Skip to content

Qwen3 0.6B Base VRAM requirements

How much VRAM Qwen3 0.6B Base needs
4-bit weights: 345 MBWith 8K context: 2.0 GBWith 32K context: 4.6 GB8-bit, 8K context: 2.2 GB
Parameters
596M
Longest context
32K
Released
Apr 2025
License
Apache 2.0
Strongest at
Not rated

How much VRAM Qwen3 0.6B Base needs

Memory Qwen3 0.6B Base needs at each precision and context length
PrecisionWeightsWith 8KWith 32K
16-bit1.1 GB2.8 GB5.4 GB
8-bit604 MB2.2 GB4.9 GB
6-bit466 MB2.1 GB4.7 GB
5-bit404 MB2.0 GB4.7 GB
4-bit345 MB2.0 GB4.6 GB
3-bit278 MB1.9 GB4.5 GB
2-bit213 MB1.9 GB4.5 GB
1.6-bit117 MB1.8 GB4.4 GB

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

GPUs and Macs that run Qwen3 0.6B Base

The best way Qwen3 0.6B Base runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit32K240 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 4060 report
RTX 50608-bit32K350 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 5060 report
RTX 30708-bit32K350 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 3070 report
RTX 3080 10 GB8-bit32K493 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 3080 10 GB report
RTX 3060 12 GB8-bit32K298 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 3060 12 GB report
Arc B5808-bit32K224 tok/s0.1 sPDF: Qwen3 0.6B Base on Arc B580 report
RTX 40708-bit32K379 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 4070 report
RTX 4070 Super8-bit32K379 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 4070 Super report
RTX 50708-bit32K457 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 5070 report
RTX 4060 Ti 16 GB8-bit32K251 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit32K350 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit32K254 tok/s0.1 sPDF: Qwen3 0.6B Base on Arc A770 16 GB report
Radeon RX 9070 XT8-bit32K328 tok/s0.1 sPDF: Qwen3 0.6B Base on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit32K457 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 4070 Ti Super report
RTX 4080 Super8-bit32K483 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 4080 Super report
RTX 5070 Ti8-bit32K541 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 5070 Ti report
RTX 50808-bit32K561 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 5080 report
Radeon RX 7900 XTX8-bit32K400 tok/s0.1 sPDF: Qwen3 0.6B Base on Radeon RX 7900 XTX report
RTX 30908-bit32K554 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 3090 report
RTX 40908-bit32K576 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 4090 report
Radeon AI PRO R97008-bit32K328 tok/s0.1 sPDF: Qwen3 0.6B Base on Radeon AI PRO R9700 report
RTX 50908-bit32K744 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 5090 report
2× RTX 30908-bit32K554 tok/s0.1 sPDF: Qwen3 0.6B Base on 2× RTX 3090 report
2× RTX 40908-bit32K576 tok/s0.1 sPDF: Qwen3 0.6B Base on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit32K661 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit32K561 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX 6000 Ada report
L40S8-bit32K530 tok/s0.1 sPDF: Qwen3 0.6B Base on L40S report
2× RTX 50908-bit32K744 tok/s0.1 sPDF: Qwen3 0.6B Base on 2× RTX 5090 report
4× RTX 30908-bit32K554 tok/s0.1 sPDF: Qwen3 0.6B Base on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit32K744 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit32K115 tok/s0.7 sPDF: Qwen3 0.6B Base on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit32K115 tok/s0.7 sPDF: Qwen3 0.6B Base on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit32K115 tok/s0.7 sPDF: Qwen3 0.6B Base on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit32K197 tok/s0.3 sPDF: Qwen3 0.6B Base on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit32K197 tok/s0.3 sPDF: Qwen3 0.6B Base on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit32K197 tok/s0.3 sPDF: Qwen3 0.6B Base on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit32K264 tok/s0.2 sPDF: Qwen3 0.6B Base on M4 Max 128 GB report
M5 Max 128 GB8-bit32K277 tok/s0.1 sPDF: Qwen3 0.6B Base on M5 Max 128 GB report
M3 Ultra 96 GB8-bit32K292 tok/s0.1 sPDF: Qwen3 0.6B Base on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit32K292 tok/s0.1 sPDF: Qwen3 0.6B Base on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit32K292 tok/s0.1 sPDF: Qwen3 0.6B Base on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit32K328 tok/s0.1 sPDF: Qwen3 0.6B Base on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit32K181 tok/s0.1 sPDF: Qwen3 0.6B Base on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit32K181 tok/s0.1 sPDF: Qwen3 0.6B Base on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit32K259 tok/s0.1 sPDF: Qwen3 0.6B Base on RTX Spark report
DGX Spark8-bit32K241 tok/s0.1 sPDF: Qwen3 0.6B Base on DGX Spark report
Dual DGX Spark8-bit32K241 tok/s0.1 sPDF: Qwen3 0.6B Base on Dual DGX Spark report
DGX Station GB3008-bit32K1,034 tok/s0.1 sPDF: Qwen3 0.6B Base on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit32K744 tok/s0.1 sPDF: Qwen3 0.6B Base on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit32K744 tok/s0.1 sPDF: Qwen3 0.6B Base on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit32K744 tok/s0.1 sPDF: Qwen3 0.6B Base on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit32K779 tok/s0.1 sPDF: Qwen3 0.6B Base on A100 80 GB SXM report
H100 SXM8-bit32K901 tok/s0.1 sPDF: Qwen3 0.6B Base on H100 SXM report
4× H200 NVL8-bit32K972 tok/s0.1 sPDF: Qwen3 0.6B Base on 4× H200 NVL report
8× H200 NVL8-bit32K972 tok/s0.1 sPDF: Qwen3 0.6B Base 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 Qwen/Qwen3-Embedding-0.6B-GGUF
vLLM
vllm serve Qwen/Qwen3-0.6B-Base
MLX (Mac)
mlx_lm.chat --model mlx-community/Qwen3-0.6B-Base-4bit

Questions

How much VRAM does Qwen3 0.6B Base need?

Qwen3 0.6B Base needs 345 MB for its weights at 4-bit, 2.0 GB with an 8K context and 4.6 GB with 32K. At 8-bit, 2.2 GB.

What is the minimum VRAM for Qwen3 0.6B Base?

About 1.9 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 RTX 4060.

How big is Qwen3 0.6B Base?

596M parameters: 1.1 GB of weights at 16-bit, 345 MB at 4-bit.

Can Qwen3 0.6B Base run on an RTX 4090?

Yes. At 8-bit it holds 32K of context and writes about 576 tokens a second.

More models