diff --git a/ompi/mca/coll/accelerator/coll_accelerator.h b/ompi/mca/coll/accelerator/coll_accelerator.h
index e707d7ec7f2..70d971cc9a8 100644
--- a/ompi/mca/coll/accelerator/coll_accelerator.h
+++ b/ompi/mca/coll/accelerator/coll_accelerator.h
@@ -5,6 +5,7 @@
  *                         reserved.
  * Copyright (c) 2014-2024 NVIDIA Corporation.  All rights reserved.
  * Copyright (c) 2024      Triad National Security, LLC. All rights reserved.
+ * Copyright (c) 2024      Advanced Micro Devices, Inc. All Rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -87,22 +88,24 @@ mca_coll_accelerator_reduce_scatter_block(const void *sbuf, void *rbuf, size_t r
  * @retval >0                The buffer belongs to a managed buffer in
  *                           device memory.
  */
-static inline int mca_coll_accelerator_check_buf(void *addr)
+static inline int mca_coll_accelerator_check_buf(void *addr, int *dev_id)
 {
     uint64_t flags;
-    int dev_id;
+
     if (OPAL_LIKELY(NULL != addr)) {
-        return opal_accelerator.check_addr(addr, &dev_id, &flags);
+        return opal_accelerator.check_addr(addr, dev_id, &flags);
     } else {
+        *dev_id = MCA_ACCELERATOR_NO_DEVICE_ID;
         return 0;
     }
 }
 
-static inline void *mca_coll_accelerator_memcpy(void *dest, const void *src, size_t size)
+static inline void *mca_coll_accelerator_memcpy(void *dest, int dest_dev, const void *src, int src_dev, size_t size,
+						opal_accelerator_transfer_type_t type)
 {
     int res;
-    res = opal_accelerator.mem_copy(MCA_ACCELERATOR_NO_DEVICE_ID, MCA_ACCELERATOR_NO_DEVICE_ID,
-                                    dest, src, size, MCA_ACCELERATOR_TRANSFER_UNSPEC);
+
+    res = opal_accelerator.mem_copy(dest_dev, src_dev, dest, src, size, type);
     if (res != 0) {
         opal_output(0, "coll/accelerator: Error in mem_copy: res=%d, dest=%p, src=%p, size=%d", res, dest, src,
                     (int) size);
diff --git a/ompi/mca/coll/accelerator/coll_accelerator_allreduce.c b/ompi/mca/coll/accelerator/coll_accelerator_allreduce.c
index ad0566cad11..75e9051ce0a 100644
--- a/ompi/mca/coll/accelerator/coll_accelerator_allreduce.c
+++ b/ompi/mca/coll/accelerator/coll_accelerator_allreduce.c
@@ -5,6 +5,7 @@
  * Copyright (c) 2014-2015 NVIDIA Corporation.  All rights reserved.
  * Copyright (c) 2022      Amazon.com, Inc. or its affiliates.  All Rights reserved.
  * Copyright (c) 2024      Triad National Security, LLC. All rights reserved.
+ * Copyright (c) 2024      Advanced Micro Devices, Inc. All Rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -37,11 +38,12 @@ mca_coll_accelerator_allreduce(const void *sbuf, void *rbuf, size_t count,
     mca_coll_accelerator_module_t *s = (mca_coll_accelerator_module_t*) module;
     ptrdiff_t gap;
     char *rbuf1 = NULL, *sbuf1 = NULL, *rbuf2 = NULL;
+    int sbuf_dev, rbuf_dev;
     size_t bufsize;
     int rc;
 
