October 24, 2021 iRendercs

Render Faster With Multi-GPU in Blender & Cycles

As you might know, EEVEE has real-time rendering capabilities that save you a lot of time in post-production, while Cycles is very flexible and brings great realism to the project, not for speed. However, with Multi-GPU settings, Cycles can nevertheless deliver an impressive level of render speed.

In this article, iRender will bring you deep knowledge about GPU rendering in Blender & Cycles. Unfortunately, GPU rendering is only supported on Windows and Linux; macOS is currently not supported.

Multi-GPU Rendering for Blender & Cycles

Supported Graphic Cards

Always make sure to install the latest drivers from the graphics card manufacturer website. These requirements are for basic Blender operation, Cycles rendering using the GPU has higher requirements.

NVIDIA: GeForce 400 and newer, Quadro Tesla GPU architecture and newer, including RTX-based cards, with NVIDIA drivers (list of all GeForce and Quadro GPUs)

AMD: GCN 1st gen and newer. Since Blender 2.91, Terascale 2 architecture is fully deprecated, try using 2.90 (albeit not supported, it might still work) [list of all AMD GPUs]

Intel: Haswell architecture and newer. [list of all Intel GPUs]

Recommended Graphic Cards

Blender’s heavy use of GPU acceleration and its rock-solid CUDA implementation makes an Nvidia graphics card a safe choice for most users. Non-CUDA cards (read: AMD cards) aren’t the best choice for rendering, but they’re still a good choice for users interested in consistent viewport performance.

If you navigate your Scene a lot by moving around the camera in Blender or playing back animations, Nvidia GPUs are a solid choice because CUDA usually supports more features and performs better. Blender’s development team is actively working with Nvidia to fully integrate OptiX features into Blender and OptiX accelerated CUDA rendering already is available alongside the existing CUDA and OpenCL implementations.

BMW with Multi-GPU Rendering for Blender & Cycles

When it comes to looking at GPU specifications, pay attention to the number of compute-cores (CUDA or OpenCL, depending on what brand you buy) and the amount of VRAM. Blender is just as hungry for VRAM as it is normal RAM, and a GPU with ample VRAM is worth paying extra for if you’re interested in sculpting, high poly modeling, or GPU rendering.

Over the past three years or so, NVIDIA’s dramatically improved the rendering performance in its stack. It was once exciting when the BMW project broke through the 60-second barrier, but now we have these new RTX GPUs cutting through the 30-second mark. The RTX 3090 comes closer to 20 seconds than it does 30 seconds. These new GPUs are really, really fast.

For all the reasons mentioned above, we definitely recommend:

  • Performance Pick: Nvidia RTX 3090 or RTX 3080
  • Value Recommend: Nvidia RTX 3060 Ti and RTX 3070
  • Budget Choice: Nvidia GTX Model

Multiple GPU support for Blender & Cycles

You’ll need a GPU with lots of CUDA Cores or Stream Processors for fast GPU Rendering and can add multiple GPUs for a near-linear increase of GPU Render Performance. Blender can take great advantage of multiple GPUs, delivering dramatic gains when a second card is added. 

Using more than one GPU will certainly speed up Cycles. In User preferences -> System -> Compute device, you can configure which Cuda devices to use. When rendering, each GPU will render one tile (following the settings on the performance tab). The more GPUs, the more tiles are rendered simultaneously, so dual GPUs will make a huge difference in Cycles by decreasing rendering time by almost one-half. 

In some cases of GPU rendering, mainly on previous versions of Blender 2.8, the whole system slowed down and became quite responsive. This issue has been fixed since version 2.8. Other than rendering, only one GPU will be used for many parallel tasks, both computing and rendering, which means that it does not even promptly process the displayed image on the interface of the operating system. Therefore, using multiple GPUs can resolve the slowdowns or lag problem, since the graphics processor will divide GPUs to render while the user can continue to use the PC for other things.

Render scaling of Blender & Cycles

In the chart above, it is clear to see that Cycles scales quite well when adding more GPUs. There’s no reason to stop at 4 GPUs out of fear of diminishing returns. Unless you’re typically rendering at samples less than 200, you’re better off with more than 4 video cards if your motherboard can handle it and you can afford the cards.

