What GPU do I need to run nari-labs/Dia-1.6B?
A 1.6B-parameter speech model. 1.6B parameters, published in F32. View on Hugging Face
Dia-1.6B is published by nari-labs on Hugging Face, with 28,683 downloads and 2,912 likes to date. It's a unlisted-architecture model built for text-to-speech, published natively in F32.
What Dia-1.6B is
Dia-1.6B is a 1.6B-parameter model published by Nari Labs on Hugging Face that synthesizes speech from text, released under Apache 2.0.
License note: fully permissive, no gating and no usage restrictions. Facts in this section are sourced from Dia-1.6B's Hugging Face model card, not benchmarked by Aquanode.
What it's used for
- Speech synthesis
- Voice agent pipelines
VRAM required & cheapest live GPU fit
Required VRAM = weight size at each precision, plus a fixed overhead for KV-cache, activations, and fragmentation. Full formula and assumptions: methodology.
| Precision | Weight size | Required VRAM | Cheapest live fit | GPUs needed | Est. $/hr (full fit) |
|---|---|---|---|---|---|
| FP32 | 6.0 GB | 7.2 GB | V100 | 1 | $0.088/hr |
| FP8 (quantized) | 1.5 GB | 1.8 GB | RTX 5060 Ti | 1 | $0.110/hr |
| INT4 (quantized) | 0.8 GB | 0.9 GB | RTX 5060 Ti | 1 | $0.110/hr |
A GPU is only matched to a row if its hardware supports that precision, and the primary recommendation is always a single-GPU fit when one exists.
INT4 requires a quantized checkpoint actually published for this model, check its Hugging Face page before relying on this row.
Cheapest way to run Dia-1.6B at its published (F32) precision: 1× V100, at $0.088/hr per GPU ($0.088/hr total). Quantizing to FP8 or INT4 (rows above) can cost less, but requires a compatible quantized checkpoint to exist for this model.
Dia-1.6B: common questions
Does Dia-1.6B fit on a 8 GB GPU?
Yes. At FP32 it needs 7.2 GB of VRAM, so an 8 GB card holds it with 0.8 GB to spare. A 6 GB card is not enough for it at FP32.
Can Dia-1.6B run in 16-bit instead of FP32?
Yes. Its published weights are FP32, 6.0 GB, or 7.2 GB once inference overhead is added. Loading the same weights in 16 bits halves that to 3.0 GB, or 3.6 GB with overhead. That moves it onto a 6 GB card instead of an 8 GB one. How much accuracy the cast costs is model-specific and is not measured here.
How many copies of Dia-1.6B fit on one V100?
2, by VRAM alone. That card carries 16.0 GB and one copy needs 7.2 GB at FP32, on a live rate of $0.088/hr for the whole card. Throughput is not modelled here, so 2 copies is not 2 times the requests served.
What is the least VRAM Dia-1.6B can run in?
0.9 GB, at INT4 (quantized), which fits a 6 GB card, against 7.2 GB at FP32. That row assumes an INT4 checkpoint has actually been published for this model, so check its Hugging Face page before planning around it.
How to run Dia-1.6B
Run Dia-1.6B with Transformers (Python)
Generic example using Hugging Face's transformers library, not from the model's own docs.
from transformers import pipeline
import soundfile as sf
tts = pipeline("text-to-speech", model="nari-labs/Dia-1.6B", device="cuda")
speech = tts("Hello from Aquanode.")
sf.write("output.wav", speech["audio"], speech["sampling_rate"])Deploy Dia-1.6B on Aquanode
Aquanode has no one-click deploy template for Dia-1.6B; you install the inference engine yourself with the commands below. Aquanode sells GPU pods billed per second, not a hosted inference API.
- Launch a bare GPU pod sized to the requirement above (1× V100 or larger).
- Open a terminal on the pod, or save one of the commands above as a startup script so it runs automatically the first time the pod boots.
- Run the command and connect to the resulting endpoint.
Launch a GPU podWeight-to-VRAM math, the fit rules, and how live prices are normalized: full methodology.
Related reading: V100 pricing and specs, and The best GPUs for AI, ranked.
