You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Multi-GPU inference using the SYCL / oneAPI backend fails with an out-of-device-memory error during a SYCL memcpy().wait() call, even though sufficient VRAM is available on each GPU. The same model and configuration work reliably on a single Intel GPU.
This appears to be a multi-GPU SYCL / Level Zero pipeline or cross-device copy issue, not a real VRAM exhaustion problem.
Model uses <3 GiB weights + ~512 MiB KV cache
The failure happens after successful model load, during inference
The error is triggered inside a SYCL memcpy + wait, suggesting:
cross-device tensor movement
pipeline parallelism
or Level Zero memory management issues
With two GPUs visible, logs show:
Summary
Multi-GPU inference using the SYCL / oneAPI backend fails with an out-of-device-memory error during a SYCL memcpy().wait() call, even though sufficient VRAM is available on each GPU. The same model and configuration work reliably on a single Intel GPU.
This appears to be a multi-GPU SYCL / Level Zero pipeline or cross-device copy issue, not a real VRAM exhaustion problem.
Environment
GPUs
Reproduction Steps
Works (single GPU)
Model loads and runs correctly
Stable inference, high tokens/sec
Fails (multi-GPU)
Model loads successfully
Fails on first inference request
Observed Error
Key Observations
This is not a real VRAM exhaustion issue
The failure happens after successful model load, during inference
The error is triggered inside a SYCL memcpy + wait, suggesting:
With two GPUs visible, logs show:
With one GPU:
Any help would be appreciated.