Redshift for Houdini on Multiple GPUs: Device Selection and Houdini 22
Redshift for Houdini uses the GPUs listed in its preferences, an environment variable or an hbatch command. What to know for multi-GPU work:
- Devices: Compute Devices in the Redshift plugin options, saved in preferences.xml
- Per process: REDSHIFT_GPUDEVICES=0,1 overrides preferences when Redshift starts
- hbatch: Redshift_setGPU -s 011 picks GPUs for one session
- Scaling: Maxon: GPU subsets let one machine render several frames at once
- Houdini 22: supported from Redshift 2026.9 (22.0.368), later builds added 22.0.429
Redshift is the main third-party GPU renderer in many Houdini studios, used from the classic Redshift ROP and, in Solaris, as a Hydra delegate. On a multi-GPU machine it can render one frame on every card, or split the cards between several renders. Which GPUs it uses depends on three settings that override each other, and getting them wrong is a common reason a 4-GPU machine renders on one card.
This guide explains where Redshift for Houdini takes its GPU list from, how to pin a render to a subset of GPUs from the command line, when splitting GPUs helps, and which Redshift builds support Houdini 22.
Checked against the Maxon Redshift for Houdini documentation (Command-line Rendering and GPU Selection, Environment Variables, Redshift CPU) and Redshift release notes, November 2026.
Where Redshift takes its GPU list from
In Houdini, Redshift’s device list is the Compute Devices list in the Redshift plugin options, which lists the GPU and CPU hardware and controls which devices render. The selection is saved in preferences.xml in the Redshift local data path. Two things can override it:
- REDSHIFT_GPUDEVICES: Maxon documents it as a comma-separated list of GPU ordinals, for example
0,1, that overrides the preferences. It is read only when Redshift initializes, so set it before starting Houdini or hbatch. - Redshift_setGPU: an hscript command for hbatch sessions.
-s 011on a three-GPU machine disables the first GPU and enables the second and third. Maxon notes that command-line selection does not updatepreferences.xml.
To find device numbers, Maxon suggests checking the device list in preferences.xml. When Hybrid Rendering is on and a CPU is enabled, Redshift can also render on the CPU alongside the GPUs.
Sources: Redshift Environment Variables, Command-line Rendering and GPU Selection, Redshift CPU.
One render on all GPUs, or several renders on subsets
By default one Redshift render uses every enabled GPU for each frame. Maxon’s documentation points out that selecting only a subset of GPUs for a job is useful to render multiple frames simultaneously on a single machine for optimal scaling, and that render managers such as Deadline and Royal Render support this selection.
Splitting costs memory and start-up: each process loads the scene and its textures, and each GPU still holds its own copy. Test both layouts on a few representative frames and compare the time per frame. Our Redshift multi-GPU guide covers how Redshift scales across cards.
Two hbatch processes, each pinned to two GPUs and rendering half the range, keep all four cards busy on two frames at a time. Write one log per process and compare the time per frame with a single render on all four GPUs.
Sources: Command-line Rendering and GPU Selection, Redshift Environment Variables.
Redshift in Solaris
In Solaris, Redshift renders as a Hydra delegate, so the same LOP network can be rendered with Karma or Redshift and sent to husk. Maxon ships Solaris support per Houdini build; the release notes list Hydra support next to the Houdini plugin support for each version. Our guide to Redshift in Solaris LOPs covers the setup, and the Karma XPU vs Redshift Hydra comparison explains when each fits.
For GPU selection the same rules apply: REDSHIFT_GPUDEVICES set in the environment of the husk process decides which cards the delegate uses.
Sources: Redshift release notes.
husk’s -R option selects the delegate on the command line, so the same saved USD can be rendered with either engine for comparison.
Redshift and Houdini 22
Redshift 2026.9 (September 2026) added support for Houdini 22.0.368 and Solaris 22.0.368, with a USD 26.05 build for Houdini 22’s USD Procedural workflow, and updated OCIO to 2.5.2. Redshift 2027.0.1 (October 2026) added Houdini 22.0.429, and a later release added 22.0.459 while dropping 21.0.700. Each Redshift release supports a list of specific Houdini builds and drops older ones, so check the release notes for your exact build before installing on a render machine.
Keep the same Redshift version on every machine that renders a shot. A render machine on a newer Redshift can produce slightly different results and may not open scenes with the same plugin settings. For the wider version checklist, see Houdini 22 on a render farm.
