feat: add native Intel XPU (torch.xpu) device support - #9401

Merged

lstein merged 41 commits into

invoke-ai:mainfrom

LexiconCode:feat/intel-xpu-support

Aug 15, 2026

Commits on Aug 13, 2026

feat(backend): add Intel XPU (torch.xpu) device support

LexiconCode
authored and
LexiconCode
committed
f3ef296
test(backend): add XPU coverage for TorchDevice

LexiconCode
authored and
LexiconCode
committed
97e1d83
build: add [xpu] extra

LexiconCode
authored and
LexiconCode
committed
cbf1cec
feat(backend): extend idle-GPU text encoder offload to XPU

LexiconCode
authored and
LexiconCode
committed
db34e71
feat(ui): show the executing GPU for XPU sessions

LexiconCode
authored and
LexiconCode
committed
1b1ad54
fix(mm): gate Krea 2 fp8 encoder casting on fp8 storage support

LexiconCode
authored and
LexiconCode
committed
7ba7b2f
chore: label XPU devices by index in load logs and fp8 help text

LexiconCode
authored and
LexiconCode
committed
55e37fd
test: cover XPU config validation, progress device and fp8 probe

LexiconCode
authored and
LexiconCode
committed
9bdd0c0
fix(nodes): recognise XPU out-of-memory errors in the Anima VAE retry

LexiconCode
authored and
LexiconCode
committed
3488824
style: wrap long vram_usage_gb ternary for ruff
LexiconCode
committed
930e702
fix: drop CUDA-only wording from progress device description
LexiconCode
committed
6ecaa8d
docs: regenerate settings data for xpu device values
LexiconCode
committed
b6bf7a3
fix: stop xpu VRAM probe from reporting an unknown total as zero
LexiconCode
committed
5d3baf0
fix: probe fp8 support on the target device, per device, without caching failures
LexiconCode
committed
4e2fa56
fix: pin torch current device when borrowing an idle GPU
LexiconCode
committed
13025e8
fix: keep idle-GPU borrows within one device type
LexiconCode
committed
7e8245f
feat: detect Intel integrated GPUs via Level Zero
LexiconCode
committed
6950ca8
feat: add xpu torch index to pins.json
LexiconCode
committed
21b7324
fix: report VRAM diagnostics for the device in use
LexiconCode
committed
f603de5
docs: record why xpu takes the CUDA VAE constants and keeps the broad OOM needle
LexiconCode
committed
25e2b4e
fix: derive rand_device metadata from the backend's devices
LexiconCode
committed
8c05cf5
docs: add Intel Arc install, driver and VRAM-reporting notes
LexiconCode
committed
99e8c71
fix: probe fp8 device support only when a model requests it
LexiconCode
committed
506ef5b
fix: query Level Zero Sysman for driver-global free VRAM on xpu
LexiconCode
committed
27c8b56
fix: make the fp8 probe mirror the runtime cast path
LexiconCode
committed
7ae8cd0
fix: degrade gracefully when a backend cannot name a device
LexiconCode
committed
789180b
fix: resolve an index-less device in the Sysman VRAM query
LexiconCode
committed
3d3c86b
fix: declare ctypes prototypes for the Level Zero calls
LexiconCode
committed
4eb540b
fix: drop the ZES_ENABLE_SYSMAN mutation from the Sysman probe
LexiconCode
committed
1fffa2a
refactor: tidy up the xpu additions after a cleanup review
LexiconCode
committed
a779812
build: pin the xpu extra to torch 2.13.0
LexiconCode
committed
82372a1
test: stub Sysman in the unknown-total xpu probe test
LexiconCode
committed
3700e37
build: teach the pins check about the xpu index
LexiconCode
committed
7cc6de3
fix: defer the xpu device pin like cuda's
LexiconCode
committed
0123ffa
docs: regenerate settings data on linux
LexiconCode
committed
0f5272b
fix(mm): handle shared memory on integrated GPUs
LexiconCode
committed
77f2d07
docs: note intel device selection and integrated-GPU memory
LexiconCode
committed
f779454
chore(ui): typegen for the xpu device values
LexiconCode
committed
c5b83b6

Commits on Aug 15, 2026

Merge branch 'main' into feat/intel-xpu-support

lstein
and
claude
committed
961c443
fix(mm): compare the integrated-GPU full-load guard against bytes still to move

lstein
and
claude
committed
0645f90
test: run the integrated-GPU cache tests on CPU-only torch

lstein
and
claude
committed
d04d443