Karma Displacement and Subdivision: Dicing Quality and GPU Memory
Karma dices subdivision and displaced surfaces into micropolygons at render time. The settings that decide quality and memory:
- Subdivision: a mesh subdivides only when its subdivisionScheme is set
- Dicing: Dicing Quality 1 ≈ one micropolygon per pixel
- Grazing: Dicing Z-Importance below 1 dices less at grazing angles
- Textures: load displacement maps as Raw, not sRGB
- Houdini 22: up to half the memory for subdivided and diced meshes on XPU
Displacement is the quickest way to add detail that holds up in close-ups: rock surfaces, terrain, skin pores, brick walls. It is also one of the quickest ways to fill GPU memory, because Karma turns every displaced or subdivided surface into micropolygons, and the number of micropolygons depends on how many pixels the surface covers.
This guide explains how Karma decides subdivision and dicing, how Dicing Quality and Dicing Z-Importance control the micropolygon count, how to load displacement textures, and what changed in Houdini 22 for memory on Karma XPU.
Checked against the SideFX Houdini 22 documentation (Karma User Guide: Geometry, Render Geometry Settings, Color management, What’s new in Karma), November 2026.
How Karma subdivides and dices
In USD, a mesh becomes a subdivision surface when its subdivisionScheme property is set to one of the supported schemes. Karma then dices it at render time, together with displaced surfaces, into micropolygons. The Dicing Quality parameter controls that resolution: SideFX defines a value of 1 as roughly one micropolygon per pixel.
Because dicing is measured in pixels, the same asset costs more memory in a close-up than in a wide shot, and more at 4K than at HD. Values above 1 add micropolygons the camera may not resolve; values below 1 save memory on background objects. A separate dicing camera can be set to control the frustum used for dicing, which keeps detail stable when the render camera moves fast.
Sources: Karma User Guide: Geometry.
Saving memory at grazing angles with Dicing Z-Importance
Large flat surfaces seen at a low angle (ground planes, walls along a corridor, terrain toward the horizon) cover many pixels but show little detail per pixel. Dicing Z-Importance on the Render Geometry Settings LOP addresses that: set below 1, it reduces the dicing rate based on the angle between each face and the dicing camera, to reduce memory use.
Apply it per object on the surfaces that need it, not globally: a character’s face turned away from the camera still needs its detail on the silhouette. Houdini 22 lists Dicing Z-Importance with the note that lower values decrease memory consumption.
Sources: Render Geometry Settings LOP, What’s new in Houdini 22: Karma.
Displacement textures and colour space
A displacement map is data, not colour. Houdini’s colour management page says it plainly for the USD UV Texture VOP: Raw loads the data unchanged and is the setting for displacement, bump and roughness maps, while sRGB converts the incoming data to scene linear and Auto follows the OCIO file rules. Houdini’s default rules treat JPG, PNG and TIFF as sRGB, so an 8-bit displacement map left on Auto is converted and displaces wrongly.
Settings that decide displacement detail and memory
From the Houdini 22 Karma geometry guide and Render Geometry Settings help.
| Setting | What it does | When to change it |
|---|---|---|
| subdivisionScheme | Marks a mesh as a subdivision surface | Off for meshes already dense enough |
| Dicing Quality | 1 ≈ one micropolygon per pixel for subdivision and displaced surfaces | Lower for background objects; raise only if displacement looks soft |
| Dicing camera | A separate camera that controls the frustum used for dicing | Keep dicing stable on moving cameras or across shots |
| Dicing Z-Importance | Below 1 reduces dicing by the angle between face and dicing camera, to reduce memory | Large grounds and walls seen at grazing angles |
| Displacement texture colour space | Raw loads data unchanged; sRGB converts | Displacement maps should load as Raw |
Use 16-bit or 32-bit EXR or TIFF for displacement maps where possible, and set the texture node to Raw. On another machine, a different OCIO setup changes how Auto resolves; Raw does not depend on it. Our guide to OCIO and ACES in Houdini covers the rest of the colour setup.
