Real Render Times and Costs from Recorded iRender Tests
This page collects our multi-GPU render benchmarks: 17 recorded runs of Redshift, Octane and Blender Cycles on dedicated servers with 2, 4 or 8 NVIDIA RTX 5090 or RTX 4090 GPUs. Each row shows the server, frame count, render time and cost at list price, with a link to the video of the run and to the render farm page for that software.
Use the numbers to choose a starting plan, then render a few test frames of your own scene to confirm. 2x, 4x and 8x servers bill per minute, and your first top-up of $50 or more gets a 100% bonus.
Last updated: November 11, 2026.
How we measure
- Real servers, real plans. Every run uses a standard iRender plan (3S, 4S, 5S, 9S or 5i) on a bare-metal, single-tenant server with AMD Threadripper PRO and 256 GB ECC RAM.
- The renderer's own clock. Render time is the time reported by the renderer or the render queue in the video. Where a row says "billed", the figure is the full session on the invoice, including setup.
- Cost at list price. Render time multiplied by the plan's hourly price on the price page. 2x, 4x and 8x plans bill per minute; sessions of 3 hours or more get 10% off the whole session. The 100% first top-up bonus is not included.
- Cost per frame is the cost divided by the frame count, rounded to the cent.
- Different scenes. Each test uses its own scene and settings. Compare servers only where a row is marked as the same scene, or in the same-scene table further down.
Redshift benchmarks (Cinema 4D)
Five Redshift runs in Cinema 4D, from a 2x RTX 4090 server to an 8x RTX 4090 server. Redshift splits each frame across every GPU in the server, so 8x servers suit long sequences on a deadline.
Redshift render times and costs
Each row is one recorded run. Scenes differ between rows unless noted.
| Test scene | Server | Frames | Render time | Cost at list price | Cost per frame | Video | Service page |
|---|---|---|---|---|---|---|---|
| Cinema 4D animation1080p | 5i4x RTX 5090 | 150 | 1h 42m · 40.84 s/frame | ~$64.66 | ~$0.43 | Watch the test | C4D Redshift render farm → |
| Fluid and particle motion paintingCinema 4D 2025.3.3, 1080p | 9S8x RTX 4090 | 360 | 1h 18m 13s render · 1h 21m billed | $70.20 | ~$0.20 | Watch the test | C4D Redshift render farm → |
| Cinema 4D animation30 fps, 1080p | 4S2x RTX 4090 | 300 | ~2h 50m · 33.95 s/frame | ~$42.44 | ~$0.14 | Watch the test | C4D Redshift render farm → |
| Sailboat animation1080p, same scene as the Octane run | 5i4x RTX 5090 | 120 | 3h 19m | ~$113.43after the 10% 3h+ discount | ~$0.95 | Watch the test | C4D Redshift render farm → |
| Cinema 4D still sequence1080p, same scene as the Octane run | 5i4x RTX 5090 | one test scene | 27m 57s | ~$17.70 | n/a | Watch the test | C4D Redshift render farm → |
For plans, licensing (Prime customers get Cinema 4D and Redshift licenses on every server they run) and the full setup guide, see the Cinema 4D Redshift render farm page.
Watch the recorded Redshift runs behind this table:
Octane benchmarks (Cinema 4D)
Five OctaneRender runs in Cinema 4D. OctaneRender is pre-installed on every server and you sign in with your own OTOY license; an Octane Studio+ subscription has no GPU limit, so it uses every card on 2x, 4x and 8x servers.
Octane render times and costs
The two car drift rows use the same scene and settings; the RTX 4090 cost is estimated from its measured seconds per frame.
| Test scene | Server | Frames | Render time | Cost at list price | Cost per frame | Video | Service page |
|---|---|---|---|---|---|---|---|
| Car drift scene1080p, 800 max samples | 5i4x RTX 5090 | 300 | 1h 18m 24s · 15.6 s/frame | ~$49.65 | ~$0.17 | Watch the test | Octane render farm → |
| Car drift scenesame scene and settings | 5S4x RTX 4090 | 300 | 18.8 s/frame (about 1h 34m) | ~$47estimated from s/frame | ~$0.16 | Watch the test | Octane render farm → |
| Cinema 4D animation1080p | 9S8x RTX 4090 | 320 | 1h 02m render · 1h 16m session billed | ~$65.87 | ~$0.21 | Watch the test | Octane render farm → |
| Sailboat animation1080p | 5i4x RTX 5090 | 120 | 2h 01m | ~$76.63 | ~$0.64 | Watch the test | Octane render farm → |
| Cinema 4D still sequence1080p | 5i4x RTX 5090 | one test scene | 18m 13s | ~$11.54 | n/a | Watch the test | Octane render farm → |
Setup steps, licensing and plan advice are on the Octane render farm page.
