Skip to content
Run LLM Locally

GLM-4.5-Air on the RTX 5090

GLM-4.5-Air at Q4_K_M needs about 73.7 GB, more than the 32 GB of the RTX 5090: part of the model runs from system RAM, which slows generation. At Q3_K_M it does not fit in VRAM.

ShareFacebookXWhatsAppLinkedInEmail

GLM-4.5-Air

Q4_K_M · 8K context

Possible but slow

The model needs about 73.7 GB but only 31 GB of VRAM is usable: 42.7 GB must be loaded into system RAM. It runs, 42% on the GPU, but more slowly.

Your hardware
RTX 5090 32 GB · 64 GB RAM
Estimated memory
~73.7 GB estimate
VRAM (usable memory)
31 GB
GPU only
No
CPU/GPU offload
Required (42.7 GB)
Generation speed
3.5–7 tok/s estimate
Largest context in fast memory
None
Confidence: low

Based on: memory bandwidth (1,792 GB/s) and model size · system RAM bandwidth (~51 GB/s)

Memory breakdown

~73.7 GB (71.9 GB–75.5 GB) needed · estimate

42.7 GB must be offloaded to system RAM

  • Model weights68 GB
  • KV cache1.4 GB
  • Runtime overhead0.8 GB
  • Safety margin3.5 GB
Assumptions: context 8K tokens · KV cache f16 · llama.cpp (CUDA) · flash attention on · weights = real file size

EstimateData updated 40 days agoSource: Run LLM Locally engine, Hugging Face config.json + model card

What makes this estimate less reliable

  • no benchmark for this hardware and model yet
  • split CPU/GPU runs vary a lot with the CPU and PCIe link
  • system RAM speed unknown (dual-channel DDR4-3200 assumed)
  • mixture-of-experts: speed depends on expert routing

Other quantizations on the RTX 5090

GLM-4.5-Air on RTX 5090 + 64 GB RAM · 8K context
GLM-4.5-Air Q8_08K~117 GB · No dataIncompatible
GLM-4.5-Air Q6_K8K~99.2 GB · No dataIncompatible
GLM-4.5-Air Q5_K_M8K~84 GB · No dataIncompatible
GLM-4.5-Air Q4_K_M8K~73.7 GB · 3.5–7 tok/s estimatePossible but slow
GLM-4.5-Air IQ4_XS8K~61.5 GB · 5–10 tok/s estimateCPU/GPU offload
GLM-4.5-Air Q3_K_M8K~58.3 GB · 5.5–11 tok/s estimateCPU/GPU offload

Alternatives

GPUs that run GLM-4.5-Air Q4_K_M entirely in VRAM

Other models that fit the RTX 5090

Hardware that runs GLM-4.5-Air Q4_K_M entirely in memory

Prices and links follow the market you choose.

Radeon PRO W7900

AMD

ExcellentRuns GLM-4.5-Air Q4_K_M entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

NVIDIA DGX Spark (GB10)

Unified memory PC

ExcellentRuns GLM-4.5-Air Q4_K_M entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

No current offer imported for this market: the link opens a marketplace search.

Frequently asked questions

Can the RTX 5090 run GLM-4.5-Air?
The model needs about 73.7 GB but only 31 GB of VRAM is usable: 42.7 GB must be loaded into system RAM. It runs, 42% on the GPU, but more slowly.
Which quantization should I use on the RTX 5090?
None fits entirely in VRAM at 8K context; the smallest, Q3_K_M, still needs offload.
What is the largest context on the RTX 5090?
At Q4_K_M, the model does not fit entirely in VRAM even at a short context.
How fast is GLM-4.5-Air (RTX 5090)?
We estimate 3.5–7 tok/s (estimate, confidence low). Based on: memory bandwidth (1,792 GB/s) and model size, system RAM bandwidth (~51 GB/s).