Intel Linux installation paths
Distinguish the pinned OMIX stack from the Ubuntu PPA route. Intel warns against mixing their packages.
dgpu-docs.intel.comKeep the primary sources close / 05
Go straight to the source.
A working shelf for setting up, making things, and understanding what your B70 is doing.
26 SOURCES / FIND YOUR NEXT STEP
SOURCE REVIEW
Upstream documentation, experimental projects, and community reports are labeled separately. Linked capabilities have not all been tested on our B70s.
01 / setup
Choose your foundation.
Hardware identity, host packages, and application dependencies.
OMIX support matrix
Release-specific hardware, host, container and component versions. Read the host and container columns separately.
dgpu-docs.intel.comB70 hardware specifications
Specifications from Intel. Your board design and negotiated PCIe link still need their own evidence.
intel.comXe-driver hardware support
Identify B70 and check the driver and distro support table before assembling an environment.
dgpu-docs.intel.comPyTorch on Intel GPUs
Native XPU installation and framework notes. Match packages to your operating system and application.
docs.pytorch.orgchipStar
HIP/CUDA source through SPIR-V. Read the project limitations before treating translation as application compatibility.
github.com02 / inference
Run models and serve requests.
Follow the backend and release that match the workload you need.
llama.cpp SYCL
Builds, device selection and backend restrictions for GGUF inference. Use instructions matching your revision.
github.comIntel LLM Scaler
Serving recipes and release notes for B60/B70. Includes embedding, reranking, speech and OCR workload leads.
github.comvLLM XPU installation
Intel-specific images, wheels and dependencies. Read this source alongside your chosen release documentation.
github.comOpenVINO
Model conversion and deployment toolkit. Check device, operation and precision support for your pipeline.
docs.openvino.aiOllama GPU support
Documents an optional Vulkan route. Evaluate model behavior and device selection independently of SYCL results.
docs.ollama.com03 / creative
Make something beyond text.
Image and video workflows, graphical tools, and model preparation.
Intel LLM Scaler Omni
Current b2 instructions target a source-built single-XPU preview, not a published b2 image.
github.comOmni ComfyUI guide
Workflow inputs, output persistence and memory options for the matching Omni environment.
github.comComfyUI requirements
Native Intel XPU requirements. Individual custom nodes can introduce additional compatibility limits.
docs.comfy.orgIntel AI Playground
A graphical option for chat and creative workloads. Check the selected workflow and platform requirements.
github.comIntel AutoRound
Prepare quantized models on supported hardware. Export format, quality and efficient inference support are separate checks.
github.com04 / diagnostics
Understand the machine.
Find drivers, monitor a workload, and get board-specific help.
Intel compute-runtime
Level Zero and OpenCL compute releases and issue tracking. Capture loader and runtime versions together.
github.comnvtop
GPU/process monitoring with Intel Xe support. Metrics depend on driver access and do not establish whole-system power.
github.comB70 driver downloads
Find the appropriate Windows driver track and check board/OEM guidance before choosing a package.
intel.comECC and reported VRAM
Intel explains a Windows case showing 28 GB on a 32 GB B70. Read the scope and error-protection tradeoff.
intel.comB70 support hub
Hardware FAQs and support links. Our troubleshooting log keeps version-specific reports and unresolved outcomes visible.
intel.com05 / community
Follow the experiments.
Independent work with methods worth reading. Compare each configuration before comparing its numbers.
PMZFX benchmark notebook
Public source for our April dataset. The reviewed explorer retains corrections and metadata gaps.
github.comIntel XPU Optimization Lab
Unofficial recipes, patches and model-specific tuning. Read quality qualifications and local-change details.
github.comIntel Arc GPU Benchmarks
B70 and B580 configurations, including different hosts and backends. Keep their test cohorts separate.
github.comTRX50 / B70 benchmarks
Threadripper, B70/B60 and wall-power work. Inspect measurement windows before interpreting energy ratios.
github.comIntel GPU performance discussion
Reproducible reports and maintainer feedback in llama.cpp. Discussion results need their own environment context.
github.com