What fits in 128 GB of unified memory
Data updated . Last reviewed the same day, when this catalog snapshot was written.
4 catalog machines have 128 GB unified. Models are ranked against the Framework Desktop 128GB, the tightest usable pool at this sticker size (96.0 GB). The tok/s estimate uses Framework Desktop 128GB bandwidth (256 GB/s) and is not a benchmark. 2 of these fits do not clear the 96 GB tier.
Models
| Model | Best quant | Estimate | Headroom | Est. tok/s |
|---|---|---|---|---|
| Llama 3.2 1B Instruct | FP16 | 3.0 GB | 93.0 GB | 104 tok/s |
| Llama 3.2 3B Instruct | FP16 | 7.7 GB | 88.3 GB | 39.8 tok/s |
| Phi-3.5 mini | FP16 | 9.2 GB | 86.8 GB | 33.5 tok/s |
| Phi-4 mini | FP16 | 9.2 GB | 86.8 GB | 33.4 tok/s |
| Gemma 3 4B | FP16 | 10.3 GB | 85.7 GB | 29.8 tok/s |
| Mistral 7B Instruct v0.3 | FP16 | 17.4 GB | 78.6 GB | 17.7 tok/s |
| Qwen2.5 7B Instruct | FP16 | 18.3 GB | 77.7 GB | 16.8 tok/s |
| Gemma 4 E4B | FP16 | 19.2 GB | 76.8 GB | 16.0 tok/s |
| DeepSeek R1 Distill Llama 8B | FP16 | 19.3 GB | 76.7 GB | 15.9 tok/s |
| Llama 3.1 8B Instruct | FP16 | 19.3 GB | 76.7 GB | 15.9 tok/s |
| Qwen3 8B | FP16 | 19.7 GB | 76.3 GB | 15.6 tok/s |
| Ministral 3 8B | FP16 | 21.4 GB | 74.6 GB | 14.4 tok/s |
| Gemma 2 9B | FP16 | 22.2 GB | 73.8 GB | 13.8 tok/s |
| Gemma 4 12B | FP16 | 28.7 GB | 67.3 GB | 10.7 tok/s |
| Gemma 3 12B | FP16 | 29.2 GB | 66.8 GB | 10.5 tok/s |
| Mistral Nemo 12B | FP16 | 29.4 GB | 66.6 GB | 10.5 tok/s |
| Ministral 3 14B | FP16 | 33.5 GB | 62.5 GB | 9.18 tok/s |
| Phi-4 | FP16 | 35.2 GB | 60.8 GB | 8.73 tok/s |
| Qwen3 14B | FP16 | 35.4 GB | 60.6 GB | 8.67 tok/s |
| Qwen2.5 14B Instruct | FP16 | 35.4 GB | 60.6 GB | 8.67 tok/s |
| gpt-oss-20b | FP16 | 50.2 GB | 45.8 GB | 35.6 tok/s |
| Mistral Small 3.2 24B | FP16 | 57.6 GB | 38.4 GB | 5.33 tok/s |
| Gemma 4 26B-A4B | FP16 | 61.9 GB | 34.1 GB | 33.7 tok/s |
| Gemma 2 27B | FP16 | 65.3 GB | 30.7 GB | 4.70 tok/s |
| Gemma 3 27B | FP16 | 65.8 GB | 30.2 GB | 4.67 tok/s |
| Qwen3.6 27B | FP16 | 66.7 GB | 29.3 GB | 4.61 tok/s |
| Qwen3.8 27B | FP16 | 66.7 GB | 29.3 GB | 4.61 tok/s |
| Qwen3 30B-A3B | FP16 | 73.3 GB | 22.7 GB | 42.7 tok/s |
| GLM-4.7-Flash | FP16 | 74.9 GB | 21.1 GB | 42.7 tok/s |
| Gemma 4 31B | FP16 | 75.1 GB | 20.9 GB | 4.09 tok/s |
| Qwen3 32B | FP16 | 78.6 GB | 17.4 GB | 3.91 tok/s |
| DeepSeek R1 Distill Qwen 32B | FP16 | 78.6 GB | 17.4 GB | 3.91 tok/s |
| Qwen2.5 32B Instruct | FP16 | 78.6 GB | 17.4 GB | 3.91 tok/s |
| Qwen3.6 35B-A3B | FP16 | 86.3 GB | 9.7 GB | 42.7 tok/s |
| DeepSeek V4 Flash | Q8 | 4.5 GB | 91.5 GB | 1505 tok/s |
| Mixtral 8x7B | Q8 | 50.7 GB | 45.3 GB | 18.7 tok/s |
| gpt-oss-120b | Q8 | 63.7 GB | 32.3 GB | 92.5 tok/s |
| Llama 3.3 70B Instruct | Q8 | 75.0 GB | 21.0 GB | 3.41 tok/s |
| DeepSeek R1 Distill Llama 70B | Q8 | 77.7 GB | 18.3 GB | 3.41 tok/s |
| Qwen2.5 72B Instruct | Q8 | 80.2 GB | 15.8 GB | 3.30 tok/s |
| Llama 4 Scout | Q4 | 65.4 GB | 30.6 GB | 25.0 tok/s |
| Mistral Small 4 | Q4 | 74.0 GB | 22.0 GB | 63.8 tok/s |
| Mixtral 8x22B | Q4 | 87.5 GB | 8.5 GB | 10.8 tok/s |
Machines at 128 GB
- Framework Desktop 128GB
96.0 GB usable · $3,449
- GMKtec EVO-X2 128GB
96.0 GB usable · $3,499
- Minisforum MS-S1 Max 128GB
96.0 GB usable · $3,879
- Beelink GTR9 Pro 128GB
96.0 GB usable · $4,349
Questions
- How many models fit in 128 GB of unified memory?
- 43 of 49 fit on the tightest config at this size, the Framework Desktop 128GB, which keeps 96.0 GB usable.
- Which catalog machines have 128 GB unified?
- Framework Desktop 128GB (75% usable), GMKtec EVO-X2 128GB (75% usable), Minisforum MS-S1 Max 128GB (75% usable), Beelink GTR9 Pro 128GB (75% usable).
- Why is the ranking based on the tightest config?
- So a model listed as fitting works on every 128 GB machine here, including ones that reserve more memory for the OS. A looser machine has more headroom than the table shows.