Question 1

How much VRAM does AI21-Jamba-Reasoning-3B need?

Accepted Answer

AI21-Jamba-Reasoning-3B (3.2B) requires approximately 7.146096897125244 GB of VRAM at full precision (FP16), 3.6 GB at 8-bit quantization (Q8), and 1.8 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 AI21-Jamba-Reasoning-3B?

Accepted Answer

We have 10 compatible GPUs indexed for AI21-Jamba-Reasoning-3B, including K80, L4, A40, A100 40GB, A10G and 5 more. 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 AI21-Jamba-Reasoning-3B?

Accepted Answer

The cheapest GPU to run AI21-Jamba-Reasoning-3B in the cloud is the K80 at $0.18/hr. Other affordable options include L4 ($0.20/hr) and A40 ($0.44/hr). Prices reflect the lowest on-demand rate across cloud providers.

Question 4

Can AI21-Jamba-Reasoning-3B run on consumer hardware?

Accepted Answer

AI21-Jamba-Reasoning-3B may be able to run on high-end consumer GPUs like the RTX 4090 (24 GB) with 4-bit quantization (Q4), which reduces VRAM usage to approximately 1.8 GB. However, performance may be limited compared to datacenter GPUs.

Question 5

What quantization options are available for AI21-Jamba-Reasoning-3B?

Accepted Answer

AI21-Jamba-Reasoning-3B supports the following quantization levels: FP16 (full precision): ~7.146096897125244 GB, Q8 (8-bit): ~3.6 GB, Q6 (6-bit): ~2.7 GB, Q4 (4-bit): ~1.8 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 1 experts (1 active), AI21-Jamba-Reasoning-3B has a larger total weight size but only activates a subset of parameters per token, which can affect effective memory usage during inference.

GPU	VRAM	Best Quant	From	Rent
K80	12 GB	FP16	$0.18/hr
L4	24 GB	FP16	$0.20/hr
A40	48 GB	FP16	$0.44/hr	Rent
A100 40GB	40 GB	FP16	$0.93/hr	Rent
A10G	24 GB	FP16	$1.01/hr
A10	24 GB	FP16	$1.29/hr
A100 80GB PCIe	80 GB	FP16	$1.39/hr	Rent
A100 80GB	80 GB	FP16	$1.39/hr	Rent
H100 MEGA	80 GB	FP16	$1.80/hr	Rent
H100 SXM	80 GB	FP16	$1.80/hr	Rent

AI21-Jamba-Reasoning-3B

VRAM Requirements

Compatible GPUs

AI21-Jamba-Reasoning-3B FAQ

VRAM requirements for AI21-Jamba-Reasoning-3B at different quantization levels
Quantization	VRAM Required
FP16	7.1 GB
Q8	3.6 GB
Q6	2.7 GB
Q4	1.8 GB