Question 1

How much VRAM does Kimi-Linear-48B-A3B-Base need?

Accepted Answer

Kimi-Linear-48B-A3B-Base (49.1B) requires approximately 112.35482425689698 GB of VRAM at full precision (FP16), 56.2 GB at 8-bit quantization (Q8), and 28.1 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 Kimi-Linear-48B-A3B-Base?

Accepted Answer

We have 9 compatible GPUs indexed for Kimi-Linear-48B-A3B-Base, including MI300X, H200 SXM, B200, B100, MI325X and 4 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 Kimi-Linear-48B-A3B-Base?

Accepted Answer

The cheapest GPU to run Kimi-Linear-48B-A3B-Base in the cloud is the MI300X at $1.99/hr. Other affordable options include H200 SXM ($2.35/hr) and B200 ($5.50/hr). Prices reflect the lowest on-demand rate across cloud providers.

Question 4

Can Kimi-Linear-48B-A3B-Base run on consumer hardware?

Accepted Answer

Kimi-Linear-48B-A3B-Base requires approximately 112.35482425689698 GB of VRAM at FP16, and even with 4-bit quantization (Q4) it needs about 28.1 GB. This exceeds the 24 GB available on the highest-end consumer GPU (RTX 4090), so Kimi-Linear-48B-A3B-Base requires datacenter or cloud GPUs.

Question 5

What quantization options are available for Kimi-Linear-48B-A3B-Base?

Accepted Answer

Kimi-Linear-48B-A3B-Base supports the following quantization levels: FP16 (full precision): ~112.35482425689698 GB, Q8 (8-bit): ~56.2 GB, Q6 (6-bit): ~42.1 GB, Q4 (4-bit): ~28.1 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.

GPU	VRAM	Best Quant	From	Rent
MI300X	192 GB	FP16	$1.99/hr	Rent
H200 SXM	141 GB	FP16	$2.35/hr	Rent
B200	192 GB	FP16	$5.50/hr
B100	192 GB	FP16	—
MI325X	256 GB	FP16	—
MI350X	288 GB	FP16	—
Gaudi 3	128 GB	Q8	—
DGX Spark (GB10)	128 GB	Q8	—
M3 Ultra (192 GB)	144 GB	FP16	—

Kimi-Linear-48B-A3B-Base

VRAM Requirements

Compatible GPUs

Kimi-Linear-48B-A3B-Base FAQ

VRAM requirements for Kimi-Linear-48B-A3B-Base at different quantization levels
Quantization	VRAM Required
FP16	112.4 GB
Q8	56.2 GB
Q6	42.1 GB
Q4	28.1 GB