October 19, 2026 Kath Nguyen

Redshift Multi-GPU Rendering in Cinema 4D: How It Scales, VRAM and When More GPUs Pay Off

Executive SummaryKey Takeaways: Redshift Multi-GPU Rendering in Cinema 4D
Scaling • VRAM • Frames vs GPUs

Redshift renders faster with more GPUs, but only speed is shared, not memory. The main points from Maxon's documentation:

  • Speed is shared: All GPUs in a machine can work on the same frame by splitting it into buckets; Maxon notes the gain can be non-linear.
  • Memory is not: Maxon support: Redshift does not combine VRAM across GPUs. Four 32 GB cards still mean a 32 GB in-VRAM limit.
  • Two modes: Render each frame on all GPUs, or render several frames at once on subsets of GPUs; long, light animations often scale better the second way.
  • Measured jobs: On iRender, 150 frames rendered in 1h 42m on 4x RTX 5090 and 360 frames in 1h 18m on 8x RTX 4090 (two different scenes).
  • On iRender: 2x, 4x and 8x RTX 5090 or RTX 4090 servers with Cinema 4D and Redshift pre-installed; split long sequences across several servers for deadlines.

Adding GPUs is the fastest way to speed up Redshift, but it does not work the way many people expect. More cards share the work of each frame; they do not add up their memory. This guide explains how Redshift uses several GPUs, what limits scaling, and how to choose between 2, 4 and 8 GPUs, based on Maxon’s documentation and recorded iRender jobs.

Checked against the Cinema 4D and Redshift documentation, November 2026.

1. How does Redshift use multiple GPUs?

Redshift splits each frame into buckets and hands them to the available GPUs, so all cards work on the same frame (Maxon: hardware recommendations). The scene itself is not split: each GPU needs its own copy in its own memory.

iRender
How Redshift uses several GPUs
speed is shared, memory is not
ONE FRAME, FOUR GPUs: BUCKETS ARE SHARED, THE SCENE IS COPIEDThe frame: buckets split across GPUseach GPU renders different buckets of the same frameThe memory: every GPU holds the whole sceneGPU 1full scene copy32 GBGPU 2full scene copy32 GBGPU 3full scene copy32 GBGPU 4full scene copy32 GB4 × 32 GB is still a 32 GB scene limit, not 128 GB✓ More GPUs = more speed per frame; more VRAM per GPU = bigger scenesMaxon support: Redshift does not combine VRAM when using multiple GPUs; each GPU can only use its own memory
GPU 1GPU 2GPU 3GPU 4
Based on Maxon support articles on multi-GPU rendering

How to read the figure. On the left, the frame is shared out bucket by bucket, which is why more GPUs make each frame faster. On the right, each card holds the whole scene, which is why four 32 GB GPUs still give a 32 GB scene limit. Maxon states it directly: Redshift does not combine VRAM when using multiple GPUs (Maxon support).

ModesRedshift multi-GPU

Two ways to use several GPUs with Redshift

From Maxon's CPU, GPU and hardware recommendations for Redshift.

ModeHow it worksBest for
All GPUs on each frameRedshift splits the frame into buckets and every GPU renders part of it. Maxon notes the gain can be non-linear.Heavy frames, stills, look-dev and test renders
Several frames at onceSeparate render processes, each using a subset of the GPUs, render different frames in parallel.Long animations of light frames where per-frame overhead dominates
Several serversSplit the frame range across machines; each renders its own frames.Deadlines; turnaround drops roughly with the number of servers
MemoryIn every mode, each GPU needs enough VRAM for the scene; data that does not fit goes out-of-core and renders slower.

A note on sources about VRAM: Maxon’s support article states that Redshift does not combine the VRAM when using multiple GPUs. Maxon’s 3D rendering product page also lists “multi-GPU memory pooling” and says Redshift shares out-of-core memory across multiple cards. This article follows the support article: plan for every GPU holding the scene in its own VRAM. Check Maxon’s latest documentation for your Redshift version.