Watch the recorded Octane runs behind this table:
Blender Cycles benchmarks
Seven Cycles runs, from a single RTX 4090 reference up to an 8x RTX 4090 server. Several runs use "one Blender per GPU": one Blender instance per card, each rendering its own frame range, which keeps every GPU busy on animations.
Cycles render times and costs
Per-frame tests show average seconds per frame without a full sequence cost. The 3S row is a single-GPU reference.
| Test scene | Server | Frames | Render time | Cost at list price | Cost per frame | Video | Service page |
|---|---|---|---|---|---|---|---|
| Cycles animation, 4K3840×2160, 512 samples, noise threshold 0.01, Blender 5.2.2 LTS, OptiX, one Blender per GPU | 5i4x RTX 5090 | 240 | 87 s/frame per GPU · about 1h 27m | ~$55.10 | ~$0.23 | Watch the test | Blender render farm → |
| Cycles animation, 4Ksame scene and settings | 5S4x RTX 4090 | 240 | 108 s/frame per GPU · about 1h 48m | ~$54 | ~$0.23 | Watch the test | Blender render farm → |
| Cycles animation1080p, 512 samples | 5i4x RTX 5090 | 120 | 72 s/frame · about 2h 24m | ~$91.20 | ~$0.76 | Watch the test | Blender render farm → |
| Cycles animation1080p | 9S8x RTX 4090 | 120 | 22.5 s/frame · 45 min | ~$39 | ~$0.33 | Watch the test | Blender render farm → |
| Interior scene, ~249,000 objects2K, 256 samples, one Blender per GPU | 5S4x RTX 4090 | 100 | 18.5 s/frame average | per-frame test | n/a | Watch the test | Blender render farm → |
| Same interior scene1080p, one Blender per GPU | 4S2x RTX 4090 | 100 | 11.5 s/frame average | per-frame test | n/a | Watch the test | Blender render farm → |
| Three production test scenessingle-GPU reference | 3S1x RTX 4090 | 3 scenes | 38.3 s/frame average | per-frame test | n/a | Watch the test | Blender render farm → |
Blender is free to run on as many servers as you like, with no render licenses. Plans and the one-Blender-per-GPU setup are on the Blender render farm page.
Watch the recorded Cycles runs behind this table:
Same scene, different hardware
These four pairs are the only fair comparisons on this page: the same scene and settings, changing only the GPU or the renderer. They are results for these scenes, not a general ranking.
Same-scene comparisons
Time difference is the share of render time saved by the faster run.
| Same-scene pair | Faster run | Slower run | Time difference | Cost at list price | Service page |
|---|---|---|---|---|---|
| Octane, car drift, 300 frames5i vs 5S | 5i · 4x RTX 509015.6 s/frame | 5S · 4x RTX 409018.8 s/frame | ~17% less time per frame | ~$49.65 vs ~$47 | Octane render farm → |
| Cycles, 4K, 240 frames5i vs 5S | 5i · 4x RTX 509087 s/frame per GPU | 5S · 4x RTX 4090108 s/frame per GPU | ~19% less time per frame | ~$55.10 vs ~$54 | Blender render farm → |
| Sailboat, 120 frames, 5iOctane vs Redshift | Octane2h 01m | Redshift3h 19m | ~39% less time | ~$76.63 vs ~$113.43 | Octane render farm → C4D Redshift render farm → |
| Still sequence, 5iOctane vs Redshift | Octane18m 13s | Redshift27m 57s | ~35% less time | ~$11.54 vs ~$17.70 | Octane render farm → C4D Redshift render farm → |
Both Octane-versus-Redshift pairs were recorded on the same 4x RTX 5090 server. Watch them side by side:
Cutting wall-clock time: split frames across servers
You can rent several servers at once and give each one a frame range. Because 2x, 4x and 8x plans bill per minute, the total cost stays close to a single-server run, while the wait drops roughly by the number of servers. Each extra server adds a few minutes for setup and first-frame compilation, which is why the estimates cost slightly more.
