CMP 170HX vs Intel Arc Pro B70
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.
CMP 170HX16 sourced runs · best 255.9 tok/s
Intel Arc Pro B7018 verified runs · best 283.1 tok/s
Specs and evidence
| CMP 170HX | Intel Arc Pro B70 | |
|---|---|---|
| VRAM | 64 GB | 32 GB |
| Sourced runs on the board | 16 | 18 |
| Models measured | 11 | 16 |
| Best output tok/s | 255.9 tok/s | 283.1 tok/s |
| Highest measured peak VRAM | 58.7 GB | 23.2 GB |
| Street price (latest snapshot) | — | $1,300 |
Head-to-head
On the 6 models measured on both cards, CMP 170HX leads on 3, Intel Arc Pro B70 on 3. The widest gap is on Qwen3.8-27B: CMP 170HX 255.9 tok/s vs Intel Arc Pro B70 27.8 tok/s (CMP 170HX 820% faster).
The measured matrix
| Model | CMP 170HX | Intel Arc Pro B70 |
|---|---|---|
| Laguna-XS-2.1 | — | 283.1 |
| Qwen3.8-27B | 255.9 | 27.8 |
| Qwen3.6-35B-A3B | 87.2 | 248.3 |
| Ornith-1.0-35B-B70-Turbo | — | 224.9 |
| Nex-N2-mini-B70-Turbo | — | 224.3 |
| SIQ-1-35B-B70-Turbo | — | 223.7 |
| Qwen-AgentWorld-35B-A3B-B70-Turbo | — | 209.0 |
| NVIDIA-Nemotron-3.5-Lightning-30B-A3B | 165.1 | 186.6 |
| LFM2.5-2.6B | — | 132.4 |
| Ornith-1.5-35B-A3B | 87.1 | 131.5 |
| Ornith-1.5-9B | 61.5 | 49.6 |
| Gemma4-26B-A4B-QAT-Uncensored-HauhauCS-Balanced-MTP | — | 58.4 |
| Qwen3.8-27B-MTP | 46.4 | — |
| Qwen3.8-Flash-Next | 46.1 | 24.0 |
| Llama-4-Scout-17B-16E-Instruct | 44.5 | — |
| Qwen3.6-27B-MTP | — | 36.0 |
| Muse-Glimmer-30B | — | 29.2 |
| gemma-4-31B-it | 28.3 | — |
| ThinkingCap-Qwen3.6-27B | — | 27.9 |
| granite-4.2-30b | 23.7 | — |
| Llama-3.3-70B-Instruct | 20.1 | — |