diff --git a/cuda_core/cuda/core/_memory/_memory_pool.pyx b/cuda_core/cuda/core/_memory/_memory_pool.pyx
index 1e9f5116c1..c3dc4114db 100644
--- a/cuda_core/cuda/core/_memory/_memory_pool.pyx
+++ b/cuda_core/cuda/core/_memory/_memory_pool.pyx
@@ -348,8 +348,8 @@ cdef int _MP_init_current(_MemPool self, int dev_id, _MemPoolOptions opts) excep
     elif opts._type == cydriver.CUmemAllocationType.CU_MEM_ALLOCATION_TYPE_PINNED \
             and opts._location == cydriver.CUmemLocationType.CU_MEM_LOCATION_TYPE_HOST_NUMA:
         IF CUDA_CORE_BUILD_MAJOR >= 13:
-            assert dev_id == 0
-            loc.id = 0
+            assert dev_id >= 0
+            loc.id = dev_id
             loc.type = opts._location
             with nogil:
                 HANDLE_RETURN(cydriver.cuMemGetMemPool(&pool, &loc, opts._type))
diff --git a/cuda_core/cuda/core/_memory/_pinned_memory_resource.pyx b/cuda_core/cuda/core/_memory/_pinned_memory_resource.pyx
index b2a9db4594..3d1e916c95 100644
--- a/cuda_core/cuda/core/_memory/_pinned_memory_resource.pyx
+++ b/cuda_core/cuda/core/_memory/_pinned_memory_resource.pyx
@@ -76,8 +76,8 @@ def _check_numa_nodes():
         if numa_count is not None and numa_count > 1:
             warnings.warn(
                 f"System has {numa_count} NUMA nodes. IPC-enabled pinned memory "
-                f"uses location ID 0, which may not work correctly with multiple "
-                f"NUMA nodes.",
+                f"targets the host NUMA node closest to the current device; "
+                f"this may not work correctly with multiple NUMA nodes.",
                 UserWarning,
                 stacklevel=3
             )
@@ -85,6 +85,38 @@ def _check_numa_nodes():
         _numa_warning_shown = True
 
 
+def _host_numa_id_for_current_device() -> int:
+    """Return host NUMA node closest to current device (fallback to 0)."""
+    cdef cydriver.CUdevice dev
+    cdef cydriver.CUresult err
+    cdef int host_numa_id
+
+    with nogil:
+        err = cydriver.cuCtxGetDevice(&dev)
+    if err in (
+        cydriver.CUresult.CUDA_ERROR_INVALID_CONTEXT,
+        cydriver.CUresult.CUDA_ERROR_NOT_INITIALIZED,
+    ):
+        return 0
+    HANDLE_RETURN(err)
+
+    with nogil:
+        err = cydriver.cuDeviceGetAttribute(
+            &host_numa_id,
+            cydriver.CUdevice_attribute.CU_DEVICE_ATTRIBUTE_HOST_NUMA_ID,
+            dev
+        )
+    if err in (
+        cydriver.CUresult.CUDA_ERROR_INVALID_VALUE,
+        cydriver.CUresult.CUDA_ERROR_NOT_SUPPORTED,
+    ):
+        return 0
+    HANDLE_RETURN(err)
+    if host_numa_id < 0:
+        return 0
+    return host_numa_id
+
+
 __all__ = ['PinnedMemoryResource', 'PinnedMemoryResourceOptions']
 
 
@@ -134,10 +166,10 @@ cdef class PinnedMemoryResource(_MemPool):
     allocations between processes, specify ``ipc_enabled=True`` in the initializer
     option. When IPC is enabled, the location type is automatically set to
     CU_MEM_LOCATION_TYPE_HOST_NUMA instead of CU_MEM_LOCATION_TYPE_HOST,
-    with location ID 0.
+    using the host NUMA node closest to the current device.
 
-    Note: IPC support for pinned memory requires a single NUMA node. A warning
-    is issued if multiple NUMA nodes are detected.
+    Note: IPC support for pinned memory can be sensitive to NUMA placement. A
+    warning is issued if multiple NUMA nodes are detected.
 
     See :class:`DeviceMemoryResource` for more details on IPC usage patterns.
     """
@@ -150,6 +182,7 @@ cdef class PinnedMemoryResource(_MemPool):
         cdef _MemPoolOptions opts_base = _MemPoolOptions()
 
         cdef bint ipc_enabled = False
+        cdef int location_id = -1
         if opts:
             ipc_enabled = opts.ipc_enabled
             if ipc_enabled and not _ipc.is_supported():
@@ -157,6 +190,7 @@ cdef class PinnedMemoryResource(_MemPool):
             if ipc_enabled:
                 # Check for multiple NUMA nodes on Linux
                 _check_numa_nodes()
+                location_id = _host_numa_id_for_current_device()
             opts_base._max_size = opts.max_size
             opts_base._use_current = False
         opts_base._ipc_enabled = ipc_enabled
@@ -166,7 +200,7 @@ cdef class PinnedMemoryResource(_MemPool):
             opts_base._location = cydriver.CUmemLocationType.CU_MEM_LOCATION_TYPE_HOST
         opts_base._type = cydriver.CUmemAllocationType.CU_MEM_ALLOCATION_TYPE_PINNED
 
-        super().__init__(0 if ipc_enabled else -1, opts_base)
+        super().__init__(location_id if ipc_enabled else -1, opts_base)
 
     def __reduce__(self):
         return PinnedMemoryResource.from_registry, (self.uuid,)
diff --git a/cuda_core/tests/test_memory.py b/cuda_core/tests/test_memory.py
index 9a88f5f483..60bc90bbdd 100644
--- a/cuda_core/tests/test_memory.py
+++ b/cuda_core/tests/test_memory.py
@@ -978,7 +978,7 @@ def test_mempool_ipc_errors(mempool_device):
 
 def test_pinned_mempool_ipc_basic():
     """Test basic IPC functionality for PinnedMemoryResource."""
-    device = Device()
+    device = Device(0)
     device.set_current()
 
     skip_if_pinned_memory_unsupported(device)
@@ -995,7 +995,11 @@ def test_pinned_mempool_ipc_basic():
     assert mr.is_ipc_enabled
     assert mr.is_device_accessible
     assert mr.is_host_accessible
-    assert mr.device_id == 0  # IPC-enabled uses location id 0
+    # For IPC-enabled pinned pools, device_id reflects the host NUMA location ID.
+    expected_numa_id = device.properties.host_numa_id
+    if expected_numa_id < 0:
+        expected_numa_id = 0
+    assert mr.device_id == expected_numa_id
 
     # Test allocation handle export
     alloc_handle = mr.get_allocation_handle()