Skip to content

Qwen1.5 1.8B Chat VRAM requirements

How much VRAM Qwen1.5 1.8B Chat needs
4-bit weights: 1.0 GBWith 8K context: 3.3 GBWith 32K context: 7.8 GB8-bit, 8K context: 4.1 GB
Parameters
1.8B
Longest context
32K
Released
Jan 2024
License
Tongyi Qianwen Research
Strongest at
Not rated

Its licence (Tongyi Qianwen Research) is not a standard open-source one: check the terms before commercial use.

How much VRAM Qwen1.5 1.8B Chat needs

Memory Qwen1.5 1.8B Chat needs at each precision and context length
PrecisionWeightsWith 8KWith 32K
16-bit3.4 GB5.7 GB10.2 GB
8-bit1.8 GB4.1 GB8.6 GB
6-bit1.4 GB3.7 GB8.2 GB
5-bit1.2 GB3.5 GB8.0 GB
4-bit1.0 GB3.3 GB7.8 GB
3-bit856 MB3.1 GB7.6 GB
2-bit657 MB2.9 GB7.4 GB
1.6-bit361 MB2.7 GB7.2 GB

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

GPUs and Macs that run Qwen1.5 1.8B Chat

The best way Qwen1.5 1.8B Chat runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit24K108 tok/s0.2 sPDF: Qwen1.5 1.8B Chat on RTX 4060 report
RTX 50608-bit24K170 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX 5060 report
RTX 30708-bit24K170 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX 3070 report
RTX 3080 10 GB8-bit32K265 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX 3080 10 GB report
RTX 3060 12 GB8-bit32K140 tok/s0.2 sPDF: Qwen1.5 1.8B Chat on RTX 3060 12 GB report
Arc B5808-bit32K117 tok/s0.2 sPDF: Qwen1.5 1.8B Chat on Arc B580 report
RTX 40708-bit32K188 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX 4070 report
RTX 4070 Super8-bit32K188 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX 4070 Super report
RTX 50708-bit32K240 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX 5070 report
RTX 4060 Ti 16 GB8-bit32K114 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit32K170 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit32K138 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on Arc A770 16 GB report
Radeon RX 9070 XT8-bit32K184 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit32K240 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX 4070 Ti Super report
RTX 4080 Super8-bit32K259 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX 4080 Super report
RTX 5070 Ti8-bit32K303 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX 5070 Ti report
RTX 50808-bit32K319 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX 5080 report
Radeon RX 7900 XTX8-bit32K249 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on Radeon RX 7900 XTX report
RTX 30908-bit32K313 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX 3090 report
RTX 40908-bit32K331 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX 4090 report
Radeon AI PRO R97008-bit32K184 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on Radeon AI PRO R9700 report
RTX 50908-bit32K497 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX 5090 report
2× RTX 30908-bit32K313 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on 2× RTX 3090 report
2× RTX 40908-bit32K331 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit32K409 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit32K319 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX 6000 Ada report
L40S8-bit32K294 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on L40S report
2× RTX 50908-bit32K497 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on 2× RTX 5090 report
4× RTX 30908-bit32K313 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit32K497 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit32K54 tok/s1.8 sPDF: Qwen1.5 1.8B Chat on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit32K54 tok/s1.8 sPDF: Qwen1.5 1.8B Chat on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit32K54 tok/s1.8 sPDF: Qwen1.5 1.8B Chat on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit32K109 tok/s0.9 sPDF: Qwen1.5 1.8B Chat on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit32K109 tok/s0.9 sPDF: Qwen1.5 1.8B Chat on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit32K109 tok/s0.9 sPDF: Qwen1.5 1.8B Chat on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit32K170 tok/s0.5 sPDF: Qwen1.5 1.8B Chat on M4 Max 128 GB report
M5 Max 128 GB8-bit32K184 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on M5 Max 128 GB report
M3 Ultra 96 GB8-bit32K201 tok/s0.3 sPDF: Qwen1.5 1.8B Chat on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit32K201 tok/s0.3 sPDF: Qwen1.5 1.8B Chat on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit32K201 tok/s0.3 sPDF: Qwen1.5 1.8B Chat on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit32K250 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit32K85 tok/s0.3 sPDF: Qwen1.5 1.8B Chat on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit32K85 tok/s0.3 sPDF: Qwen1.5 1.8B Chat on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit32K118 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on RTX Spark report
DGX Spark8-bit32K109 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on DGX Spark report
Dual DGX Spark8-bit32K109 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on Dual DGX Spark report
DGX Station GB3008-bit32K956 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit32K497 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit32K497 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit32K497 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit32K539 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on A100 80 GB SXM report
H100 SXM8-bit32K709 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on H100 SXM report
4× H200 NVL8-bit32K832 tok/s0.1 sPDF: Qwen1.5 1.8B Chat on 4× H200 NVL report
8× H200 NVL8-bit32K832 tok/s0.1 sPDF: Qwen1.5 1.8B Chat 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 second-state/Qwen1.5-1.8B-Chat-GGUF
vLLM
vllm serve Qwen/Qwen1.5-1.8B-Chat

Questions

How much VRAM does Qwen1.5 1.8B Chat need?

Qwen1.5 1.8B Chat needs 1.0 GB for its weights at 4-bit, 3.3 GB with an 8K context and 7.8 GB with 32K. At 8-bit, 4.1 GB.

What is the minimum VRAM for Qwen1.5 1.8B Chat?

About 3.1 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 Qwen1.5 1.8B Chat?

1.8B parameters: 3.4 GB of weights at 16-bit, 1.0 GB at 4-bit.

Can Qwen1.5 1.8B Chat run on an RTX 4090?

Yes. At 8-bit it holds 32K of context and writes about 331 tokens a second.

More models