2. Why doesn't Redshift scale perfectly with more GPUs?

Part of every frame is work the GPUs cannot share, and some settings affect how evenly work is spread. Maxon notes that rendering a single frame on all GPUs can give a non-linear gain, and that large buckets can reduce multi-GPU performance unless the image is high resolution (Maxon: System settings).

iRender
What limits Redshift multi-GPU scaling
REDSHIFT
Time for one frame (illustrative)1 GPUfixed partbucket rendering, shared between GPUs2 GPUsbuckets halve4 GPUsa quarter each8 GPUsthe fixed part now dominatesScene export, data upload and writing the image do not shrink with more GPUsThe larger the fixed part, the less each extra GPU helps; caching heavy deformers and simulations shrinks it
Fixed per frameShared bucket work
Proportions are illustrative, not measured

How to read the figure. Each bar is one frame on 2, 4 and 8 GPUs. The fixed part (scene export, data upload and writing the image) stays the same length, while the shared bucket-rendering part halves each time the GPU count doubles. The fixed part therefore becomes a larger share of the frame, and each extra GPU helps less.

The table lists each limiting factor and what to do about it.

Scalingwhat limits it

What decides how well Redshift scales

From Maxon documentation; check each factor on your own scene.

FactorWhy it mattersWhat to do
Scene preparation timeExporting and preparing the scene happens before the GPUs start each frame and does not get faster with more GPUs.Check the per-frame log; cache heavy deformers and simulations
Bucket sizeMaxon notes large buckets can reduce performance with multiple GPUs unless the frame is high resolution; default is 128x128.Keep the default unless you have a reason
VRAM per GPUVRAM is not combined across GPUs; out-of-core data is slower.Watch the memory in the Feedback Display
Mixed GPUsEach card must still hold the scene, so the smallest card sets the in-VRAM limit.Use matching GPUs for heavy scenes
NoteMaxon recommends a stronger CPU when driving four or more GPUs.

3. How to choose between 2, 4 and 8 GPUs (step by step)

Settings checklist (copy or follow line by line):

Settings checklist
1. Does the scene fit one GPU?
Redshift Feedback Display → geometry + texture memory
> 24 GB → use 32 GB RTX 5090 GPUs
2. Time 2–3 representative frames
2 GPUs (4S / 4i) → seconds per frame = A
4 or 8 GPUs (5i / 9i) → seconds per frame = B
speed-up = A ÷ B // 2× GPUs giving far less than 2× = high per-frame overhead
3. Cost per frame = seconds per frame ÷ 3600 × hourly price
4. Long, light sequences → several frames at once (GPU subsets or servers)
5. Full job → best cost per frame × frame count

Step 1: Check the scene fits one GPU

Open the Feedback Display and watch geometry and texture memory (Maxon: Memory). If the scene does not fit a 24 GB card, use 32 GB RTX 5090 GPUs.

Step 2: Time a frame on 2 GPUs

Render 2 or 3 representative frames and note the seconds per frame.

Step 3: Time the same frames on 4 or 8 GPUs

Compare seconds per frame and cost per frame. If doubling GPUs gives far less than double speed, per-frame overhead is high.

Step 4: Try several frames at once

For long, light sequences, render frame ranges in parallel on subsets of GPUs or on several servers.

Step 5: Plan the full job

Multiply the best cost per frame by your frame count. Our cost guide shows the calculation.

iRender
Choosing 2, 4 or 8 GPUs for Redshift: five steps
D1
Redshift Feedback DisplayGeometry + texture memoryfits a 24 GB card?1Test frames on 2 GPUsSeconds per frameA2Same frames on 4 or 8 GPUsSeconds per frameB (speed-up = A ÷ B)3Cost per framecompareSeveral frames at once-redshift-gpu 0 -redshift-gpu 1 (job 1)4-redshift-gpu 2 -redshift-gpu 3 (job 2)Plan the jobbest cost per frame × frame count5
Step to follow
Mockup based on Maxon Redshift documentation

How to read the figure. The badges follow Steps 1 to 5 above. Badge 1 is the memory check in the Feedback Display, badges 2 and 3 are the timings on 2 GPUs and on 4 or 8 GPUs, with speed-up and cost per frame, badge 4 is two jobs on one server chosen with -redshift-gpu, and badge 5 is the plan from the best cost per frame.

