Hardware
CPU only · DDR5-6400
CPU only · DDR5-6400 has 8 / 16 / 32 / 64 / 128 GB System RAM and 102.4 GB/s. See which open models fit, at which quant, and roughly how fast. Figures below use the default memory (32 GB). Updated 2026-10-24. Estimate from memory bandwidth, not a benchmark.
Dual-channel DDR5-6400: 2 × 6.4 GT/s × 8 bytes = 102.4 GB/s. Source.
RSS for new models on this device
Models that fit
| Model | Verdict | Quant | Context | tok/s |
|---|---|---|---|---|
| gemma 4 E4B it qat 7.5B | Runs on CPU | UD-Q4_K_XL | 8K | 11–16 est. |
| Sharp Spark X2.5 4B 4.1B | Runs on CPU | Q6_K | 8K | 12–16 est. |
| NVIDIA Nemotron 3 Nano 4B 4.0B | Runs on CPU | Q8_0 | 8K | 10.0–14 est. |
| LFM2.5 2.6B 2.7B | Runs on CPU | BF16 | 8K | 9.5–13 est. |
| LFM2.5 VL 3B 2.7B | Runs on CPU | BF16 | 8K | 9.5–13 est. |
| Qwen3.8 2B Distill 1.9B | Runs on CPU | BF16 | 8K | 14–19 est. |
| Qwen3.5 2B 1.9B | Runs on CPU | BF16 | 8K | 15–20 est. |
| DeepSeek R1 Distill Qwen 1.5B 1.8B | Runs on CPU | BF16 | 8K | 15–20 est. |
| Qwen3 1.7B 1.7B | Runs on CPU | BF16 | 8K | 13–18 est. |
| Qwen3 VL 2B Instruct 1.7B | Runs on CPU | BF16 | 8K | 13–18 est. |
| Qwen3 VL 2B Thinking 1.7B | Runs on CPU | BF16 | 8K | 13–18 est. |
| Qwen2.5 1.5B Instruct 1.5B | Runs on CPU | F16 | 8K | 17–23 est. |
| Qwen2.5 Coder 1.5B Instruct 1.5B | Runs on CPU | F16 | 8K | 17–23 est. |
| Llama 3.2 1B Instruct 1.2B | Runs on CPU | BF16 | 8K | 20–28 est. |
| LFM2.5 1.2B Instruct 1.2B | Runs on CPU | BF16 | 8K | 22–29 est. |
| gemma 3 1b it 1.0B | Runs on CPU | BF16 | 8K | 25–35 est. |
| Qwen3.5 0.8B 0.75B | Runs on CPU | BF16 | 8K | 35–47 est. |
| Qwen3 0.6B 0.60B | Runs on CPU | BF16 | 8K | 26–36 est. |
| Qwen2.5 0.5B Instruct 0.49B | Runs on CPU | F16 | 8K | 51–70 est. |
| Qwen2.5 32B Instruct 32.8B | Runs on CPU | Q5_K_M | 8K | 2.2–3.0 est. |
| Qwen3 32B 32.8B | Runs on CPU | UD-Q5_K_XL | 8K | 2.2–3.0 est. |
| Qwen2.5 Coder 32B Instruct 32.8B | Runs on CPU | Q5_K_M | 8K | 2.2–3.0 est. |
| Qwen3 VL 32B Instruct 32.8B | Runs on CPU | UD-Q5_K_XL | 8K | 2.2–3.0 est. |
| DeepSeek R1 Distill Qwen 32B 32.8B | Runs on CPU | Q5_K_M | 8K | 2.2–3.0 est. |
| GLM 4 32B 0414 32.6B | Runs on CPU | UD-Q5_K_XL | 8K | 2.4–3.3 est. |
| gemma 4 31B it 30.7B | Runs on CPU | UD-Q5_K_XL | 8K | 1.9–2.6 est. |
| Swift Qwen3.8 27B 27.3B | Runs on CPU | Q5_K_M | 8K | 2.7–3.7 est. |
| Qwen3.8 27B 27.3B | Runs on CPU | UD-Q6_K_XL | 8K | 2.2–3.0 est. |
| gemma 3 27b it 27.0B | Runs on CPU | UD-Q6_K_XL | 8K | 2.0–2.8 est. |
| medgemma 27b text it 27.0B | Runs on CPU | UD-Q6_K_XL | 8K | 2.0–2.8 est. |
| Qwen3.6 27B 26.9B | Runs on CPU | Q6_K | 8K | 2.4–3.3 est. |
| Qwen3.5 27B 26.9B | Runs on CPU | Q6_K | 8K | 2.4–3.3 est. |
| Kwaipilot KAT Coder V2.5 Dev 34.7B | Runs on CPU | Q5_K_M | 8K | 5.4–7.4 est. |
| Qwen3.8 35B A3B Distill 35.5B | Runs on CPU | Q5_K_M | 8K | 5.4–7.3 est. |
| Cyber Tiel Coder 35B A3B GGUF MTP 35.5B | Runs on CPU | Q5_K_M (est.) | 8K | 5.4–7.3 est. |
| Qwen3.6 35B A3B 34.7B | Runs on CPU | UD-Q5_K_S | 8K | 5.5–7.4 est. |
| Qwen3.5 35B A3B 34.7B | Runs on CPU | Q5_K_S | 8K | 5.5–7.5 est. |
| Devstral Small 2 24B Instruct 2512 23.6B | Runs on CPU | Q8_0 | 4K | 2.2–3.0 est. |
| Mistral Small 3.2 24B Instruct 2506 23.6B | Runs on CPU | Q8_0 | 4K | 2.2–3.0 est. |
| Nemotron 3 Nano 30B A3B 31.6B | Runs on CPU | Q5_K_S | 8K | 5.5–7.5 est. |
94 more models in the catalog do not fit this default profile, or ranked below the first 40 that do.
Estimate from memory bandwidth, not a benchmark. Method.
Example: Runs on CPU on your CPU only · DDR5-6400 at UD-Q4_K_XL, 8K context, about 11–16 tok/s. Weights 3.93 GB plus 0.14 GB of KV cache — 4.41 GB of 25.2 GB usable. Up to 128K context still fits. Weights shown in the verdict use GiB (the same 1024³ counting as GPU memory).