GLM · 106.9B parameters · 12B active
GLM-4.5-Air: GPU memory requirements and which GPU to rent
Zhipu's GLM-4.5-Air has 107 billion parameters with 12 billion active per token, 46 layers and grouped-query attention — a mid-size mixture of experts that fits one 141 GB GPU at FP8.
- Weights at FP8107 GB214 GB at 16-bit · 59 GB at 4-bit
- FP8, 8k context, 1 sequence114 GBweights + KV cache + runtime
- Suggested here2× A100$1,298 a month · dedicated
Inference
How much GPU memory GLM-4.5-Air needs to serve
Total memory for one sequence at each context length, and for 16 concurrent sequences at 8k — weights, KV cache at 16-bit and runtime overhead included.
| Weights precision | Weights | 4k context | 32k context | 128k context | 8k × 16 sequences | Suggested here, 8k |
|---|---|---|---|---|---|---|
| 16-bit (BF16 / FP16) | 214 GB | 225 GB | 230 GB | 249 GB | 249 GB | 4× A100 · $2,596/mo |
| 8-bit (FP8 / INT8) | 107 GB | 113 GB | 119 GB | 137 GB | 137 GB | 2× A100 · $1,298/mo |
| 4-bit (GPTQ, AWQ, Q4) | 59 GB | 63 GB | 69 GB | 87 GB | 87 GB | 1× A100 · $649/mo |
GB are 109 bytes. Context lengths are shown for scale where the model supports them; an FP8 KV cache halves the cache part of each figure. As a mixture of experts, every expert stays resident, so memory follows the total parameter count, not the active one.
Hardware
GPU servers that run GLM-4.5-Air
For FP8 inference with an 8k context and one sequence: the configurations we would suggest, in order — one large card ahead of several small ones.
Fine-tuning GLM-4.5-Air
Memory for a 2k-token sequence and a micro-batch of 4, with gradient checkpointing. Full fine-tuning keeps gradients and Adam states for every parameter; LoRA and QLoRA train small adapters on a frozen base.
Architecture at a glance
Grouped-query attention: every layer keeps keys and values for every token of the context. Figures from the model's published configuration; method on the calculator page.
GLM-4.5-Air hardware: questions answered
How much VRAM does GLM-4.5-Air need?
For the weights alone: about 214 GB at 16-bit, 107 GB at FP8 and 59 GB at 4-bit. Serving it at FP8 with an 8k-token context and one sequence takes about 114 GB in total once the KV cache and the runtime are counted; long contexts and many concurrent users add the KV cache on top — about 0.19 GB per 1,000 tokens at 16-bit.
Can GLM-4.5-Air run on a single GPU?
Yes, at FP8 with a moderate context: one NVIDIA H200 (141 GB HBM3e) holds it, for $1,690 a month here. Longer contexts or more concurrent users may need a larger card or a second GPU.
What is the cheapest GPU setup to run GLM-4.5-Air?
For FP8 inference with an 8k context and one sequence, the configuration we would suggest is 2× A100 · $1,298/mo; at 4-bit, 1× A100 · $649/mo. Suggestions favour one large card over several small ones: splitting a model across PCIe cards costs speed.
How much GPU memory do I need to fine-tune GLM-4.5-Air?
With a 2k-token sequence and a micro-batch of 4: about 74 GB for QLoRA on a 4-bit base, 235 GB for LoRA on a 16-bit base, and 1,727 GB for a full fine-tune with Adam in mixed precision. Gradient checkpointing is assumed; sharding the optimizer across GPUs spreads the full fine-tune over several cards.
Run GLM-4.5-Air on 2× A100 for $1,298 a month
Dedicated, single-tenant and live in under 10 minutes — with vLLM, the Hugging Face stack or Ollama preinstalled if you want. Paid in BTC, ETH or USDT, with no KYC.