Skip to content

GLM 4.5V VRAM requirements

How much VRAM GLM 4.5V needs
4-bit weights: 60.8 GBWith 8K context: 63.1 GBWith 32K context: 67.4 GB8-bit, 8K context: 109 GB
Parameters
108B
Active per token
14.3B
Longest context
64K
Released
Aug 2025
License
MIT
Strongest at
Text and images

How much VRAM GLM 4.5V needs

Memory GLM 4.5V needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 64K
16-bit201 GB203 GB207 GB213 GB
8-bit107 GB109 GB113 GB119 GB
6-bit82.3 GB84.6 GB88.9 GB94.6 GB
5-bit71.3 GB73.6 GB78.0 GB83.7 GB
4-bit60.8 GB63.1 GB67.4 GB73.2 GB
3-bit49.0 GB51.3 GB55.6 GB61.4 GB
2-bit37.6 GB39.9 GB44.2 GB50.0 GB
1.6-bit20.7 GB23.0 GB27.3 GB33.1 GB

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

GPUs and Macs that run GLM 4.5V

The best way GLM 4.5V runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: GLM 4.5V on RTX 4060 report
RTX 5060Too bigPDF: GLM 4.5V on RTX 5060 report
RTX 3070Too bigPDF: GLM 4.5V on RTX 3070 report
RTX 3080 10 GBToo bigPDF: GLM 4.5V on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: GLM 4.5V on RTX 3060 12 GB report
Arc B580Too bigPDF: GLM 4.5V on Arc B580 report
RTX 4070Too bigPDF: GLM 4.5V on RTX 4070 report
RTX 4070 SuperToo bigPDF: GLM 4.5V on RTX 4070 Super report
RTX 5070Too bigPDF: GLM 4.5V on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: GLM 4.5V on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: GLM 4.5V on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: GLM 4.5V on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: GLM 4.5V on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: GLM 4.5V on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: GLM 4.5V on RTX 4080 Super report
RTX 5070 TiToo bigPDF: GLM 4.5V on RTX 5070 Ti report
RTX 5080Too bigPDF: GLM 4.5V on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: GLM 4.5V on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: GLM 4.5V on RTX 3090 report
RTX 4090Too bigPDF: GLM 4.5V on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: GLM 4.5V on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: GLM 4.5V on RTX 5090 report
2× RTX 30902-bit32K91 tok/s1.6 sPDF: GLM 4.5V on 2× RTX 3090 report
2× RTX 40902-bit32K97 tok/s0.7 sPDF: GLM 4.5V on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB2-bit36K118 tok/s1.4 sPDF: GLM 4.5V on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada2-bit36K93 tok/s1.0 sPDF: GLM 4.5V on RTX 6000 Ada report
L40S2-bit36K86 tok/s1.7 sPDF: GLM 4.5V on L40S report
2× RTX 50903-bit56K120 tok/s0.5 sPDF: GLM 4.5V on 2× RTX 5090 report
4× RTX 30906-bit32K48 tok/s0.8 sPDF: GLM 4.5V on 4× RTX 3090 report
RTX PRO 6000 Blackwell6-bit44K83 tok/s0.7 sPDF: GLM 4.5V on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: GLM 4.5V on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: GLM 4.5V on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: GLM 4.5V on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: GLM 4.5V on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: GLM 4.5V on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB2-bit40K33 tok/s31 sPDF: GLM 4.5V on Mac mini M4 Pro 64 GB report
M4 Max 128 GB6-bit44K29 tok/s15 sPDF: GLM 4.5V on M4 Max 128 GB report
M5 Max 128 GB6-bit44K31 tok/s4.4 sPDF: GLM 4.5V on M5 Max 128 GB report
M3 Ultra 96 GB4-bit36K44 tok/s8.6 sPDF: GLM 4.5V on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit64K29 tok/s8.6 sPDF: GLM 4.5V on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit64K29 tok/s8.6 sPDF: GLM 4.5V on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit64K39 tok/s2.7 sPDF: GLM 4.5V on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB2-bit40K26 tok/s8.8 sPDF: GLM 4.5V on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB6-bit44K12 tok/s8.8 sPDF: GLM 4.5V on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit36K13 tok/s3.1 sPDF: GLM 4.5V on RTX Spark report
DGX Spark8-bit36K12 tok/s3.1 sPDF: GLM 4.5V on DGX Spark report
Dual DGX Spark8-bit64K12 tok/s1.8 sPDF: GLM 4.5V on Dual DGX Spark report
DGX Station GB3008-bit64K172 tok/s0.2 sPDF: GLM 4.5V on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit64K67 tok/s0.4 sPDF: GLM 4.5V on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit64K67 tok/s0.2 sPDF: GLM 4.5V on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit64K67 tok/s0.1 sPDF: GLM 4.5V on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM5-bit20K101 tok/s2.0 sPDF: GLM 4.5V on A100 80 GB SXM report
H100 SXM5-bit20K141 tok/s0.6 sPDF: GLM 4.5V on H100 SXM report
4× H200 NVL8-bit64K137 tok/s0.2 sPDF: GLM 4.5V on 4× H200 NVL report
8× H200 NVL8-bit64K137 tok/s0.1 sPDF: GLM 4.5V 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 mradermacher/GLM-4.5V-i1-GGUF
vLLM
vllm serve zai-org/GLM-4.5V
MLX (Mac)
mlx_lm.chat --model mlx-community/Huihui-GLM-4.5V-abliterated-mxfp4

Questions

How much VRAM does GLM 4.5V need?

GLM 4.5V needs 60.8 GB for its weights at 4-bit, 63.1 GB with an 8K context and 67.4 GB with 32K. At 8-bit, 109 GB.

What is the minimum VRAM for GLM 4.5V?

About 39.9 GB at 2-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 2× RTX 3090.

How big is GLM 4.5V?

108B parameters (14.3B active per token): 201 GB of weights at 16-bit, 60.8 GB at 4-bit.

Can GLM 4.5V run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the 2× RTX 3090.

More models