary in this case.
Pay attention to the fact that RAM with CL14 will work faster than CL16; consider the latency indicator of the RAM.
Graphics Card for Rendering
V-Ray GPU (formerly V-Ray RT) supports OpenCL technology and works well with AMD graphics cards, but this is more of an exception. It is better to choose Nvidia instead. Radeon cards from AMD are not suitable for rendering in Octane and Redshift. To use GPU renderers like Octane or Redshift, we need Nvidia graphics cards as they are the ones that support CUDA technology.
So, it is clear that the top-end cards are the most powerful but also the most expensive. When choosing a graphics card, we primarily look for the best balance in different price categories, considering the optimal price-to-performance ratio. Buying a good used GPU card on the market is a stroke of luck, especially after the decline of cryptocurrency. Therefore, we rely on the prices in the store for new hardware.
If we consider affordable graphics cards, the best performance in terms of the investment value can be seen in the GTX 1070 and RTX 2070, based on the price-to-performance ratio. Compared to the GTX 1060/RTX 2060, which only have 6 GB of VRAM, the GTX 1070 and RTX 2070 have 8 GB of VRAM. Keep this in mind if you want to save money.
If you have no budget constraints, you can confidently purchase the RTX 2080 (8 GB VRAM) or GTX 1080Ti (11 GB VRAM). In the long run, the RTX series will have an advantage as it has hardware support for RT (ray tracing) cores. However, if you don't want to wait, the best choice would be the GTX 1080Ti (specifically the Ti version) due to its good performance and 11 GB of video memory.
From the high-end series such as the RTX 2080Ti and Titans (Titan V and Titan RTX), you can consider these options. They provide 24 GB of video memory and excellent performance improvement. For processing highly polygonal scenes with up to 200 million unique objects, you will need graphics cards with 11 GB or more of video memory, but this is not necessary for everyone, so it's subjective.
One more thing: when installing multiple graphics cards for GPU rendering in a single system, it is recommended to use cards with the same amount of video memory or cards from the same class. This is because the video memory will be limited by the smaller value and will not be summed up. For example, if you install a GTX 1080Ti (11 GB) and a GTX 1060 (6 GB) in one system, the rendering will be limited by the 6 GB of video memory.
Cooling for the Graphics Card
Let's consider the Founders Edition graphics cards; they come with a blower-type cooling system that exhausts hot air from the rear, which facilitates the assembly of a multi-GPU setup that will be well-cooled. However, blower-style coolers tend to be louder compared to regular cooling systems.
If you don't like noise, then you should choose graphics cards with regular air cooling, but it might be challenging to place them in a single system block to avoid overheating. The most effective cooling solution is water cooling, which provides better performance, but it is expensive and requires additional space for the water reservoir and radiator.
Power Supply Unit (PSU)
If you have a computer with a single graphics card, a 500W power supply should be sufficient. Most GPUs consume around 180-250W. If you plan to add another graphics card, add around 250W per card, and make sure that the power supply unit matches the required power.