Add Arc native Windows support

README.md

@@ -65,7 +65,7 @@ The installer sets up everything else. Two or three cards can share the model ([
 Experimental, written and tested by community members on their own machines:
 
 - **Older graphics cards** (Tesla P40 / V100, GTX 10, Radeon VII / MI50, RX 6700 XT, RX 5500 XT): [Older GPUs](docs/OLDER_GPUS.md).
-- **Intel Arc**, built from source on Linux: [Intel Arc](docs/INTEL_ARC.md).
+- **Intel Arc**, built from source on Linux or Windows: [Intel Arc](docs/INTEL_ARC.md).
 - **AMD Ryzen AI Max (Strix Halo)**, built from source on Linux: [Strix Halo](docs/STRIX_HALO.md).
 - **Older processors without AVX2**: they work, but slowly. [Older CPUs](docs/INSTALL.md#older-cpus-experimental).
 

docs/INTEL_ARC.md

@@ -9,7 +9,8 @@ The code is in `sycl/`, and the port's own notes, measurements and maintenance p
 we have not run it on an Arc. Every result on Arc hardware below comes from the community. The NVIDIA and AMD
 engines are unchanged: the Intel build is a separate CMake target, off by default.
 
-There is **no ready-made Intel engine** in the release zips. You build it from source on Linux.
+There is **no ready-made Intel engine** in the release zips. You build it from source, on Linux (this page) or on
+Windows ([INTEL_WINDOWS.md](INTEL_WINDOWS.md)).
 
 ## What has been run, and by whom
 
@@ -48,9 +49,9 @@ Full flags, per-request timings and engine logs are in the report linked in the
 - **Unchanged:** the CUDA engine's greedy output, checked byte-identical against the gated 0.1.39 build (Q2_0 and Coder).
   The HIP build is not affected (the option is off by default).
 
-**Not tested by anyone yet:** Windows (no native build path; see below), Arc on WSL2 for the 0.1.39 port, the Alchemist
-A-series, integrated Arc GPUs (the B390 / Panther Lake in #515; the shared-memory planning does not exist yet), and
-images (not wired on Intel).
+**Not tested by anyone yet:** Arc on WSL2 for the 0.1.39 port, integrated Arc GPUs (the B390 / Panther Lake in #515;
+the shared-memory planning does not exist yet), and images (not wired on Intel). Windows runs natively on an Arc Pro
+B70 since this version ([INTEL_WINDOWS.md](INTEL_WINDOWS.md)); other cards on Windows are untested.
 
 ## What you need
 
@@ -157,11 +158,15 @@ Things that matter on an Arc (details in INTEL.md):
 
 ## Windows
 
-There is no Windows path yet. `setup --backend sycl` on Windows stops and points here. oneAPI exists for Windows,
-but `sycl/CMakeLists.txt` uses GCC-style flags (`-mavx512f`, `-fp-model=precise`, `-qmkl`) and the runner is a
-bash/Docker script, so a native Windows build would need work. Nobody has tried it. WSL2 with an Arc has been
-used by one tester (the B580 row above), but setup cannot detect the card there, because it reads `/sys/class/drm`,
-which WSL2 does not have.
+The port runs natively on Windows 11, without WSL2 or Docker: `sycl\tools\build_windows.bat` builds the engine with
+oneAPI's `icx` and Visual Studio's linker into `engine\`, and `START-HERE.bat --backend sycl` installs and starts a
+model the usual way. It was built and run on an Arc Pro B70 (the card dedicated to the model, an NVIDIA card driving
+the desktop): the Coder IQ1_M at 60 tok/s on the first chat (2026-10-05), Swift 1.5 IQ3_XXS with part of its experts
+in a pinned RAM mirror at 53-64 tok/s decode over several days of use (2026-10-07 to 09). What the Windows build adds
+(a chunked VRAM arena for WDDM, a VRAM heartbeat for a display-less card, the DXGI memory budget, the card choice by
+Level Zero index, the engine's environment, Sysman telemetry) and what is still open is in
+[INTEL_WINDOWS.md](INTEL_WINDOWS.md). WSL2 has been used by one tester (the B580 row above), but setup cannot detect
+the card there, because it reads `/sys/class/drm`, which WSL2 does not have; the native path is the one to use.
 
 ## Reporting a problem
 

setup.py

@@ -4549,12 +4549,14 @@ def sycl_setup(argv) -> int:
     with the Intel steps swapped in. Nothing of the CUDA / HIP paths is used or changed."""
     say()
     say("  Intel Arc (--backend sycl): supported since 0.1.40.2 on Linux, tested on an Arc Pro B70 (xe) and an Arc A750 "
-        "(i915); other Arc cards and driver versions are untested, and reports help (docs/INTEL.md).")
+        "(i915), and on Windows, tested on an Arc Pro B70; other Arc cards and driver versions are untested, and "
+        "reports help (docs/INTEL.md).")
     if WIN:
-        fail("the Intel Arc engine has no Windows setup yet (no ready-made Intel engine either)",
-             "run it on Linux (Ubuntu 24.04 with Intel's GPU driver and oneAPI): docs/INTEL_ARC.md")
-    say("  There is no ready-made Intel engine: it is built from source with Intel oneAPI (icpx + oneMKL),")
-    say("  docs/INTEL_ARC.md. Setup continues with sycl/setup_intel.py.")
+        say("  There is no ready-made Intel engine: it is built from source with Intel oneAPI (icx + oneMKL) by")
+        say("  sycl\\tools\\build_windows.bat, docs/INTEL_WINDOWS.md. Setup continues with sycl\\setup_intel.py.")
+    else:
+        say("  There is no ready-made Intel engine: it is built from source with Intel oneAPI (icpx + oneMKL),")
+        say("  docs/INTEL_ARC.md. Setup continues with sycl/setup_intel.py.")
     rest, skip = [], False
     for x in argv:                                     # setup_intel.py drives this setup through its AMD path
         if skip:
@@ -4667,7 +4669,7 @@ def main() -> int:
     ap.add_argument("--backend", choices=["cuda", "hip", "sycl"],
                     help="cuda = NVIDIA (default), hip = AMD RX 7900 / 7800 / 7700 XT, RX 9060 XT / 9070 / AI PRO R9700 on "
                          "Linux or Windows (chosen by itself when the PC has no NVIDIA card Strata can use), "
-                         "sycl = Intel Arc, EXPERIMENTAL: Linux, built from source (docs/INTEL_ARC.md)")
+                         "sycl = Intel Arc, EXPERIMENTAL: Linux or Windows, built from source (docs/INTEL_ARC.md)")
     ap.add_argument("--skip-build", action="store_true", help=argparse.SUPPRESS)
     a = ap.parse_args()
     if a.source: