Not another fork. Superb LLM inference server for Intel Arc Pro B70 | 50+% vLLM 🔥| 5x llama.cpp 💩
-
Updated
Nov 21, 2026 - C++
Not another fork. Superb LLM inference server for Intel Arc Pro B70 | 50+% vLLM 🔥| 5x llama.cpp 💩
Community-maintained llama.cpp SYCL Docker optimized for Intel Arc Pro B70 (Battlemage). Latest oneAPI/compute-runtime/IGC, B70-specific AOT build, all features enabled (Flash Attention, MTP, reorder kernels, no GGML_SYCL_DISABLE_OPT).
b70ctl — TUI that installs and serves qualified LLM model packs on Intel Arc Pro B70 GPUs (vLLM/XPU): Qwen3.8 27B, Qwen3.6 35B A3B, Qwen3.8 Flash Next, MTP, DFlash2
Pioneer documentation: OpenFOAM CFD GPU offloading on Intel Arc Pro B70 Pro (Battlemage Xe2). Hardware excellent (FP64 96%, kernel latency CUDA-par), software stack not ready (GPU 66% idle, no working strong preconditioner in Ginkgo 1.10 SYCL). 14 bugs documented + direct VRAM/profiling data.
To associate your repository with the b70 topic, visit your repo's landing page and select "manage topics."