Skip to content

GLM 4 9B 0414 VRAM requirements

How much VRAM GLM 4 9B 0414 needs
4-bit weights: 5.3 GBWith 8K context: 6.5 GBWith 32K context: 7.4 GB8-bit, 8K context: 10.5 GB
Parameters
9.4B
Longest context
32K
Released
Apr 2025
License
MIT
Strongest at
Not rated

How much VRAM GLM 4 9B 0414 needs

Memory GLM 4 9B 0414 needs at each precision and context length
PrecisionWeightsWith 8KWith 32K
16-bit17.5 GB18.7 GB19.6 GB
8-bit9.3 GB10.5 GB11.4 GB
6-bit7.2 GB8.4 GB9.3 GB
5-bit6.2 GB7.4 GB8.3 GB
4-bit5.3 GB6.5 GB7.4 GB
3-bit4.3 GB5.5 GB6.4 GB
2-bit3.3 GB4.5 GB5.4 GB
1.6-bit1.8 GB3.0 GB3.9 GB

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

GPUs and Macs that run GLM 4 9B 0414

The best way GLM 4 9B 0414 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40604-bit32K34 tok/s1.2 sPDF: GLM 4 9B 0414 on RTX 4060 report
RTX 50604-bit32K55 tok/s1.0 sPDF: GLM 4 9B 0414 on RTX 5060 report
RTX 30704-bit32K55 tok/s0.9 sPDF: GLM 4 9B 0414 on RTX 3070 report
RTX 3080 10 GB6-bit32K68 tok/s0.6 sPDF: GLM 4 9B 0414 on RTX 3080 10 GB report
RTX 3060 12 GB8-bit28K26 tok/s1.4 sPDF: GLM 4 9B 0414 on RTX 3060 12 GB report
Arc B5808-bit28K23 tok/s1.0 sPDF: GLM 4 9B 0414 on Arc B580 report
RTX 40708-bit28K36 tok/s0.6 sPDF: GLM 4 9B 0414 on RTX 4070 report
RTX 4070 Super8-bit28K36 tok/s0.5 sPDF: GLM 4 9B 0414 on RTX 4070 Super report
RTX 50708-bit28K48 tok/s0.6 sPDF: GLM 4 9B 0414 on RTX 5070 report
RTX 4060 Ti 16 GB8-bit32K21 tok/s0.8 sPDF: GLM 4 9B 0414 on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit32K32 tok/s0.8 sPDF: GLM 4 9B 0414 on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit32K28 tok/s0.9 sPDF: GLM 4 9B 0414 on Arc A770 16 GB report
Radeon RX 9070 XT8-bit32K38 tok/s0.5 sPDF: GLM 4 9B 0414 on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit32K48 tok/s0.4 sPDF: GLM 4 9B 0414 on RTX 4070 Ti Super report
RTX 4080 Super8-bit32K52 tok/s0.4 sPDF: GLM 4 9B 0414 on RTX 4080 Super report
RTX 5070 Ti8-bit32K62 tok/s0.4 sPDF: GLM 4 9B 0414 on RTX 5070 Ti report
RTX 50808-bit32K66 tok/s0.3 sPDF: GLM 4 9B 0414 on RTX 5080 report
Radeon RX 7900 XTX8-bit32K55 tok/s0.8 sPDF: GLM 4 9B 0414 on Radeon RX 7900 XTX report
RTX 30908-bit32K65 tok/s0.5 sPDF: GLM 4 9B 0414 on RTX 3090 report
RTX 40908-bit32K69 tok/s0.2 sPDF: GLM 4 9B 0414 on RTX 4090 report
Radeon AI PRO R97008-bit32K38 tok/s0.5 sPDF: GLM 4 9B 0414 on Radeon AI PRO R9700 report
RTX 50908-bit32K116 tok/s0.2 sPDF: GLM 4 9B 0414 on RTX 5090 report
2× RTX 30908-bit32K65 tok/s0.3 sPDF: GLM 4 9B 0414 on 2× RTX 3090 report
2× RTX 40908-bit32K69 tok/s0.1 sPDF: GLM 4 9B 0414 on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit32K90 tok/s0.3 sPDF: GLM 4 9B 0414 on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit32K66 tok/s0.2 sPDF: GLM 4 9B 0414 on RTX 6000 Ada report
L40S8-bit32K60 tok/s0.3 sPDF: GLM 4 9B 0414 on L40S report
2× RTX 50908-bit32K116 tok/s0.1 sPDF: GLM 4 9B 0414 on 2× RTX 5090 report