Sources: Color management in Houdini.
Memory on Karma XPU and what Houdini 22 changed
Diced geometry is built for each GPU, because Karma XPU memory is not pooled. A shot that fits at HD can run out of memory at 4K with the same dicing settings, since twice the resolution in each direction means roughly four times the pixels to dice for. Check per-GPU memory on the frame where the displaced asset is largest on screen.
Houdini 22 helps directly: SideFX states that memory usage for subdivided and diced meshes on Karma XPU is reduced by up to half. Houdini 21 added data deduplication for identical geometry and primvars. If a scene ran out of memory on Houdini 21, it is worth re-testing on Houdini 22 before reducing detail; see our guide to moving to Houdini 22 on a render farm.
Sources: What’s new in Houdini 22: Karma, What’s new in Houdini 21: Karma, Karma XPU.
4K has exactly four times the pixels of 1920 × 1080, so a displaced close-up tested at HD needs re-checking at delivery resolution before the full render.
SaaS or IaaS render farm for displacement-heavy shots?
Which farm type fits a displacement-heavy shot
Both render displacement; they differ in how easily you can tune dicing against memory.
| Situation | Usually a better fit | Why |
|---|---|---|
| Dicing already tuned, scene fits in 32 GB | SaaS (managed farm) | Upload and submit |
| Close-ups near the VRAM limit | IaaS with 32 GB cards | Lower dicing or Z-Importance per object and re-test on the spot |
| Delivery resolution higher than your test renders | IaaS | Test the final resolution on the target GPU first |
| Testing Houdini 22 memory savings | IaaS | Install both versions and compare on one frame |
| Short one-off render | SaaS | No setup time |
Rule of thumb: if dicing is tuned at final resolution and fits on one card, a managed farm is the least work. While you are still balancing detail against memory, a machine you control saves round trips.
Rendering displacement on an iRender server
On iRender you rent a whole Windows workstation, install your own Houdini build and sign in with your own SideFX license, so the scene opens exactly as on your desk:
- Upload the project and displacement textures through iRender Drive with relative paths.
- Start a 4x RTX 5090 server (32 GB per card), which we recommend for Karma XPU; 8x servers also render Karma XPU, but it does not get proportionally faster.
- Install your Houdini build. The Karma XPU kernels compile on the first render and the cache stays on your saved server image.
- Render the closest frame at final resolution and check per-GPU memory in the render stats.
- Render the range with husk or the USD Render ROP.
Displacement checklist
Use this as a final check before you start the full range on any machine.
Frequently asked questions
1. What does Dicing Quality mean in Karma?
It sets how finely Karma dices subdivision and displaced surfaces. A value of 1 means roughly one micropolygon per pixel.
2. How do I make a mesh subdivide in Karma?
Set its USD subdivisionScheme property to a supported scheme. Meshes without it render as polygons.
3. How do I reduce displacement memory in Karma?
Lower Dicing Quality on background objects, set Dicing Z-Importance below 1 on surfaces seen at grazing angles, and test on Houdini 22, which cut subdivided-mesh memory on XPU.
4. Which colour space should a displacement map use?
Raw. It loads the data unchanged; sRGB and Auto can convert the values.
5. Why does my displaced scene run out of memory at 4K but not at HD?
Dicing is measured in pixels, so higher resolutions create more micropolygons, on every GPU with Karma XPU.
6. Did Houdini 22 reduce displacement memory?
SideFX states memory for subdivided and diced meshes on Karma XPU is reduced by up to half in Houdini 22.
7. What is a dicing camera?
A separate camera that controls the frustum used for dicing, so detail stays stable when the render camera moves.
Render your Houdini Karma XPU 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.
Sources: Karma User Guide: Geometry, Render Geometry Settings LOP, Color management in Houdini, What’s new in Houdini 22: Karma, What’s new in Houdini 21: Karma, Karma XPU. Checked November 2026.
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).


