November 17, 2026 Maddie Quach

Houdini TOPs for Rendering: USD Render Node, Batches and Resume

TOPs / PDGRender a sequence through TOPs
From SideFX documentation

TOPs turns a frame range into work items, sends each one to husk through a scheduler, and skips frames that already exist. The settings that matter most:

  • Node: USD Render Scene renders a LOP network or a .usd file
  • Batches: Frames per Batch 1 = one job per frame; higher values group frames
  • Resume: Cache Mode Automatic skips frames already on disk
  • Scheduler: Local Scheduler: one Karma XPU job at a time, retries on failure
  • Debug: Create Husk Command Attribute shows the exact command each job ran

A ROP render is one long process: if it stops at frame 63, you have to work out which frames are missing and start again. TOPs (PDG) treats every frame as a separate work item with its own status, log and output file, so a scheduler can retry failures, skip finished frames and run the same network on a workstation, on HQueue or on Deadline.

This guide covers the TOP nodes that render USD with husk, how work items and batches are formed, Local Scheduler settings for a multi-GPU machine, and how to resume a sequence. It is the TOPs companion to our husk command-line guide.

Checked against the SideFX Houdini 22 node help for USD Render, USD Render Scene and Local Scheduler, November 2026.

Which TOP node renders USD

Two TOP nodes render USD, and both run the Houdini husk executable for each job:

  • USD Render Scene can render a LOP network in the current scene or a .usd file on disk, and reads the output paths from the USD hierarchy. With Input USD = LOP Node it writes the network to disk first.
  • USD Render Files only renders USD files that already exist, for example files written upstream.

Karma is the default delegate; the Render Delegate parameter selects another Hydra client. Both nodes expose the usual render settings prim, camera and resolution overrides.

iRender
How a TOPs Render Job Reaches husk
Schematic
LOP network/stage/OUTUSD Render SceneTOP nodeframe 1001frame 1002 ✓ doneframe 1003 ✓ doneframe 1004frame 1005work itemsLocal Schedulerslots · retries · RAMhuskone job eachCache Mode = Automatic: frames already on diskare marked cooked and never sent to huskEach job runs the husk executable;the full command can be stored as an attribute
data flowskipped (cached)scheduler
Schematic based on SideFX TOP node docs

The diagram shows the part that saves the most time: with Cache Mode set to Automatic, a work item whose expected output already exists on disk is marked as cooked without being scheduled. Rerunning the network after a crash therefore renders only what is missing.

Sources: USD Render Scene, USD Render Files.

Work items and batches

Evaluate Using decides how work items are created: Single Frame makes one work item per upstream item, Frame Range uses a start, end and step, and Custom Range takes a value pattern. By default each work item is cooked as a separate job.

iRender
Work Items vs Batches for 8 Frames
Schematic
Frames per Batch = 11 job1 job1 job1 job1 job1 job1 job1 job8 jobs · each job starts its own husk processFrames per Batch = 41 job1 job2 jobs · one husk process renders 4 framesAll Frames in One Batch1 job1 job · fewest process starts, but a retry covers the whole batch
scheduled jobframe
Schematic based on SideFX USD Render TOP docs

Frames per Batch groups work items so one husk process renders several frames, and All Frames in One Batch sends the whole range as a single job. Fewer jobs mean fewer process starts, but a retry then covers the whole batch, and SideFX notes a limitation for batched renders: husk cannot evaluate work item attributes in file paths when it renders several frames. For Karma XPU sequences, one frame per job is the safest starting point; try small batches only if process start-up is a large share of each frame.

Sources: USD Render TOP.

Scheduling renders on one multi-GPU workstation

The Local Scheduler runs work items on the machine Houdini is running on. Its defaults are tuned for CPU work such as simulations, where several jobs run side by side. A Karma XPU job already uses every GPU it can see, and each GPU holds its own copy of the scene, so two renders on the same GPUs compete for the same VRAM.

ReferenceLocal Scheduler

Local Scheduler settings that matter for GPU renders

All parameter names are from the Houdini 22 Local Scheduler help.

ParameterWhat it doesFor Karma XPU jobs
Total SlotsNumber of slots the scheduler can use at onceKeep render jobs to one at a time: each XPU job already uses every GPU
Slots per Work ItemHow many slots one job consumesAlternative way to stop two renders sharing the GPUs
SingleRuns only one work item at a timeSimplest switch for a render-only network
On Task Failure / Maximum RetriesReport, warn, retry or ignore a failed jobRetry once or twice to ride out a driver hiccup
Minimum Available MemoryRAM required before a job startsAvoids starting a frame the machine cannot hold
Maximum MemoryRAM cap per work item, then mark failed or succeededStops one runaway frame from freezing the machine
Environment VariablesExtra variables for every work itemPass KARMA_XPU_* settings to all jobs
Why one job at a time: Karma XPU memory is not pooled, so two renders on the same GPUs each need room for a full copy of the scene.