4× RTX 30908-bit32K65 tok/s0.2 sPDF: GLM 4 9B 0414 on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit32K116 tok/s0.1 sPDF: GLM 4 9B 0414 on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB6-bit32K13 tok/s12 sPDF: GLM 4 9B 0414 on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit32K10 tok/s12 sPDF: GLM 4 9B 0414 on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit32K10 tok/s12 sPDF: GLM 4 9B 0414 on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit32K22 tok/s5.8 sPDF: GLM 4 9B 0414 on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit32K22 tok/s5.8 sPDF: GLM 4 9B 0414 on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit32K22 tok/s5.8 sPDF: GLM 4 9B 0414 on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit32K38 tok/s2.9 sPDF: GLM 4 9B 0414 on M4 Max 128 GB report
M5 Max 128 GB8-bit32K43 tok/s0.8 sPDF: GLM 4 9B 0414 on M5 Max 128 GB report
M3 Ultra 96 GB8-bit32K48 tok/s1.6 sPDF: GLM 4 9B 0414 on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit32K48 tok/s1.6 sPDF: GLM 4 9B 0414 on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit32K48 tok/s1.6 sPDF: GLM 4 9B 0414 on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit32K66 tok/s0.5 sPDF: GLM 4 9B 0414 on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit32K16 tok/s1.7 sPDF: GLM 4 9B 0414 on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit32K16 tok/s1.7 sPDF: GLM 4 9B 0414 on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit32K22 tok/s0.6 sPDF: GLM 4 9B 0414 on RTX Spark report
DGX Spark8-bit32K20 tok/s0.6 sPDF: GLM 4 9B 0414 on DGX Spark report
Dual DGX Spark8-bit32K20 tok/s0.3 sPDF: GLM 4 9B 0414 on Dual DGX Spark report
DGX Station GB3008-bit32K325 tok/s0.1 sPDF: GLM 4 9B 0414 on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit32K116 tok/s0.1 sPDF: GLM 4 9B 0414 on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit32K116 tok/s0.1 sPDF: GLM 4 9B 0414 on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit32K116 tok/s0.1 sPDF: GLM 4 9B 0414 on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit32K129 tok/s0.4 sPDF: GLM 4 9B 0414 on A100 80 GB SXM report
H100 SXM8-bit32K193 tok/s0.1 sPDF: GLM 4 9B 0414 on H100 SXM report
4× H200 NVL8-bit32K251 tok/s0.1 sPDF: GLM 4 9B 0414 on 4× H200 NVL report
8× H200 NVL8-bit32K251 tok/s0.1 sPDF: GLM 4 9B 0414 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 bartowski/THUDM_GLM-4-9B-0414-GGUF
vLLM
vllm serve zai-org/GLM-4-9B-0414
MLX (Mac)
mlx_lm.chat --model mlx-community/GLM-4-9B-0414-4bit

Questions

How much VRAM does GLM 4 9B 0414 need?

GLM 4 9B 0414 needs 5.3 GB for its weights at 4-bit, 6.5 GB with an 8K context and 7.4 GB with 32K. At 8-bit, 10.5 GB.

What is the minimum VRAM for GLM 4 9B 0414?

About 5.5 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 GLM 4 9B 0414?

9.4B parameters: 17.5 GB of weights at 16-bit, 5.3 GB at 4-bit.

Can GLM 4 9B 0414 run on an RTX 4090?

Yes. At 8-bit it holds 32K of context and writes about 69 tokens a second.

More models