Splitting one server into two render jobs

Maxon’s Redshift documentation shows how to choose GPUs for a Cinema 4D command-line render with -redshift-gpu, repeated once per device (Maxon: Command-line Rendering and GPU Selection). The example runs two jobs on a 4-GPU server, each on two GPUs and half of a 200-frame range.

iRender
Two Redshift jobs on one 4-GPU server
cmd
1:: GPU ids: the AllComputeDevices line in Redshift's preferences.xml
2:: e.g. value="0:NVIDIA GeForce RTX 5090,1:...,2:...,3:..."
3set C4D="C:\Program Files\Maxon Cinema 4D 2026\Commandline.exe"
4set SCENE="D:\project\shot010.c4d"
5
6:: job 1: GPUs 0 and 1, frames 0-99
7start "job1" %C4D% -redshift-gpu 0 -redshift-gpu 1 -render %SCENE% -frame 0 99 1
8
9:: job 2: GPUs 2 and 3, frames 100-199
10start "job2" %C4D% -redshift-gpu 2 -redshift-gpu 3 -render %SCENE% -frame 100 199 1
Edit the Cinema 4D path, scene and frame rangesMaxon: Command-line Rendering and GPU Selection · Cinema 4D and the Command Line

Maxon notes that GPUs chosen on the command line are not written to the Redshift preferences, so Cinema 4D opened normally keeps the devices ticked in the System tab. Maxon support also notes that each machine running the Commandline renderer with Redshift GPU needs its own Redshift license (Maxon: Redshift licenses and render nodes).

iRender
Picking the GPU count: measure on your own scene
CHOOSE
s / frameGPUs in the server2 GPUs4 GPUs8 GPUsABperfect scaling1Fits 24 GB?if not, use 32 GB RTX 50902Time 2 GPUs (A), then 4 or 8 (B)speed-up = A ÷ B3Far less than 2× per doubling?run frame ranges on several serversCost per frame = seconds per frame ÷ 3,600 × hourly price; multiply by your frame count
Your measured framesPerfect scaling
Curve is illustrative; use your own timings

How to read the figure. Compare your measured curve with perfect scaling on the chart of seconds per frame for 2, 4 and 8 GPUs: the measured line flattens. The notes beside it repeat the test: check the scene fits 24 GB or use a 32 GB RTX 5090, time 2 GPUs (A) and 4 or 8 GPUs (B), divide A by B, and if doubling GPUs gives far less than double speed, render frame ranges on several servers.

One quick test tells you which way your scene leans:

Quick check: if 8 GPUs are barely faster than 4 on your scene, split the frame range across two 4-GPU servers instead; per-frame preparation is likely dominating.

Checklist: choosing the GPU count

Tick each line as you test, so the full job runs on the server size you measured.

ChecklistRedshift multi-GPU checklist
0 / 7 done
Tip: tick each item as you set up the job; the ticks reset when the page reloads.

4. How fast are real Redshift jobs on multi-GPU servers?

Recorded iRender jobs: 150 frames rendered in 1h 42m on 4x RTX 5090 (40.84 s per frame), and 360 frames in 1h 18m on 8x RTX 4090. They are different scenes, so they show real job times rather than a scaling curve. Full results are in our Redshift benchmark.

Controlled test: 1x vs 4x RTX 4090 on the same scenes

To measure scaling directly, we timed the same Cinema 4D Redshift scenes on one RTX 4090 and on four, on the same base machine (AMD Ryzen Threadripper PRO 3955WX, 256 GB RAM), at 1920×1080 and 1024 samples:

  • Simple scene, adaptive threshold 0.005: 36 s on 1x RTX 4090, 12 s on 4x RTX 4090, so 3.0× faster.
  • Complex scene, threshold 0.002: 4 min 19 s on 1x, 1 min 19 s on 4x, so about 3.3× faster.

Neither result reaches 4×. Splitting a frame across GPUs has overhead, the CPU, RAM or disk can become the limit, and a light scene may not keep four GPUs fully busy. In the test, the single card’s usage swung between 50% and 90% on the simple scene and stayed around 80–99% on the complex one, which is why the heavier scene scaled better.

Recorded Redshift jobs on iRender multi-GPU servers.

5. Multi-GPU server or a render farm that splits frames: which fits?

A SaaS render farm usually spreads frames across many machines, so turnaround for a finished scene can be short, and you do not manage hardware. An IaaS render farm such as iRender gives you a whole multi-GPU server: every GPU works on each frame, you can test frames first, and you can rent several servers to split a long sequence. Choose by what the job needs: final scene and shortest turnaround, or testing and control on the render hardware.

