What GPU do you need to run GLM-4.7 Flash?
31B params 128K max context ReasoningTool use
The smallest GPU memory that runs GLM-4.7 Flash well is 20 GB (Q4_K_M, e.g. RTX 4000 Ada). 47 of 132 GPUs run it well.
MMLU 82 SWE-bench Verified 59.2
VRAM needed per quantization
| Quantization | Weights | 16K context | 32K context | Full 128K |
|---|---|---|---|---|
| Q4_K_M | 18.3 GB | 19.1 GB | 19.9 GB | 24.7 GB |
| Q8_0 | 31.8 GB | 32.6 GB | 33.4 GB | 38.2 GB |
Weights plus fp16 KV cache. Leave ~1 GB headroom for the inference engine.
Cheapest GPUs that can run GLM-4.7 Flash
Lowest launch MSRP among GPUs that run its smallest quantization (Q4_K_M) well, with at least 16K context.
As an Amazon Associate, I earn from qualifying purchases.
GPUs that can run GLM-4.7 Flash
Best quantization that fits on each GPU, with the context and speed it gets. For chips with several memory sizes, the smallest that works is shown.
NVIDIA
- RTX 4000 Ada 20 GB Q4_K_M 33K context · ~19 tok/s
- RTX 3090 24 GB Q4_K_M 113K context · ~48 tok/s
- RTX 3090 Ti 24 GB Q4_K_M 113K context · ~52 tok/s
- RTX 4090 24 GB Q4_K_M 113K context · ~52 tok/s
- RTX 4500 Ada 24 GB Q4_K_M 113K context · ~22 tok/s
- RTX A5000 24 GB Q4_K_M 113K context · ~40 tok/s
- RTX PRO 4000 Blackwell 24 GB Q4_K_M 113K context · ~35 tok/s
- RTX Spark 24 GB Q4_K_M 113K context · ~16 tok/s
- Tesla P40 24 GB Q4_K_M 113K context · ~18 tok/s
- RTX 5090 32 GB Q4_K_M 128K context · ~93 tok/s
- RTX 5000 Ada 32 GB Q4_K_M 128K context · ~30 tok/s
- RTX PRO 4500 Blackwell 32 GB Q4_K_M 128K context · ~46 tok/s
- Tesla V100 32GB 32 GB Q4_K_M 128K context · ~47 tok/s
- L40S 48 GB Q8_0 128K context · ~26 tok/s
- RTX 6000 Ada 48 GB Q8_0 128K context · ~29 tok/s
- RTX A6000 48 GB Q8_0 128K context · ~23 tok/s
- RTX PRO 5000 Blackwell 48 GB Q8_0 128K context · ~41 tok/s
- RTX PRO 5000 Blackwell 72GB 72 GB Q8_0 128K context · ~41 tok/s
- RTX PRO 5500 Blackwell 84 GB Q8_0 128K context · ~43 tok/s
- RTX PRO 6000 Blackwell 96 GB Q8_0 128K context · ~55 tok/s
- DGX Spark (GB 10) 128 GB Q8_0 128K context · ~8 tok/s
AMD
- Radeon RX 7900 XT 20 GB Q4_K_M 33K context · ~41 tok/s
- Radeon RX 7900 XTX 24 GB Q4_K_M 113K context · ~50 tok/s
- Radeon AI PRO R9700 32 GB Q4_K_M 128K context · ~33 tok/s
- Ryzen AI Max (Strix Halo) 32 GB Q4_K_M 128K context · ~13 tok/s
- Ryzen AI Max 400 (Gorgon Halo) 64 GB Q8_0 128K context · ~8 tok/s
Intel
- Arc Pro B65 32 GB Q4_K_M 128K context · ~32 tok/s
- Arc Pro B70 32 GB Q4_K_M 128K context · ~32 tok/s
Apple
- M2 24 GB Q4_K_M 113K context · ~5 tok/s
- M3 24 GB Q4_K_M 113K context · ~5 tok/s
- M4 24 GB Q4_K_M 113K context · ~6 tok/s
- M4 Pro 24 GB Q4_K_M 113K context · ~14 tok/s
- M5 24 GB Q4_K_M 113K context · ~8 tok/s
- M5 Pro 24 GB Q4_K_M 113K context · ~16 tok/s
- M6 24 GB Q4_K_M 113K context · ~9 tok/s
- M1 Max 32 GB Q4_K_M 128K context · ~21 tok/s
- M1 Pro 32 GB Q4_K_M 128K context · ~10 tok/s
- M2 Max 32 GB Q4_K_M 128K context · ~21 tok/s
- M2 Pro 32 GB Q4_K_M 128K context · ~10 tok/s
- M3 Max 36 GB Q8_0 83K context · ~9 tok/s
- M3 Pro 36 GB Q8_0 83K context · ~5 tok/s
- M4 Max 36 GB Q8_0 83K context · ~13 tok/s
- M5 Max 36 GB Q8_0 83K context · ~14 tok/s
- M1 Ultra 64 GB Q8_0 128K context · ~24 tok/s
- M2 Ultra 64 GB Q8_0 128K context · ~24 tok/s
- M3 Ultra 96 GB Q8_0 128K context · ~25 tok/s
- M5 Ultra 96 GB Q8_0 128K context · ~37 tok/s