dexiio

What hardware runs Qwen3 14B?

Qwen3 14B is a 14.77B-parameter dense model (apache-2.0 license). Held over from the 3.0 line because nothing in 3.5 lands between 9B and 27B. Still the solid dense mid-card pick.

All figures assume an f16 KV cache, a 0.6 GB display reserve on the GPU, and 64 GB of DDR5 system RAM for the offload tiers. Tune these in the calculator.

Minimum VRAM by quant and context

QuantFile size8K16K32K
Q2_K *5.4 GB8.5 GB9.8 GB12.4 GB
Q3_K_M *6.8 GB9.9 GB11.2 GB13.9 GB
Q4_K_M8.4 GB11.5 GB12.8 GB15.4 GB
Q5_K_M9.8 GB12.9 GB14.2 GB16.8 GB
Q6_K11.3 GB14.4 GB15.7 GB18.3 GB
Q8_014.6 GB17.7 GB19 GB21.7 GB
IQ4_XS7.6 GB10.7 GB12 GB14.6 GB

Full-GPU figures: weights + f16 KV cache + overhead. * below our recommended floor of Q4_K_M.

GPU compatibility

GPU8K16K32K64K128K
NVIDIA GeForce RTX 5090Q8_0Q8_0Q8_0
NVIDIA GeForce RTX 5080Q6_KQ6_KQ4_K_M
NVIDIA GeForce RTX 5070 TiQ6_KQ6_KQ4_K_M
NVIDIA GeForce RTX 5070Q4_K_MIQ4_XSIQ4_XS
NVIDIA GeForce RTX 5060 Ti 16GBQ6_KQ6_KQ4_K_M
NVIDIA GeForce RTX 5060IQ4_XSIQ4_XSIQ4_XS
NVIDIA GeForce RTX 4090Q8_0Q8_0Q8_0
NVIDIA GeForce RTX 4080 SUPERQ6_KQ6_KQ4_K_M
NVIDIA GeForce RTX 4070 Ti SUPERQ6_KQ6_KQ4_K_M
NVIDIA GeForce RTX 4070Q4_K_MIQ4_XSIQ4_XS
NVIDIA GeForce RTX 4060 Ti 16GBQ6_KQ6_KQ4_K_M
NVIDIA GeForce RTX 4060IQ4_XSIQ4_XSIQ4_XS
NVIDIA GeForce RTX 3090Q8_0Q8_0Q8_0
NVIDIA GeForce RTX 3080 10GBIQ4_XSIQ4_XSIQ4_XS
NVIDIA GeForce RTX 3070IQ4_XSIQ4_XSIQ4_XS
NVIDIA GeForce RTX 3060 TiIQ4_XSIQ4_XSIQ4_XS
NVIDIA GeForce RTX 3060 12GBQ4_K_MIQ4_XSIQ4_XS
NVIDIA GeForce RTX 2080 TiIQ4_XSIQ4_XSIQ4_XS
NVIDIA GeForce GTX 1080 TiIQ4_XSIQ4_XSIQ4_XS
AMD Radeon RX 9070 XTQ6_KQ6_KQ4_K_M
AMD Radeon RX 7900 XTXQ8_0Q8_0Q8_0
AMD Radeon RX 7900 XTQ8_0Q8_0Q6_K
AMD Radeon RX 7800 XTQ6_KQ6_KQ4_K_M
AMD Radeon RX 7600 XTQ6_KQ6_KQ4_K_M
AMD Radeon RX 6800 XTQ6_KQ6_KQ4_K_M
AMD Radeon RX 6700 XTQ4_K_MIQ4_XSIQ4_XS
Intel Arc B580Q4_K_MIQ4_XSIQ4_XS
Intel Arc A770 16GBQ6_KQ6_KQ4_K_M
Apple M2 (16GB)Q4_K_MIQ4_XSIQ4_XS
Apple M2 Pro (32GB)Q8_0Q8_0Q8_0
Apple M2 Max (64GB)Q8_0Q8_0Q8_0
Apple M2 Ultra (128GB)Q8_0Q8_0Q8_0
Apple M3 (16GB)Q4_K_MIQ4_XSIQ4_XS
Apple M3 Pro (36GB)Q8_0Q8_0Q8_0
Apple M3 Max (64GB)Q8_0Q8_0Q8_0
Apple M3 Ultra (96GB)Q8_0Q8_0Q8_0
Apple M4 (16GB)Q4_K_MIQ4_XSIQ4_XS
Apple M4 Pro (48GB)Q8_0Q8_0Q8_0
Apple M4 Max (64GB)Q8_0Q8_0Q8_0
CPU only / integrated graphicsQ8_0Q8_0Q8_0

Fits on GPUExpert offloadPartial offloadCPU only

Quant guidance

Our floor for Qwen3 14B is Q4_K_M — below that, quality degrades faster than the VRAM savings are worth. Prefer the highest quant that still lands "Fits on GPU" at your context length in the table above.

Recommended run command

Q4_K_M at 32K on a NVIDIA GeForce RTX 5080-class GPU (Fits on GPU):

llama-server -m Qwen3-14B-Q4_K_M.gguf -c 32768 --flash-attn -ngl 99

Frequently asked questions

How much VRAM does Qwen3 14B need?

At Q4_K_M and 32K context, Qwen3 14B needs about 15.4 GB of VRAM to run fully on GPU (weights + KV cache + overhead).

What is the smallest GPU that can run Qwen3 14B?

The NVIDIA GeForce RTX 5080 (16 GB) is the smallest GPU in our set that runs Qwen3 14B well at 32K context.

What quantization should I use for Qwen3 14B?

We recommend Q4_K_M or higher. Q4_K_M weighs 8.4 GB (4.88 bits/weight); going below Q4_K_M costs noticeable quality on a model this size.

How long a context can Qwen3 14B handle?

Qwen3 14B supports up to 40K tokens. KV cache grows linearly with context.

Can I run Qwen3 14B without a GPU?

Yes, at reduced speed: on CPU with 64 GB of DDR5 it manages roughly 3 tokens/sec at 8K context (Q8_0).

Related guides