Question 1

How much VRAM does Hy3-preview need?

Accepted Answer

Hy3-preview (298.8B) requires approximately 663.8545989990234 GB of VRAM at full precision (FP16), 331.9 GB at 8-bit quantization (Q8), and 166 GB at 4-bit quantization (Q4). The actual VRAM usage may be slightly higher due to KV cache and runtime overhead.

Question 2

What GPUs can run Hy3-preview?

Accepted Answer

We currently have no compatible GPUs indexed for Hy3-preview. This may be because the model's VRAM requirements exceed available GPUs in our database, or VRAM data hasn't been computed yet.

Question 3

What is the cheapest GPU to run Hy3-preview?

Accepted Answer

We don't currently have cloud pricing data for GPUs compatible with Hy3-preview. Cloud pricing data is updated regularly — check back soon.

Question 4

Can Hy3-preview run on consumer hardware?

Accepted Answer

Hy3-preview requires approximately 663.8545989990234 GB of VRAM at FP16, and even with 4-bit quantization (Q4) it needs about 166.0 GB. This exceeds the 24 GB available on the highest-end consumer GPU (RTX 4090), so Hy3-preview requires datacenter or cloud GPUs.

Question 5

What quantization options are available for Hy3-preview?

Accepted Answer

Hy3-preview supports the following quantization levels: FP16 (full precision): ~663.8545989990234 GB, Q8 (8-bit): ~331.9 GB, Q6 (6-bit): ~248.9 GB, Q4 (4-bit): ~166 GB. Lower quantization reduces VRAM usage at the cost of some quality loss. Q8 offers a good balance between quality and memory savings, while Q4 provides maximum compression. As a Mixture of Experts (MoE) model with 192 experts (8 active), Hy3-preview has a larger total weight size but only activates a subset of parameters per token, which can affect effective memory usage during inference.

Hy3-preview

VRAM Requirements

Compatible GPUs

Hy3-preview FAQ

VRAM requirements for Hy3-preview at different quantization levels
Quantization	VRAM Required
FP16	663.9 GB
Q8	331.9 GB
Q6	248.9 GB
Q4	166 GB