November 10, 2026 Maddie Quach

Hair and Fur on Karma XPU: Curves, Procedural Hair and GPU Memory

Hair & furRender groomed hair and fur on Karma XPU
From SideFX documentation

Hair is where Karma XPU is strongest and where GPU memory runs out first. Key points:

  • Curves: Karma XPU renders rounded tubes only; ribbons need Karma CPU
  • Topology: wrong vertex counts for the curve basis make curves disappear
  • Sampling: SideFX: hair resolves better with Path Traced convergence on XPU
  • Procedural: Houdini Procedural: Hair builds curves at render time and saves disk
  • Memory: every rendered curve sits on every GPU; KARMA_XPU_OPTIX_CURVE_OPT_LEVEL trades memory for speed

Creatures, characters and grass mean millions of thin curves. Karma XPU handles them well, and SideFX itself notes that scenes with lots of hair tend to resolve better with Karma XPU. They are also among the heaviest data in a scene: every rendered curve has to fit in each GPU, and cached hair can run to large files per frame that have to reach the render machine.

This guide covers how Karma renders curves, the topology rules that make curves vanish, the render-time Hair Procedural, sampling for fur, and how to keep curve memory under control on 32 GB cards.

Checked against the SideFX Houdini 22 documentation (Karma User Guide: Geometry, Karma XPU, Houdini Procedural: Hair), November 2026.

How Karma renders curves

Karma renders curves as rounded tubes by default and can also render ribbons and oriented ribbons, with the curve basis overridable at render time. On Karma XPU only rounded curves are supported, and the XPU page adds that disabling backface culling on curves is not supported. If a groom was set up as ribbons for Karma CPU, expect it to render as tubes on XPU.

iRender
How Karma Draws Curves
Schematic
Rounded tubedefault; Karma CPU and XPURibbonKarma CPU onlyOriented ribbonKarma CPU onlyKARMA_XPU_OPTIX_CURVE_OPT_LEVEL: 1 = lower memory, 2 = default, 3 = higher performance
works on XPUCPU only
Based on SideFX Karma User Guide: Geometry and Karma XPU page

Topology matters. USD BasisCurves are either linear or cubic, and SideFX warns that different curve types need different numbers of vertices for a strand to be valid; incorrect topologies can make curves disappear or look strange. The Karma guide recommends non-periodic linear curves for most cases and pinned cubic B-splines when you need smoother curves.

Sources: Karma User Guide: Geometry, Karma XPU.

Render-time hair with the Houdini Hair Procedural

Caching every render curve for every frame produces large files that take time to write, upload and load. The Houdini Procedural: Hair LOP, available since Houdini 19.5, provides a render-time hair solution that saves disk space. In Generate mode it creates new curves interpolated from existing guides; in Deform mode it deforms existing render curves using the guides. Its Animated Procedural option decides whether it is evaluated before each frame or only at the start of the render.

iRender
Cached Hair vs Render-Time Hair
Schematic
Cached hairHoudini Hair Proceduralevery curve, every frame, on diskupload + loadGPU: all curvesguides + skin onlyat renderGPU: all curveslarge caches to write and movesmall caches, curves built in huskBoth end the same on the GPU: the procedural saves disk and transfer, not VRAM.
cached curvesguideson the GPU
Schematic based on SideFX Hair Procedural and Karma XPU docs

The procedural runs inside husk at render time, so test it on the render machine: Python run by husk can check out an Engine license, according to SideFX’s licensing FAQ. The saving is in disk and upload size; the generated curves still occupy GPU memory during the render.

Sources: Houdini Procedural: Hair LOP, SideFX FAQ: husk licenses.

Sampling noisy fur

Thin curves are hard to sample: each pixel sees many strands, so fur stays noisy longer than surfaces. SideFX’s Karma guide gives direct advice: scenes with lots of hair or fur tend to resolve better with Path Traced convergence and with Karma XPU, and relying heavily on secondary samples to clean up noisy fur gets very expensive. Raise pixel samples and use the denoiser before adding secondary samples.

Shading also changes the cost. Houdini 21 updated the fur shader, and Houdini 21 lets Karma displace curve vertices when a displacement shader is applied to curves. Keep displacement on curves for close-ups, where it is visible.