On iRender you rent a whole server with 2x, 4x or 8x RTX 5090 or RTX 4090 GPUs, with Cinema 4D and Redshift pre-installed; Prime members get activated Cinema 4D and Redshift licenses. Extra render nodes need their own Redshift license according to Maxon, so check licensing before splitting a job across machines. Real render times are in our Redshift benchmark, and plans are on the Cinema 4D Redshift render farm page.

iRender
How iRender works: your computer controls your own multi-GPU server
IaaS render farm
Your PC or MacNo powerful GPU neededRemote desktopLive screen1:1 like your PCYOUR OWN GPU SERVERbare-metal · not sharedGPUGPUGPUGPUGPUGPUGPUGPU1, 2, 4 or 8 × RTX 5090 (32 GB) or RTX 4090 (24 GB)Threadripper PRO · 256 GB ECC RAM · 2 TB NVMe SSDFull admin access · your setup is saved between sessions1Register, top upFirst $50+ top-up doubled2Upload your projectiRender Drive, free3Start and connectRemote desktop, any OS4Render, shut downBilling stops, per minute✓ Your workstation, just more GPUs: no queue to wait, no farm plugin, no job submission, no settings changed
Your mouse and keyboardServer screen streamed back
Schematic, not to scale

How to read the figure. The top row is the setup: your own PC or Mac connects by remote desktop to a whole bare-metal server with 1, 2, 4 or 8 GPUs, so you open your Cinema 4D and Redshift project there exactly as at home. The numbered row is the order of a session: register and top up, upload to iRender Drive before the server runs, start and connect, render a test frame and the job, then shut down so billing stops.

6. Redshift multi-GPU FAQ

Does Redshift use multiple GPUs?
Yes. Redshift can use all GPUs in a machine to render each frame, splitting it into buckets. It can also render several frames at once, each on a subset of GPUs.

Does Redshift combine VRAM from multiple GPUs?
No. According to Maxon support, Redshift does not combine VRAM when using multiple GPUs; each GPU can only use its own memory. Data that does not fit is handled out-of-core, which is slower.

Is 8 GPUs twice as fast as 4 GPUs in Redshift?
Not always. Maxon notes that rendering a single frame with all GPUs can give a non-linear gain, and the per-frame scene preparation does not speed up with more GPUs. Test your own scene on both and compare seconds per frame.

Should I render one frame on all GPUs or several frames at once?
Heavy frames usually benefit from all GPUs per frame. Long animations of light frames, where scene preparation is a large share of each frame, can finish sooner by rendering several frames at once on subsets of GPUs.

Can I mix an RTX 5090 and an RTX 4090 for Redshift?
Redshift can use both, but each GPU must hold the scene, so the smaller card’s 24 GB sets the in-VRAM limit for scenes rendered on both.

Do I need a separate Redshift license for each machine?
According to Maxon, the Redshift GPU license in a Cinema 4D subscription covers your main workstation; extra render nodes need their own Redshift license to render on the GPU.

Accelerate your production pipeline with high-performance IaaS render farm services and dedicated bare-metal workstations powered by up to 8x NVIDIA GeForce RTX 5090 (32GB GDDR7), high-frequency AMD Ryzen™ Threadripper™ PRO processors, 256GB ECC RAM, and dedicated 2TB NVMe SSD storage tailored for your pipeline:
, , , , , , ,

Kath Nguyen

Kath Nguyen works in customer support at iRender, helping artists move Blender and Cinema 4D projects to multi-GPU servers. She has written more than 140 articles for the iRender blog.
Contact

iRENDER GPU

Render Farm Pricing
Quick Start
Multi-GPU Benchmarks
iRender GPU App
Blog
Documentation
The Render Farm
The Company
Terms & Conditions

INTEGRATIONS

Autodesk Maya
Blender
Cinema 4D
Houdini
Karma XPU
Lumion
KeyShot
Unreal Engine
Twinmotion
Redshift
Octane
And many more…

iRENDER TEAM

Elite support 7 days a week
SATURDAY – SUNDAY: 6:00 AM – 11:59 PM
(UTC+7)
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.

Contact
이 페이지는 한국어로도 제공됩니다. 한국어로 보기 →