Skip to content

GLM 4.1V 9B Thinking VRAM requirements

How much VRAM GLM 4.1V 9B Thinking needs
4-bit weights: 5.8 GBWith 8K context: 7.0 GBWith 32K context: 7.9 GB8-bit, 8K context: 11.4 GB
Parameters
10.3B
Longest context
64K
Released
Jun 2025
License
MIT
Strongest at
Text and images

How much VRAM GLM 4.1V 9B Thinking needs

Memory GLM 4.1V 9B Thinking needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 64K
16-bit19.2 GB20.3 GB21.3 GB22.5 GB
8-bit10.2 GB11.4 GB12.3 GB13.5 GB
6-bit7.9 GB9.0 GB10.0 GB11.2 GB
5-bit6.8 GB8.0 GB8.9 GB10.2 GB
4-bit5.8 GB7.0 GB7.9 GB9.2 GB
3-bit4.7 GB5.9 GB6.8 GB8.0 GB
2-bit3.6 GB4.8 GB5.7 GB7.0 GB
1.6-bit2.0 GB3.2 GB4.1 GB5.3 GB

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

GPUs and Macs that run GLM 4.1V 9B Thinking

The best way GLM 4.1V 9B Thinking runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40604-bit20K31 tok/s1.3 sPDF: GLM 4.1V 9B Thinking on RTX 4060 report
RTX 50604-bit20K50 tok/s1.1 sPDF: GLM 4.1V 9B Thinking on RTX 5060 report
RTX 30704-bit20K50 tok/s1.0 sPDF: GLM 4.1V 9B Thinking on RTX 3070 report
RTX 3080 10 GB6-bit16K62 tok/s0.7 sPDF: GLM 4.1V 9B Thinking on RTX 3080 10 GB report
RTX 3060 12 GB6-bit64K31 tok/s1.6 sPDF: GLM 4.1V 9B Thinking on RTX 3060 12 GB report
Arc B5806-bit64K27 tok/s1.1 sPDF: GLM 4.1V 9B Thinking on Arc B580 report
RTX 40706-bit64K42 tok/s0.7 sPDF: GLM 4.1V 9B Thinking on RTX 4070 report
RTX 4070 Super6-bit64K42 tok/s0.6 sPDF: GLM 4.1V 9B Thinking on RTX 4070 Super report
RTX 50706-bit64K55 tok/s0.7 sPDF: GLM 4.1V 9B Thinking on RTX 5070 report
RTX 4060 Ti 16 GB8-bit64K19 tok/s0.9 sPDF: GLM 4.1V 9B Thinking on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit64K29 tok/s0.9 sPDF: GLM 4.1V 9B Thinking on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit64K26 tok/s1.0 sPDF: GLM 4.1V 9B Thinking on Arc A770 16 GB report
Radeon RX 9070 XT8-bit64K35 tok/s0.6 sPDF: GLM 4.1V 9B Thinking on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit64K43 tok/s0.5 sPDF: GLM 4.1V 9B Thinking on RTX 4070 Ti Super report
RTX 4080 Super8-bit64K47 tok/s0.4 sPDF: GLM 4.1V 9B Thinking on RTX 4080 Super report
RTX 5070 Ti8-bit64K57 tok/s0.5 sPDF: GLM 4.1V 9B Thinking on RTX 5070 Ti report
RTX 50808-bit64K61 tok/s0.4 sPDF: GLM 4.1V 9B Thinking on RTX 5080 report
Radeon RX 7900 XTX8-bit64K50 tok/s0.9 sPDF: GLM 4.1V 9B Thinking on Radeon RX 7900 XTX report
RTX 30908-bit64K59 tok/s0.6 sPDF: GLM 4.1V 9B Thinking on RTX 3090 report
RTX 40908-bit64K63 tok/s0.2 sPDF: GLM 4.1V 9B Thinking on RTX 4090 report
Radeon AI PRO R97008-bit64K35 tok/s0.6 sPDF: GLM 4.1V 9B Thinking on Radeon AI PRO R9700 report
RTX 50908-bit64K106 tok/s0.2 sPDF: GLM 4.1V 9B Thinking on RTX 5090 report
2× RTX 30908-bit64K59 tok/s0.3 sPDF: GLM 4.1V 9B Thinking on 2× RTX 3090 report
2× RTX 40908-bit64K63 tok/s0.1 sPDF: GLM 4.1V 9B Thinking on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit64K82 tok/s0.3 sPDF: GLM 4.1V 9B Thinking on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit64K61 tok/s0.2 sPDF: GLM 4.1V 9B Thinking on RTX 6000 Ada report
L40S8-bit64K55 tok/s0.4 sPDF: GLM 4.1V 9B Thinking on L40S report
2× RTX 50908-bit64K106 tok/s0.1 sPDF: GLM 4.1V 9B Thinking on 2× RTX 5090 report
