dgx-spark-qwen38 / Models / MiniMax-H3
Run MiniMax-H3 on a DGX Spark
Video with sound, by MiniMax. On SGLang.
MiniMax's video model, which makes the sound with the picture, served with the SGLang cookbook's MiniMax-H3 recipe as upstream wrote it. On an ASUS Ascent GX10 with 128 GB: 126.2 to 127.9 s, with the Turbo adapter, for 4 s of 480P video.
Measured
126.2 to 127.9s
with the Turbo adapter, for 4 s of 480P video
- 622.8 to 632.5 s at 50 steps without it
- Sound made with the picture
- From a sentence
- Measured on
- an ASUS Ascent GX10 with 128 GB, the project's reference box
- How
- 4 s at 480P, timed on the box
- Source
- docs/video-lane.md
Load it
From the cockpit: open the Lanes view, pick MiniMax-H3 and press Load. From a terminal, in the repo folder on the box:
./prepare-lane.sh video # the first time only, builds the lane
./switch-model.sh video # switches the boot, then prints the commands to runswitch-model.sh never restarts an engine itself: it prints the exact stop and start commands. The cockpit's Load runs them for you.
On a 64 GB DGX Spark
No: it needs a 128 GB DGX Spark. A 64 GB box serves Nemotron 3.5 Bolt and Qwen3.8-27B: what runs on 64 GB.
Use it
From the cockpit: its Video view drives the lane. The lane listens on 127.0.0.1:30022, on the box itself, and not through the API on port 30001.
Install it
One command, on stock DGX OS. The same command updates it later.
curl -fsSL https://raw.githubusercontent.com/hasso5703/dgx-spark-qwen38/main/get.sh | bashQuestions about MiniMax-H3 on a DGX Spark
Can a DGX Spark run MiniMax-H3?
Yes. MiniMax's video model, which makes the sound with the picture, served with the SGLang cookbook's MiniMax-H3 recipe as upstream wrote it. The installer of dgx-spark-qwen38 sets it up, and the cockpit loads it with one button.
How fast is MiniMax-H3 on a DGX Spark?
126.2 to 127.9 s, with the Turbo adapter, for 4 s of 480P video, measured on an ASUS Ascent GX10 with 128 GB. 622.8 to 632.5 s at 50 steps without it. Sound made with the picture. From a sentence.
Does MiniMax-H3 run on a 64 GB DGX Spark?
No: it needs a 128 GB DGX Spark.
The other lanes
- QwenQwen3.8-27B71.4 tok/s, greedy median, one request
- QwenQwen3.8-Flash-Next 176B47.9 tok/s, on code, with FLASH_TIER=context
- NVIDIAFlash FP4-Local55.4 tok/s, greedy median, at its default effort
- NVIDIANemotron 3.5 Bolt140 tok/s, rewriting code, one request, reasoning off
- NVIDIANemotron 3.5 Super VL69.5 tok/s, rewriting code, one request, reasoning off
- Ornith AIOrnith 1.5 122B42.4 tok/s, greedy median, one request
- Aleph AlphaKolibri-148.0 tok/s, one request
- QwenQwen-Image 2.17.5 s, with its Turbo, per 1024×1024 image