Skip to content

Qwen3 Coder 30B A3B Instruct VRAM requirements

How much VRAM Qwen3 Coder 30B A3B Instruct needs
4-bit weights: 17.2 GBWith 8K context: 18.8 GBWith 32K context: 21.0 GB8-bit, 8K context: 31.8 GB
Parameters
30.5B
Active per token
3B
Longest context
256K
Released
Jul 2025
License
Apache 2.0
Strongest at
Coding

How much VRAM Qwen3 Coder 30B A3B Instruct needs

Memory Qwen3 Coder 30B A3B Instruct needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit56.9 GB58.4 GB60.7 GB69.7 GB
8-bit30.2 GB31.8 GB34.0 GB43.0 GB
6-bit23.3 GB24.9 GB27.1 GB36.1 GB
5-bit20.2 GB21.8 GB24.0 GB33.0 GB
4-bit17.2 GB18.8 GB21.0 GB30.0 GB
3-bit13.9 GB15.5 GB17.7 GB26.7 GB
2-bit10.7 GB12.2 GB14.5 GB23.5 GB
1.6-bit5.9 GB7.4 GB9.7 GB18.7 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run Qwen3 Coder 30B A3B Instruct

The best way Qwen3 Coder 30B A3B Instruct runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Qwen3 Coder 30B A3B Instruct on RTX 4060 report
RTX 5060Too bigPDF: Qwen3 Coder 30B A3B Instruct on RTX 5060 report
RTX 3070Too bigPDF: Qwen3 Coder 30B A3B Instruct on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Qwen3 Coder 30B A3B Instruct on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Qwen3 Coder 30B A3B Instruct on RTX 3060 12 GB report
Arc B580Too bigPDF: Qwen3 Coder 30B A3B Instruct on Arc B580 report
RTX 4070Too bigPDF: Qwen3 Coder 30B A3B Instruct on RTX 4070 report
RTX 4070 SuperToo bigPDF: Qwen3 Coder 30B A3B Instruct on RTX 4070 Super report
RTX 5070Too bigPDF: Qwen3 Coder 30B A3B Instruct on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Qwen3 Coder 30B A3B Instruct on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Qwen3 Coder 30B A3B Instruct on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Qwen3 Coder 30B A3B Instruct on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Qwen3 Coder 30B A3B Instruct on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Qwen3 Coder 30B A3B Instruct on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Qwen3 Coder 30B A3B Instruct on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Qwen3 Coder 30B A3B Instruct on RTX 5070 Ti report
RTX 5080Too bigPDF: Qwen3 Coder 30B A3B Instruct on RTX 5080 report
Radeon RX 7900 XTX5-bit16K125 tok/s0.8 sPDF: Qwen3 Coder 30B A3B Instruct on Radeon RX 7900 XTX report
RTX 30905-bit16K163 tok/s0.5 sPDF: Qwen3 Coder 30B A3B Instruct on RTX 3090 report
RTX 40905-bit16K169 tok/s0.2 sPDF: Qwen3 Coder 30B A3B Instruct on RTX 4090 report
Radeon AI PRO R97006-bit64K93 tok/s0.5 sPDF: Qwen3 Coder 30B A3B Instruct on Radeon AI PRO R9700 report
RTX 50906-bit64K207 tok/s0.2 sPDF: Qwen3 Coder 30B A3B Instruct on RTX 5090 report
2× RTX 30908-bit144K129 tok/s0.3 sPDF: Qwen3 Coder 30B A3B Instruct on 2× RTX 3090 report
2× RTX 40908-bit144K135 tok/s0.1 sPDF: Qwen3 Coder 30B A3B Instruct on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit152K160 tok/s0.3 sPDF: Qwen3 Coder 30B A3B Instruct on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit152K131 tok/s0.2 sPDF: Qwen3 Coder 30B A3B Instruct on RTX 6000 Ada report
L40S8-bit152K123 tok/s0.3 sPDF: Qwen3 Coder 30B A3B Instruct on L40S report
2× RTX 50908-bit256K184 tok/s0.1 sPDF: Qwen3 Coder 30B A3B Instruct on 2× RTX 5090 report
