Intel Arc Pro B70 x4 vs RTX 3060 x2
Best sourced output tok/s per model (single-stream generation) — engine and version, quantization, context and a source link on every row. Hover a value for the quant.
Intel Arc Pro B70 x44 sourced runs · best 161.9 tok/s
RTX 3060 x229 verified runs · best 198.1 tok/s
Specs and evidence
| Intel Arc Pro B70 x4 | RTX 3060 x2 | |
|---|---|---|
| VRAM | 32 GB | 12 GB |
| Sourced runs on the board | 4 | 29 |
| Models measured | 4 | 29 |
| Best output tok/s | 161.9 tok/s | 198.1 tok/s |
| Highest measured peak VRAM | — | 21.7 GB |
| Street price (latest snapshot) | — | — |
Head-to-head
Only one model has been measured on both cards (Muse-Glimmer-30B). On that single data point Intel Arc Pro B70 x4 records 161.9 tok/s and RTX 3060 x2 records 56.0 tok/s — treat it as one data point, not a ranking.
The measured matrix
| Model | Intel Arc Pro B70 x4 | RTX 3060 x2 |
|---|---|---|
| Qwen3-1.7B | — | 198.1 |
| Muse-Glimmer-30B | 161.9 | 56.0 |
| MiniCPM5-1B | — | 125.9 |
| Laguna-S-2.1 | 125.5 | — |
| Trinity-Mini | — | 109.0 |
| Phi-4-mini-reasoning | — | 108.2 |
| gemma-4-26B-A4B-it-qat | — | 95.3 |
| MiniMax-M2.7 | 94.5 | — |
| Qwen3-30B-A3B-Thinking-2507 | — | 94.3 |
| gpt-oss-20b | — | 93.4 |
| DeepSeek-V4-Flash-180B | 80.8 | — |
| OpenReasoning-Nemotron-7B | — | 68.0 |
| WebWorld-8B | — | 62.3 |
| gemma-4-26B-A4B-it | — | 62.1 |
| Carnice-9b | — | 56.1 |
| NVIDIA-Nemotron-Nano-12B-v2 | — | 40.4 |
| gemma-3-12b-it | — | 39.4 |
| Qwen3-14B | — | 35.8 |
| Nemotron-Cascade-14B-Thinking | — | 35.7 |
| Phi-4-reasoning | — | 35.4 |
| WebWorld-14B | — | 35.4 |
| InternVL3-14B | — | 35.1 |
| DeepSeek-R1-Distill-Qwen-14B | — | 35.1 |
| phi-4 | — | 34.2 |
| DeepHermes-3-Mistral-24B-Preview | — | 24.5 |
| reka-flash-3 | — | 23.9 |
| mistralai_Mistral-Small-3.2-24B-Instruct-2506 | — | 21.8 |
| Qwen3.6-27B | — | 18.9 |
| Tess-4-27B | — | 18.8 |
| GLM-Z1-32B-0414 | — | 16.6 |
| Olmo-3-32B-Think | — | 16.4 |
| gemma-4-31B-it | — | 15.8 |