Intel Arc Pro B70 for local LLMs
8 measured results from 3 builds, each linked to its source.
| Memory | 32 GB per card |
|---|---|
| Memory bandwidth | not listed |
| Typical EU price | not tracked yet |
Can it run? Gemma 4 26B-A4B on Intel Arc Pro B70 Ornith 1.5 35B-A3B on Intel Arc Pro B70 Qwen3.6 35B-A3B on Intel Arc Pro B70 Qwen3.8 27B on 2x Intel Arc Pro B70 Qwen3.8 27B on Intel Arc Pro B70 Qwen3.8 Flash-Next on 2x Intel Arc Pro B70
All builds with the Intel Arc Pro B70
| Build | GPU(s) | Memory GB (total) | Results |
|---|---|---|---|
| Dual Intel Arc Pro B70 (neural.download lab) | 2x Intel Arc Pro B70 32 GB | 64 | 2 |
| Single Intel Arc Pro B70 32 GB (neural.download lab) | Intel Arc Pro B70 32 GB | 32 | 3 |
| Single Intel Arc Pro B70 with vLLM XPU cookbook | Intel Arc Pro B70 32 GB | 32 | 3 |
All measured results
(MTP, DFlash, DSpark and similar draft-and-verify methods give much higher tok/s on predictable text and are not comparable 1:1 with plain decode)
| Model | Build | Ran on | Quant | Engine | tok/s | Details |
|---|---|---|---|---|---|---|
| Gemma 4 26B-A4B | Single Intel Arc Pro B70 32 GB (neural.download lab) | Intel Arc Pro B70 | UD-Q8_K_XL | llama.cpp SYCL (pinned build, oneAPI 2026.0) | 122.16 | 📏 Single-user, cache-cold decode median (lab package/repro README) |
| Ornith 1.5 35B-A3B | Single Intel Arc Pro B70 32 GB (neural.download lab) | Intel Arc Pro B70 | Q4_K_M | llama.cpp SYCL + neural.download fusion patches; author marks this headline 'pending' | 131.46 | ⏳ author marks pending📏 Single-user, cache-cold decode median (lab package/repro README) |
| Qwen3.6 35B-A3B | Single Intel Arc Pro B70 with vLLM XPU cookbook | Intel Arc Pro B70 | GPTQ Int4 | vLLM XPU b70 image, no MTP | 120 | context: short prompt📏 Greedy, one request at a time, 512-token answers (README "Numbers" table) |
| Qwen3.8 27B | Single Intel Arc Pro B70 with vLLM XPU cookbook | Intel Arc Pro B70 | GPTQ Int4 (asym) | vLLM XPU b70 image + MTP, reasoning prompt | 96 | ⚡ spec-decode (MTP)context: short prompt📏 Greedy, one request at a time, 512-token answers (README "Numbers" table) |
| Qwen3.8 27B | Single Intel Arc Pro B70 with vLLM XPU cookbook | Intel Arc Pro B70 | GPTQ Int4 (asym) | vLLM XPU b70 image + MTP, prose prompt | 66 | ⚡ spec-decode (MTP)context: short prompt📏 Greedy, one request at a time, 512-token answers (README "Numbers" table) |
| Qwen3.8 27B | Dual Intel Arc Pro B70 (neural.download lab) | 2x Intel Arc Pro B70 | FP8 | vLLM XPU, TP2, MTP 5 | 46.01 | ⚡ spec-decode (MTP)📏 Lab README/scoreboard |
| Qwen3.8 27B | Single Intel Arc Pro B70 32 GB (neural.download lab) | Intel Arc Pro B70 | Q4_K_M | llama.cpp SYCL + neural.download TP1 stack, no speculation | 27.8 | 📏 Single-user, cache-cold decode median (lab package/repro README) |
| Qwen3.8 Flash-Next | Dual Intel Arc Pro B70 (neural.download lab) | 2x Intel Arc Pro B70 | IQ3 | neural.download runtime, TP2 | 58.56 | 📏 Lab README/scoreboardpartly in system RAM |
All Intel Arc Pro B70 builds: see also all GPUs · best LLMs by memory size · download the dataset.
FAQ
How fast is the Intel Arc Pro B70 for local LLMs?
The fastest plain-decode result we have is 122.16 tok/s on Gemma 4 26B-A4B (UD-Q8_K_XL, llama.cpp SYCL (pinned build, oneAPI 2026.0)) in Single Intel Arc Pro B70 32 GB (neural.download lab). Speed depends heavily on the model, quant and engine, so compare like with like in the table. RigBase lists 8 measured results from 3 builds.
How much memory does the Intel Arc Pro B70 have?
32 GB per card.
Which models have been measured on the Intel Arc Pro B70?
Gemma 4 26B-A4B, Ornith 1.5 35B-A3B, Qwen3.6 35B-A3B, Qwen3.8 27B, Qwen3.8 Flash-Next.
Where do these numbers come from?
Every row links to a build page with its original public source (benchmark logs, READMEs, forum posts). RigBase does not re-run the benchmarks. Results marked ⚡ use speculative decoding and are not comparable 1:1 with plain decode.
Referral link: if you sign up with Google and add $10, Runpod gives you and RigBase bonus credit (terms). It never affects our data.