Skip to content

Qwen3 1.7B Base VRAM requirements

How much VRAM Qwen3 1.7B Base needs
4-bit weights: 1.0 GBWith 8K context: 2.6 GBWith 32K context: 5.3 GB8-bit, 8K context: 3.4 GB
Parameters
1.7B
Longest context
32K
Released
Apr 2025
License
Apache 2.0
Strongest at
Not rated

How much VRAM Qwen3 1.7B Base needs

Memory Qwen3 1.7B Base needs at each precision and context length
PrecisionWeightsWith 8KWith 32K
16-bit3.2 GB4.9 GB7.5 GB
8-bit1.7 GB3.4 GB6.0 GB
6-bit1.3 GB3.0 GB5.6 GB
5-bit1.1 GB2.8 GB5.4 GB
4-bit1.0 GB2.6 GB5.3 GB
3-bit802 MB2.5 GB5.1 GB
2-bit615 MB2.3 GB4.9 GB
1.6-bit338 MB2.0 GB4.6 GB

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

GPUs and Macs that run Qwen3 1.7B Base

The best way Qwen3 1.7B Base runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit32K96 tok/s0.2 sPDF: Qwen3 1.7B Base on RTX 4060 report
RTX 50608-bit32K150 tok/s0.2 sPDF: Qwen3 1.7B Base on RTX 5060 report
RTX 30708-bit32K150 tok/s0.2 sPDF: Qwen3 1.7B Base on RTX 3070 report
RTX 3080 10 GB8-bit32K234 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX 3080 10 GB report
RTX 3060 12 GB8-bit32K124 tok/s0.3 sPDF: Qwen3 1.7B Base on RTX 3060 12 GB report
Arc B5808-bit32K103 tok/s0.2 sPDF: Qwen3 1.7B Base on Arc B580 report
RTX 40708-bit32K166 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX 4070 report
RTX 4070 Super8-bit32K166 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX 4070 Super report
RTX 50708-bit32K212 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX 5070 report
RTX 4060 Ti 16 GB8-bit32K101 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit32K150 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit32K122 tok/s0.2 sPDF: Qwen3 1.7B Base on Arc A770 16 GB report
Radeon RX 9070 XT8-bit32K162 tok/s0.1 sPDF: Qwen3 1.7B Base on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit32K212 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX 4070 Ti Super report
RTX 4080 Super8-bit32K228 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX 4080 Super report
RTX 5070 Ti8-bit32K266 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX 5070 Ti report
RTX 50808-bit32K281 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX 5080 report
Radeon RX 7900 XTX8-bit32K219 tok/s0.1 sPDF: Qwen3 1.7B Base on Radeon RX 7900 XTX report
RTX 30908-bit32K275 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX 3090 report
RTX 40908-bit32K291 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX 4090 report
Radeon AI PRO R97008-bit32K162 tok/s0.1 sPDF: Qwen3 1.7B Base on Radeon AI PRO R9700 report
RTX 50908-bit32K435 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX 5090 report
2× RTX 30908-bit32K275 tok/s0.1 sPDF: Qwen3 1.7B Base on 2× RTX 3090 report
2× RTX 40908-bit32K291 tok/s0.1 sPDF: Qwen3 1.7B Base on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit32K359 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit32K281 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX 6000 Ada report
L40S8-bit32K259 tok/s0.1 sPDF: Qwen3 1.7B Base on L40S report
2× RTX 50908-bit32K435 tok/s0.1 sPDF: Qwen3 1.7B Base on 2× RTX 5090 report
4× RTX 30908-bit32K275 tok/s0.1 sPDF: Qwen3 1.7B Base on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit32K435 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit32K48 tok/s2.1 sPDF: Qwen3 1.7B Base on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit32K48 tok/s2.1 sPDF: Qwen3 1.7B Base on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit32K48 tok/s2.1 sPDF: Qwen3 1.7B Base on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit32K96 tok/s1.0 sPDF: Qwen3 1.7B Base on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit32K96 tok/s1.0 sPDF: Qwen3 1.7B Base on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit32K96 tok/s1.0 sPDF: Qwen3 1.7B Base on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit32K149 tok/s0.5 sPDF: Qwen3 1.7B Base on M4 Max 128 GB report
M5 Max 128 GB8-bit32K161 tok/s0.2 sPDF: Qwen3 1.7B Base on M5 Max 128 GB report
M3 Ultra 96 GB8-bit32K176 tok/s0.3 sPDF: Qwen3 1.7B Base on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit32K176 tok/s0.3 sPDF: Qwen3 1.7B Base on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit32K176 tok/s0.3 sPDF: Qwen3 1.7B Base on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit32K218 tok/s0.1 sPDF: Qwen3 1.7B Base on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit32K75 tok/s0.3 sPDF: Qwen3 1.7B Base on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit32K75 tok/s0.3 sPDF: Qwen3 1.7B Base on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit32K105 tok/s0.1 sPDF: Qwen3 1.7B Base on RTX Spark report
DGX Spark8-bit32K96 tok/s0.1 sPDF: Qwen3 1.7B Base on DGX Spark report
Dual DGX Spark8-bit32K96 tok/s0.1 sPDF: Qwen3 1.7B Base on Dual DGX Spark report
DGX Station GB3008-bit32K828 tok/s0.1 sPDF: Qwen3 1.7B Base on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit32K435 tok/s0.1 sPDF: Qwen3 1.7B Base on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit32K435 tok/s0.1 sPDF: Qwen3 1.7B Base on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit32K435 tok/s0.1 sPDF: Qwen3 1.7B Base on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit32K472 tok/s0.1 sPDF: Qwen3 1.7B Base on A100 80 GB SXM report
H100 SXM8-bit32K617 tok/s0.1 sPDF: Qwen3 1.7B Base on H100 SXM report
4× H200 NVL8-bit32K723 tok/s0.1 sPDF: Qwen3 1.7B Base on 4× H200 NVL report
8× H200 NVL8-bit32K723 tok/s0.1 sPDF: Qwen3 1.7B 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 eulogik/Bharat-Tiny-LLM-v3
vLLM
vllm serve Qwen/Qwen3-1.7B-Base

Questions

How much VRAM does Qwen3 1.7B Base need?

Qwen3 1.7B Base needs 1.0 GB for its weights at 4-bit, 2.6 GB with an 8K context and 5.3 GB with 32K. At 8-bit, 3.4 GB.

What is the minimum VRAM for Qwen3 1.7B Base?

About 2.5 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 RTX 4060.

How big is Qwen3 1.7B Base?

1.7B parameters: 3.2 GB of weights at 16-bit, 1.0 GB at 4-bit.

Can Qwen3 1.7B Base run on an RTX 4090?

Yes. At 8-bit it holds 32K of context and writes about 291 tokens a second.

More models