One server vs three servers
Single-server figures are measured; three-server figures are estimates that include a few minutes of setup per server.
| Job | One server (measured) | Three servers, frame ranges split (estimate) | Service page |
|---|---|---|---|
| Octane, 300 frames5i, 4x RTX 5090 | ~78 min · ~$49.65 | ~31 min · ~$59 (3 x ~31 min billed) | Octane render farm → |
| Redshift, 360 frames9S, 8x RTX 4090 | ~81 min · $70.20 | ~30 min · ~$76 | C4D Redshift render farm → |
| Cycles, 120 frames, 1080p5i, 4x RTX 5090 | ~2h 24m · ~$91.20 | ~53 min · ~$101 | Blender render farm → |
How to read these numbers
- VRAM is per GPU. Every GPU loads the whole scene, so the limit is 24 GB on RTX 4090 and 32 GB on RTX 5090 plans. More GPUs make a scene render faster; they do not make room for a bigger scene.
- Your scene will differ. Samples, resolution, volumes, hair and textures change render time far more than the server does. Use these runs to choose a starting plan, then time a short frame range of your own scene.
- 1x plans bill a minimum of one hour. The 3S and 3i plans bill one hour first, then per minute; 2x, 4x and 8x plans bill per minute from the start.
Benchmark your own scene on iRender
- Create an account and top up. Your first top-up of $50 or more gets a 100% bonus, which covers setup and test renders.
- Upload your project to iRender Drive. Transfer and storage are free, and you can upload before you start a server.
- Start a 2x, 4x or 8x server, open your scene and render a short frame range.
- Multiply your seconds per frame by your frame count to plan the full job, and split frame ranges across servers if the deadline is tight.
The iRender GPU app is an all-in-one desktop app: you manage your account, start and connect to servers, and transfer files from one place.
Download the iRender GPU appWindowsWindows 10 / 11macOSApple SiliconmacOSIntelLinuxUbuntu · DebianNo install? Launch the Web App →Frequently asked questions
Were these benchmarks run on the servers customers rent?
Yes. Every test ran on iRender GPU Cloud Workstation plans (3S, 4S, 5S, 9S, 5i) with the same hardware and hourly prices listed on the price page. Each row links to the video of the run.
Can I compare rows from different tables?
Only rows marked as the same scene. The tests use different scenes, frame counts and settings, so a fast row in one table does not mean that server or renderer is faster in general. The “Same scene, different hardware” table lists the fair comparisons.
How is the cost column calculated?
Render time multiplied by the plan’s list price per hour. 2x, 4x and 8x plans are billed per minute; 1x plans (3S, 3i) bill a minimum of one hour, then per minute. Sessions of 3 hours or more get 10% off the whole session, as in the Redshift sailboat row. Upload and setup time are not included unless the row says “billed”.
Is an RTX 5090 server worth it over an RTX 4090 server?
In both same-scene tests the 4x RTX 5090 server finished about 17–19% faster per frame, and the total cost came out about the same, because the 5i costs more per hour. The other reason to choose RTX 5090 is memory: 32 GB per GPU instead of 24 GB.
Is VRAM combined across the GPUs in a server?
No. Redshift, Octane and Cycles each load the scene on every GPU, so the largest scene a server can hold in VRAM is set by one card: 24 GB on RTX 4090 and 32 GB on RTX 5090 plans. Beyond that, out-of-core features use system RAM (256 GB ECC on every server), which is slower.
How can I benchmark my own scene?
Create an account, upload your project to iRender Drive for free before you start a server, then render a short frame range on a 2x, 4x or 8x plan, which bills per minute. Your first top-up of $50 or more gets a 100% bonus, which covers setup and test renders.
Sources: iRender benchmark videos on the iRender Official YouTube channel, linked in each row; plan prices from the iRender price page.
“We focus on people and the joy of creation!”
"iRender Farm’s green data center is powered by solar energy, as part of our natural and eco-friendly commitment to a sustainable future."
Multi-GPU Render Farm Insights
Master multi-GPU cloud rendering. Practical guides for Redshift, Octane and Blender Cycles on dedicated render servers with AMD Threadripper PRO, 256 GB ECC RAM and 2–8x RTX 4090/5090 GPUs on NVMe storage.
"We all know that talent is never enough but devotion is the only way to success"
