Question 1

How much VRAM does NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16 need?

Accepted Answer

NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16 (123.6B) requires approximately 276.29214248657223 GB of VRAM at full precision (FP16), 138.1 GB at 8-bit quantization (Q8), and 69.1 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 NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16?

Accepted Answer

We have 1 compatible GPUs indexed for NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16, including MI350X. Compatibility is determined by whether the GPU's VRAM can fit the model's memory footprint, potentially using quantization to reduce memory usage.

Question 3

What is the cheapest GPU to run NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16?

Accepted Answer

We don't currently have cloud pricing data for GPUs compatible with NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16. Cloud pricing data is updated regularly — check back soon.

Question 4

Can NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16 run on consumer hardware?

Accepted Answer

NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16 requires approximately 276.29214248657223 GB of VRAM at FP16, and even with 4-bit quantization (Q4) it needs about 69.1 GB. This exceeds the 24 GB available on the highest-end consumer GPU (RTX 4090), so NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16 requires datacenter or cloud GPUs.

Question 5

What quantization options are available for NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16?

Accepted Answer

NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16 supports the following quantization levels: FP16 (full precision): ~276.29214248657223 GB, Q8 (8-bit): ~138.1 GB, Q6 (6-bit): ~103.6 GB, Q4 (4-bit): ~69.1 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 512 experts (22 active), NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16 has a larger total weight size but only activates a subset of parameters per token, which can affect effective memory usage during inference.

NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16

VRAM Requirements

Compatible GPUs

NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16 FAQ

VRAM requirements for NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16 at different quantization levels
Quantization	VRAM Required
FP16	276.3 GB
Q8	138.1 GB
Q6	103.6 GB
Q4	69.1 GB