Intel Arc Pro B70 vs RTX 5090
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 B7018 sourced runs · best 283.1 tok/s
RTX 509027 verified runs · best 446.7 tok/s
Specs and evidence
| Intel Arc Pro B70 | RTX 5090 | |
|---|---|---|
| VRAM | 32 GB | 32 GB |
| Sourced runs on the board | 18 | 27 |
| Models measured | 16 | 17 |
| Best output tok/s | 283.1 tok/s | 446.7 tok/s |
| Highest measured peak VRAM | 23.2 GB | 31.3 GB |
| Street price (latest snapshot) | $1,300 | — |
Head-to-head
On the 6 models measured on both cards, Intel Arc Pro B70 leads on 2, RTX 5090 on 4. The widest gap is on Qwen3.8-27B: Intel Arc Pro B70 27.8 tok/s vs RTX 5090 435.1 tok/s (RTX 5090 1464% faster).
The measured matrix
| Model | Intel Arc Pro B70 | RTX 5090 |
|---|---|---|
| NVIDIA-Nemotron-3.5-Lightning-30B-A3B | 186.6 | 446.7 |
| Qwen3.8-27B | 27.8 | 435.1 |
| Ornith-1.0-35B-PrismaAURA-4.75bit-vllm-MTP | — | 283.1 |
| Laguna-XS-2.1 | 283.1 | 216.7 |
| Kimi-Linear-48B-A3B-Instruct | — | 280.8 |
| Ornith-1.5-35B-A3B | 131.5 | 258.8 |
| North-Mini-Code-1.0 | — | 258.4 |
| Qwen3.6-35B-A3B | 248.3 | — |
| Ornith-1.0-35B-B70-Turbo | 224.9 | — |
| Recursive Flint 40B | — | 224.8 |
| Nex-N2-mini-B70-Turbo | 224.3 | — |
| SIQ-1-35B-B70-Turbo | 223.7 | — |
| Qwen3.6-35B-A3B-MTP | — | 222.9 |
| Qwen-AgentWorld-35B-A3B-B70-Turbo | 209.0 | — |
| KAT-Coder-V2.5-Dev-AWQ-W4A16-ASYM | — | 201.2 |
| Ornith-1.0-35B | — | 196.6 |
| LFM2.5-2.6B | 132.4 | — |
| Muse-Glimmer-30B | 29.2 | 124.8 |
| Qwen3.6-27B | — | 120.9 |
| Huihui-Qwen3.8-27B-abliterated | — | 117.7 |
| Qwen3.8-27B-5090-goldilocks | — | 93.5 |
| Gemma4-26B-A4B-QAT-Uncensored-HauhauCS-Balanced-MTP | 58.4 | — |
| Ornith-1.5-9B | 49.6 | — |
| Qwen3.6-27B-MTP | 36.0 | 28.3 |
| ThinkingCap-Qwen3.6-27B | 27.9 | — |
| DeepSeek-V4-Flash-0731 | — | 26.4 |
| Qwen3.8-Flash-Next | 24.0 | — |