4× RTX 30908-bit256K129 tok/s0.2 sPDF: Qwen3 Coder 30B A3B Instruct on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit256K184 tok/s0.1 sPDF: Qwen3 Coder 30B A3B Instruct on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Qwen3 Coder 30B A3B Instruct on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Qwen3 Coder 30B A3B Instruct on Mac mini M4 24 GB report
Mac mini M4 32 GB4-bit24K40 tok/s12 sPDF: Qwen3 Coder 30B A3B Instruct on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Qwen3 Coder 30B A3B Instruct on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit32K48 tok/s5.9 sPDF: Qwen3 Coder 30B A3B Instruct on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit152K48 tok/s5.9 sPDF: Qwen3 Coder 30B A3B Instruct on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit256K69 tok/s2.9 sPDF: Qwen3 Coder 30B A3B Instruct on M4 Max 128 GB report
M5 Max 128 GB8-bit256K73 tok/s0.9 sPDF: Qwen3 Coder 30B A3B Instruct on M5 Max 128 GB report
M3 Ultra 96 GB8-bit256K79 tok/s1.7 sPDF: Qwen3 Coder 30B A3B Instruct on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit256K79 tok/s1.7 sPDF: Qwen3 Coder 30B A3B Instruct on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit256K79 tok/s1.7 sPDF: Qwen3 Coder 30B A3B Instruct on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit256K93 tok/s0.5 sPDF: Qwen3 Coder 30B A3B Instruct on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit152K40 tok/s1.7 sPDF: Qwen3 Coder 30B A3B Instruct on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit256K40 tok/s1.7 sPDF: Qwen3 Coder 30B A3B Instruct on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit256K55 tok/s0.6 sPDF: Qwen3 Coder 30B A3B Instruct on RTX Spark report
DGX Spark8-bit256K51 tok/s0.7 sPDF: Qwen3 Coder 30B A3B Instruct on DGX Spark report
Dual DGX Spark8-bit256K51 tok/s0.4 sPDF: Qwen3 Coder 30B A3B Instruct on Dual DGX Spark report
DGX Station GB3008-bit256K284 tok/s0.1 sPDF: Qwen3 Coder 30B A3B Instruct on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit256K184 tok/s0.1 sPDF: Qwen3 Coder 30B A3B Instruct on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit256K184 tok/s0.1 sPDF: Qwen3 Coder 30B A3B Instruct on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K184 tok/s0.1 sPDF: Qwen3 Coder 30B A3B Instruct on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit256K196 tok/s0.4 sPDF: Qwen3 Coder 30B A3B Instruct on A100 80 GB SXM report
H100 SXM8-bit256K236 tok/s0.1 sPDF: Qwen3 Coder 30B A3B Instruct on H100 SXM report
4× H200 NVL8-bit256K261 tok/s0.1 sPDF: Qwen3 Coder 30B A3B Instruct on 4× H200 NVL report
8× H200 NVL8-bit256K261 tok/s0.1 sPDF: Qwen3 Coder 30B A3B Instruct on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF
vLLM
vllm serve Qwen/Qwen3-Coder-30B-A3B-Instruct
MLX (Mac)
mlx_lm.chat --model mlx-community/Qwen3-Coder-30B-A3B-Instruct-4bit

Questions

How much VRAM does Qwen3 Coder 30B A3B Instruct need?

Qwen3 Coder 30B A3B Instruct needs 17.2 GB for its weights at 4-bit, 18.8 GB with an 8K context and 21.0 GB with 32K. At 8-bit, 31.8 GB.

What is the minimum VRAM for Qwen3 Coder 30B A3B Instruct?

About 15.5 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the Radeon RX 7900 XTX.

How big is Qwen3 Coder 30B A3B Instruct?

30.5B parameters (3B active per token): 56.9 GB of weights at 16-bit, 17.2 GB at 4-bit.

Can Qwen3 Coder 30B A3B Instruct run on an RTX 4090?

Yes. At 5-bit it holds 16K of context and writes about 169 tokens a second.

More models