Sources: Karma User Guide: Geometry, What’s new in Houdini 21: Karma.

Keeping curve memory inside one GPU

Karma XPU memory is not pooled, so a 4x RTX 5090 server gives a groom 32 GB per card, not 128 GB. Curve count, segments per curve and per-vertex attributes (width, colour, UVs) all multiply. Ways to bring a heavy groom down:

  • Reduce curve density for characters far from camera; generate full density only for close-ups.
  • Drop per-vertex attributes the shader does not read.
  • Use fewer segments on short fur.
  • Try the curve optimisation level below.
iRender
Lower curve memory on Karma XPU
houdini.env
1# houdini.env: trade curve memory for speed on XPU
2# 1 = lower memory, 2 = default, 3 = higher performance
3KARMA_XPU_OPTIX_CURVE_OPT_LEVEL = 1
Test one heavy frame at 1 and 2 and compare memory and timeSideFX Karma XPU docs

SideFX documents KARMA_XPU_OPTIX_CURVE_OPT_LEVEL as a trade between curve memory and performance: 1 favours lower memory, 2 is the default, 3 favours performance. Houdini 21 also added data deduplication, which reduces memory when identical geometry or primvars appear. For budgets per card, see how much VRAM Houdini GPU rendering needs.

Sources: Karma XPU, What’s new in Houdini 21: Karma.

SaaS or IaaS render farm for hair and fur?

CompareNeutral guide

Which farm type fits a groom-heavy shot

Both render curves; they differ in how easily you can test memory and the procedural on the target GPU.

SituationUsually a better fitWhy
Cached hair that already fits comfortably in 32 GBSaaS (managed farm)Upload caches and submit
Hair Procedural evaluated at render timeIaaS (remote workstation)Check the procedural and its license on the machine that renders
Groom close to the VRAM limitIaaS with 32 GB cardsCompare curve optimisation levels on the heaviest frame
Very large hair cachesIaaSUpload once and keep them between sessions
Short one-off renderSaaSNo setup time

Rule of thumb: if the groom is cached and fits on one card, a managed farm is the least work. With render-time procedurals or tight memory, a machine you control lets you test on the spot.

Rendering hair on an iRender server

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

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:

  1. Upload the project with guide or hair caches through iRender Drive.
  2. 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.
  3. Install your Houdini build. The Karma XPU kernels compile on the first render and the cache stays on your saved server image.
  4. Render the frame with the most hair on screen and check per-GPU memory and the device list.
  5. Render the range with husk or the USD Render ROP.

Hair and fur checklist

Use this as a final check before you start the full range on any machine.

ChecklistHair and fur checklist
0 / 7 done
Tip: tick each item as you set up the job; the ticks reset when the page reloads.

Frequently asked questions

1. Does Karma XPU render hair ribbons?

No. Karma XPU supports rounded curves only. Ribbons and oriented ribbons need Karma CPU.

2. Why do some hair curves disappear in Karma?

Each curve basis needs a minimum number of vertices per strand. Invalid topology can make curves disappear or render strangely.

3. What is the Houdini Hair Procedural?

A LOP that generates or deforms hair at render time from guides, so you store guides instead of every render curve.

4. How do I reduce noise in fur with Karma?

SideFX recommends Path Traced convergence with Karma XPU; relying on many secondary samples gets very expensive for fur.

5. How do I lower hair memory on Karma XPU?

Reduce curve density and attributes, and try KARMA_XPU_OPTIX_CURVE_OPT_LEVEL=1, which favours lower memory.

6. Do four GPUs share the memory of one groom?

No. Karma XPU memory is not pooled; every GPU holds all rendered curves.

7. Does Houdini 21 support displacement on curves in Karma?

Yes. When a displacement shader is applied to curves, Karma displaces the curve vertices, similar to point meshes.

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.

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).

Accelerate your production pipeline with high-performance IaaS render farm services and dedicated bare-metal workstations powered by 4x 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:
, , , , , , ,

Maddie Quach

Maddie Quach, from iRender's customer support team, has written more than 100 articles for the iRender blog on Maya, Houdini, Arnold, Karma XPU and Redshift.
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 →