Sources: Maxon: Redshift 2026.9 / 2027.0 release notes, Redshift release notes, Digital Production: Redshift 2026.9.
SaaS or IaaS render farm for Redshift in Houdini?
Which farm type fits Redshift for Houdini
Both can render Redshift; they differ in who controls versions and GPU layout.
| Situation | Usually a better fit | Why |
|---|---|---|
| Farm runs your exact Redshift and Houdini builds | SaaS (managed farm) | Submit and let the farm schedule GPUs |
| You want to split GPUs between frames yourself | IaaS (remote workstation) | Set REDSHIFT_GPUDEVICES per process and compare layouts |
| Specific Redshift build for your Houdini build | IaaS | Install exactly the versions you tested |
| Karma and Redshift in the same LOP network | IaaS | Both delegates on one machine with the same paths |
| Short one-off job | SaaS | No setup time if versions match |
Rule of thumb: if a managed farm matches your builds, it is the least work. If you rely on a specific build or a custom GPU split, a machine you control lets you set it up exactly.
Rendering Redshift for Houdini on an iRender server
On iRender you rent a whole Windows workstation, install Houdini and the Redshift build that matches it, and sign in with your own SideFX and Maxon licenses:
- Upload the project, textures and proxies through iRender Drive with relative paths.
- Start a multi-GPU server; Redshift renders on every GPU ticked in Compute Devices.
- Install your Houdini build and the matching Redshift for Houdini build.
- Check Compute Devices, or set REDSHIFT_GPUDEVICES per process if you split GPUs.
- Render one frame, check the log for the device list, then render the range.
Redshift for Houdini checklist
Use this as a final check before you start the full range on any machine.
Frequently asked questions
1. How do I choose which GPUs Redshift uses in Houdini?
Tick them in Compute Devices in the Redshift plugin options (saved in preferences.xml), or set REDSHIFT_GPUDEVICES before Redshift starts.
2. What does Redshift_setGPU do?
It is an hscript command for hbatch sessions that enables GPUs with a bitmask, for example -s 011. It does not change preferences.xml.
3. Is it faster to split GPUs between frames?
Maxon says GPU subsets let one machine render several frames at once for optimal scaling. Each process loads the scene, so test on your frames.
4. Does Redshift support Houdini 22?
Yes. Redshift 2026.9 added Houdini 22.0.368, and 2027.0.1 added 22.0.429.
5. When is REDSHIFT_GPUDEVICES read?
Only when Redshift initializes, so set it in the environment before starting Houdini, hbatch or husk.
6. Can Redshift use the CPU too?
Yes, with Hybrid Rendering enabled and a CPU ticked in Compute Devices.
7. How do I find Redshift’s GPU device numbers?
Check the device list in preferences.xml in the Redshift data path; each GPU is listed with its number.
8. Does Redshift_setGPU change my saved preferences?
No. Maxon notes that command-line GPU selection does not update preferences.xml.
Render your Houdini scenes on iRender
iRender is an IaaS GPU render farm: you rent a bare-metal server with 4x NVIDIA RTX 5090 (32 GB) GPUs, an AMD Ryzen™ Threadripper™ PRO CPU, 256 GB ECC RAM and a 2 TB NVMe SSD, connect by remote desktop and work on it like your own workstation, with full administrator access. You install Houdini and use your own SideFX license; your server image is saved between sessions, so Houdini, your packages and the Karma XPU kernel cache stay installed. For Karma XPU we recommend the 4-GPU servers.
Multi-GPU servers are billed per minute (1-GPU servers have a one-hour minimum), and sessions of 3 hours or more get 10% off. Our 4-GPU servers are liquid-cooled; in our own measurements the GPUs hold 60–70°C under sustained render load. Elite support is available 7 days a week by hotline, live chat, Skype, WhatsApp and email.
This video walks through rendering a Houdini Karma XPU scene on an iRender server in five steps:
New users get a 100% bonus on the first top-up (minimum $50). See the Karma XPU render farm page for server options, or download the iRender GPU app.
If Karma XPU renders are slow on your machine, iRender’s Karma XPU render farm gives you a dedicated remote server where you install Houdini, bring your own license and render on multiple RTX 5090 GPUs (for Karma XPU, 4 GPUs is usually the most efficient setup).