4× RTX 30908-bit64K59 tok/s0.2 sPDF: GLM 4.1V 9B Thinking on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit64K106 tok/s0.2 sPDF: GLM 4.1V 9B Thinking on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB6-bit40K12 tok/s13 sPDF: GLM 4.1V 9B Thinking on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit64K9.3 tok/s13 sPDF: GLM 4.1V 9B Thinking on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit64K9.3 tok/s13 sPDF: GLM 4.1V 9B Thinking on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit64K20 tok/s6.4 sPDF: GLM 4.1V 9B Thinking on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit64K20 tok/s6.4 sPDF: GLM 4.1V 9B Thinking on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit64K20 tok/s6.4 sPDF: GLM 4.1V 9B Thinking on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit64K35 tok/s3.2 sPDF: GLM 4.1V 9B Thinking on M4 Max 128 GB report
M5 Max 128 GB8-bit64K39 tok/s0.9 sPDF: GLM 4.1V 9B Thinking on M5 Max 128 GB report
M3 Ultra 96 GB8-bit64K44 tok/s1.8 sPDF: GLM 4.1V 9B Thinking on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit64K44 tok/s1.8 sPDF: GLM 4.1V 9B Thinking on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit64K44 tok/s1.8 sPDF: GLM 4.1V 9B Thinking on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit64K61 tok/s0.6 sPDF: GLM 4.1V 9B Thinking on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit64K15 tok/s1.8 sPDF: GLM 4.1V 9B Thinking on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit64K15 tok/s1.8 sPDF: GLM 4.1V 9B Thinking on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit64K20 tok/s0.6 sPDF: GLM 4.1V 9B Thinking on RTX Spark report
DGX Spark8-bit64K18 tok/s0.6 sPDF: GLM 4.1V 9B Thinking on DGX Spark report
Dual DGX Spark8-bit64K18 tok/s0.4 sPDF: GLM 4.1V 9B Thinking on Dual DGX Spark report
DGX Station GB3008-bit64K306 tok/s0.1 sPDF: GLM 4.1V 9B Thinking on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit64K106 tok/s0.1 sPDF: GLM 4.1V 9B Thinking on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit64K106 tok/s0.1 sPDF: GLM 4.1V 9B Thinking on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit64K106 tok/s0.1 sPDF: GLM 4.1V 9B Thinking on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit64K119 tok/s0.4 sPDF: GLM 4.1V 9B Thinking on A100 80 GB SXM report
H100 SXM8-bit64K179 tok/s0.1 sPDF: GLM 4.1V 9B Thinking on H100 SXM report
4× H200 NVL8-bit64K235 tok/s0.1 sPDF: GLM 4.1V 9B Thinking on 4× H200 NVL report
8× H200 NVL8-bit64K235 tok/s0.1 sPDF: GLM 4.1V 9B Thinking 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/GLM-4.1V-9B-Thinking-GGUF
vLLM
vllm serve zai-org/GLM-4.1V-9B-Thinking
MLX (Mac)
mlx_lm.chat --model mlx-community/GLM-4.1V-9B-Thinking-8bit

Questions

How much VRAM does GLM 4.1V 9B Thinking need?

GLM 4.1V 9B Thinking needs 5.8 GB for its weights at 4-bit, 7.0 GB with an 8K context and 7.9 GB with 32K. At 8-bit, 11.4 GB.

What is the minimum VRAM for GLM 4.1V 9B Thinking?

About 5.9 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 GLM 4.1V 9B Thinking?

10.3B parameters: 19.2 GB of weights at 16-bit, 5.8 GB at 4-bit.

Can GLM 4.1V 9B Thinking run on an RTX 4090?

Yes. At 8-bit it holds 64K of context and writes about 63 tokens a second.

More models