Second, scaling doesn’t need to be expensive. Let’s find out the solution we provide you in the next part.

NVlink Support

If you use less memory/video memory, Blender may take a long time to complete rendering or it appears CUDA error: Out of memory. For massive scenes requiring a lot of memory, it is an incredible resource. There are two ways to solve this problem:

  • Method 1: Smaller resolution textures

One way to reduce memory usage is by using smaller resolution textures. For example, 8k, 4k, 2k, and 1k image textures take up respectively 256MB, 64MB, 16MB and 4MB of memory.

  • Method 2: Use NVlink

One crucial field for architectural visualization that is about to get some significant updates is Cycles when working with multiple GPUs. Today, you could easily use more than one GPU for a rendering in Cycles, which would split the load. Each GPU could handle part of the rendering, which cut in half your render times in many cases.

The main problem with that solution in Cycles, when using “old” interfaces like SLi (for NVIDIA), is that your GPUs would still act as two different cards. A new technology from NVIDIA called NVLink is different and can make multiple cards work like a single one. That is also valid for available memory.

To connect the cards, you need an additional piece of hardware called a bridge. For instance, if you want to go for an overkill setup, a dual GeForce RTX 3090 connected with the NVLink, the final amount of memory available for your scenes in Blender will be 48 GB.

The Best Blender Cloud Rendering Service

iRender is a Professional GPU-Acceleration Cloud Rendering Service provider in HPC optimization for rendering tasks, CGI, VFX with over 20.000 customers and being appreciated in many global rankings (e.g. CGDirector, Lumion Official, Radarrender, InspirationTuts CAD, All3DP). Moreover, iRender officially joins the Blender Foundation’s Development Fund as a corporate Diamond member. Contributions from corporate members directly support core Blender development, for generally approved projects on blender.org.

There are a variety of servers from single- GPU 1x RTX 3090 to Multi-GPU 2/4/6 x RTX 3090, you can choose one server that is suitable for your demand and your software to get started. Besides, we offer clients the powerful processor Intel Xeon W-2245 @ 3.90GHz/AMD Ryzen Threadripper Pro 3955WX @ 3.90GHz and 512 GB – 1 TB SSD to load your scene quickly.

NVLink (Request) will help you increase the amount of VRAM to 48GB. This is a technology co-developed by Nvidia and IBM with the aim of expanding the data bandwidth between the GPU and CPU 5 to 12 times faster than the PCI Express interface. These servers are sure to satisfy Blender artists/ studios with very complex and large scenes.

You can start at 3.8 USD to use the beast RTX 3090. You can rest assured that our pricing schemes are the cheapest you have ever found, with rates three times lower than that of other solutions.  Besides our hard-to-beat prices, we also offer generous volume discounts, which provide you with credit bonuses depending on how much you top up. Moreover, you can always save up to 30% with our daily/weekly rental.

With all the outstanding advantages mentioned above, you must have found yourself the perfect choice for Blender & Cycles. We provide unmatched support tailored to your specific needs and goals. SPECIAL OFFER for this October: Get 50% BONUS for all newly registered users. Let’s CHECK IT OUT!

For further queries, please feel free to contact Ms. Jenny: [email protected]. Enjoy a trial now to stop wasting pointless hours of rendering. We do the rendering, the creativity is yours!

iRender – Happy Rendering!

Source: Blender.org

Related Posts

, , , , , , , , , , , , , , , , , , , , , , , , , , , , ,


Autodesk Maya
Autodesk 3DS Max
Cinema 4D
Daz Studio
Nvidia Iray
Unreal Engine
And many more…


iRender Core – GPU Render Engine
GPU HUB. – Decentralized GPU Computing
Chip Render Farm


Hotline: (+84) 912-785-500
Skype: iRender Support
Email: [email protected]
Address 1: 68 Circular Road #02-01, 049422, Singapore.
Address 2: No.22 Thanh Cong Street, Hanoi, Vietnam.

[email protected]