Skip to content
localmodel.run

audio model · musicgen · Linux

Can I run MusicGen medium on Nvidia DGX Spark (128GB)?

Compatibility verdict memory check
Yes, it runs fast on this GPU

Yes. MusicGen medium runs on Nvidia DGX Spark (128GB) at fp32 (~14 GB of ~112 GB usable).

Needs ~14 GB Device usable ~112 GB

Runs at fp32 using ~14 GB of ~112 GB usable.

Peak memory
~14 GB
Usable on device
~112 GB
Device memory
128 GB
Quant
fp32
Share on X Share on Reddit

How to run it

Use AudioCraft or HF Transformers at fp32. It loads as a single model and runs on a GPU.

Model musicgen
Type
Music generation
Parameters
1.5B
Peak memory
~14 GB at fp32
License
CC-BY-NC-4.0
Full MusicGen medium requirements →
Device Linux
Memory
128 GB unified
Usable for weights
~112 GB
Power draw
~140 W
Best runtime
llama.cpp (CUDA) / Ollama (CUDA)
Best models for Nvidia DGX Spark (128GB) →

You could also run

Run MusicGen medium on other hardware

Nvidia GeForce RTX 4060 Ti (16GB)Nvidia GeForce RTX 4080 (16GB)Nvidia GeForce RTX 5080 (16GB)Nvidia GeForce RTX 5070 Ti (16GB)Nvidia GeForce RTX 5060 Ti (16GB)AMD Radeon RX 9070 XT (16GB)

FAQ

Can Nvidia DGX Spark (128GB) run MusicGen medium?

Yes. MusicGen medium runs on Nvidia DGX Spark (128GB) at fp32 (~14 GB of ~112 GB usable).

How much memory does MusicGen medium need?

Nvidia DGX Spark (128GB) has room to spare. At fp32 the realistic peak is ~14 GB of memory.

What do I use to run MusicGen medium locally?

MusicGen medium runs in AudioCraft or HF Transformers. It needs a GPU.

Sources

VRAM figures are sourced peak-usage anchors at the noted quant; catalog updated 2026-11-01. See methodology.