Set the scheduler so only one render job runs at a time, add a retry, and set a memory cap below the machine RAM. On a farm, the HQueue, Deadline and Tractor schedulers apply the same idea per render node.

iRender
A starting setup for Karma XPU on one 4-GPU machine
settings
1# USD Render Scene TOP
2Input USD LOP Node (/stage/OUT)
3Evaluate Using Frame Range 1001 1100 1
4Frames per Batch 1
5Cache Mode Automatic
6Checkpoint Interval on, 600 s
7Create Husk Command on (stores @huskcommand)
8Extra Husk Arguments -V 2e --make-output-path
9
10# Local Scheduler
11Total Slots Custom, 1 render job at a time
12On Task Failure Retry Task, Maximum Retries 2
13Maximum Memory on, set below the machine RAM
Parameter names as in the Houdini 22 node helpSideFX: USD Render Scene, USD Render, Local Scheduler TOPs

Put Karma XPU environment variables, such as KARMA_XPU_DISABLE_EMBREE_DEVICE, in the scheduler’s Environment Variables list so every job gets them. Variables are documented on the SideFX Karma XPU page.

Sources: Local Scheduler, Karma XPU.

Checkpoints, time limits and resuming a sequence

Three settings on the USD Render node make long sequences safer:

  • Cache Mode = Automatic: finished frames are never rendered twice. Use Write Files only when you want to overwrite everything.
  • Checkpoint Interval: writes a checkpoint file and a partial image every N seconds, so a long frame can be inspected and, for Karma, picked up again instead of starting from zero.
  • Time Limit: cancels a frame after N seconds; No Partial Results on Timeout decides whether the unfinished image is kept.

If you drive husk directly instead, --skip-existing-frames gives the same resume behaviour as Cache Mode.

Sources: USD Render TOP, husk.

Finding out what a job actually ran

When a frame fails on a farm but works locally, compare the commands. Turn on Create Husk Command Attribute and each work item gets a huskcommand attribute with the full command its job executed; copy it into a terminal on the render machine to reproduce the error. The outputimage attribute holds the output path pattern.

Extra Husk Arguments passes flags the node has no parameter for, for example -V 2e for elapsed times in the log. Command Override runs a wrapper script instead of husk and passes it the generated command, which is useful to set per-job environment variables. For the Karma XPU checks to run before any of this, see the Karma XPU pre-flight checklist.

Sources: USD Render TOP.

SaaS or IaaS render farm for a TOPs render network?

CompareNeutral guide

Which farm type fits a TOPs workflow

Both can render the frames; they differ in whether your TOP network runs as you built it.

SituationUsually a better fitWhy
Plain frame range, no wedges or custom stepsSaaS (managed farm)Submit the scene; the farm splits and schedules frames
Network with wedges, file checks or post-render stepsIaaS (remote workstation)Your TOP network runs unchanged, including custom Python or HDAs
You want to tune scheduler slots, retries and memory capsIaaSYou edit the Local Scheduler on the machine that renders
Debugging a frame that fails only on another machineIaaSYou can copy @huskcommand into a terminal on that machine
Small one-off renderSaaSNo setup time

Rule of thumb: a managed farm suits a straight frame range. If your network does more than render (wedges, checks, conversions), a machine you control runs it exactly as you built it.

Running a TOPs render on an iRender server

iRender
Works 1:1 like your own PC: we keep your setup, you just get more GPUs
IaaS render farm
Your workstationscene.hip · .usd1 GPUYour iRender serverscene.hip · .usdup to 8 GPUs=same setupWindows or Ubuntu desktopYour software, in the version you chooseYour plugins and scriptsYour licensesYour scene and render settings, unchanged1 GPUGPU power: the only difference1, 2, 4 or 8 × RTX 5090 / 4090✓ Only the GPU count changes: open your scene and render the way you do at home
Same on both machinesTip: upload the full project folder with relative paths so textures are found

On iRender you rent a whole Windows workstation with your own Houdini installed, so the TOP network runs with the Local Scheduler just as it does on your desk:

  1. Upload the project with $HIP-relative paths and caches through iRender Drive.
  2. Start a 4x RTX 5090 server, which we recommend for Karma XPU; 8x servers also render Karma XPU, but it does not get proportionally faster.
  3. Install your Houdini build and sign in with your own SideFX license. The Karma XPU kernels compile on the first render and the cache stays on your saved server image.
  4. Set the Local Scheduler to one render job at a time with a retry, and turn on Cache Mode Automatic.
  5. Cook the network; if the session is interrupted, cook it again and only the missing frames render.

TOPs render checklist

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

ChecklistTOPs render 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. Which TOP node should I use to render Karma?

USD Render Scene renders a LOP network or a .usd file; USD Render Files renders .usd files only. Both run husk for each job.

2. How do I make TOPs skip frames that are already rendered?

Set Cache Mode to Automatic. A work item whose expected output exists on disk is marked as cooked without being scheduled.

3. Should I batch frames for Karma XPU?

Start with Frames per Batch 1. Batching reduces process starts, but a retry then covers the whole batch and husk cannot evaluate work item attributes in paths for batched frames.

4. Why should only one render job run at a time on a GPU machine?

Karma XPU uses every GPU it detects and memory is not pooled, so two jobs on the same GPUs each need room for a full copy of the scene.

5. How do I see the exact husk command a work item ran?

Turn on Create Husk Command Attribute. The full command is stored in the huskcommand attribute.

6. Does TOPs need a Houdini license on the render machine?

The Houdini session that cooks the network needs a Houdini or Engine license; the husk jobs themselves render with a Karma license. See SideFX’s licensing pages for your edition.

7. What is Frames per Batch in the USD Render TOP?

It groups work items so one husk process renders several frames. The default of 1 cooks each work item as a separate job.

8. How do I pass extra husk flags from TOPs?

Use the Extra Husk Arguments parameter on the USD Render node, for example -V 2e for elapsed times in the log.

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 →