     bufsize = opal_datatype_span(&dtype->super, count, &gap);
-    rc = mca_coll_accelerator_check_buf((void *)sbuf);
+    rc = mca_coll_accelerator_check_buf((void *)sbuf, &sbuf_dev);
     if (rc < 0) {
         return rc;
     }
@@ -50,10 +52,11 @@ mca_coll_accelerator_allreduce(const void *sbuf, void *rbuf, size_t count,
         if (NULL == sbuf1) {
             return OMPI_ERR_OUT_OF_RESOURCE;
         }
-        mca_coll_accelerator_memcpy(sbuf1, sbuf, bufsize);
+        mca_coll_accelerator_memcpy(sbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, sbuf, sbuf_dev,
+                                    bufsize, MCA_ACCELERATOR_TRANSFER_DTOH);
         sbuf = sbuf1 - gap;
     }
-    rc = mca_coll_accelerator_check_buf(rbuf);
+    rc = mca_coll_accelerator_check_buf(rbuf, &rbuf_dev);
     if (rc < 0) {
         return rc;
     }
@@ -63,7 +66,8 @@ mca_coll_accelerator_allreduce(const void *sbuf, void *rbuf, size_t count,
             if (NULL != sbuf1) free(sbuf1);
             return OMPI_ERR_OUT_OF_RESOURCE;
         }
-        mca_coll_accelerator_memcpy(rbuf1, rbuf, bufsize);
+        mca_coll_accelerator_memcpy(rbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, rbuf, rbuf_dev,
+                                    bufsize, MCA_ACCELERATOR_TRANSFER_DTOH);
         rbuf2 = rbuf; /* save away original buffer */
         rbuf = rbuf1 - gap;
     }
@@ -73,7 +77,8 @@ mca_coll_accelerator_allreduce(const void *sbuf, void *rbuf, size_t count,
     }
     if (NULL != rbuf1) {
         rbuf = rbuf2;
-        mca_coll_accelerator_memcpy(rbuf, rbuf1, bufsize);
+        mca_coll_accelerator_memcpy(rbuf, rbuf_dev, rbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, bufsize,
+                                    MCA_ACCELERATOR_TRANSFER_HTOD);
         free(rbuf1);
     }
     return rc;
diff --git a/ompi/mca/coll/accelerator/coll_accelerator_exscan.c b/ompi/mca/coll/accelerator/coll_accelerator_exscan.c
index 4933cbcedd3..d77dd5a9b46 100644
--- a/ompi/mca/coll/accelerator/coll_accelerator_exscan.c
+++ b/ompi/mca/coll/accelerator/coll_accelerator_exscan.c
@@ -5,6 +5,7 @@
  * Copyright (c) 2014-2015 NVIDIA Corporation.  All rights reserved.
  * Copyright (c) 2022      Amazon.com, Inc. or its affiliates.  All Rights reserved.
  * Copyright (c) 2024      Triad National Security, LLC. All rights reserved.
+ * Copyright (c) 2024      Advanced Micro Devices, Inc. All Rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -29,11 +30,12 @@ int mca_coll_accelerator_exscan(const void *sbuf, void *rbuf, size_t count,
     mca_coll_accelerator_module_t *s = (mca_coll_accelerator_module_t*) module;
     ptrdiff_t gap;
     char *rbuf1 = NULL, *sbuf1 = NULL, *rbuf2 = NULL;
+    int sbuf_dev, rbuf_dev;
     size_t bufsize;
     int rc;
 
     bufsize = opal_datatype_span(&dtype->super, count, &gap);
-    rc = mca_coll_accelerator_check_buf((void *)sbuf);
+    rc = mca_coll_accelerator_check_buf((void *)sbuf, &sbuf_dev);
     if (rc < 0) {
         return rc;
     }
@@ -43,10 +45,11 @@ int mca_coll_accelerator_exscan(const void *sbuf, void *rbuf, size_t count,
         if (NULL == sbuf1) {
             return OMPI_ERR_OUT_OF_RESOURCE;
         }
-        mca_coll_accelerator_memcpy(sbuf1, sbuf, bufsize);
+        mca_coll_accelerator_memcpy(sbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, sbuf, sbuf_dev, bufsize,
+                                    MCA_ACCELERATOR_TRANSFER_DTOH);
         sbuf = sbuf1 - gap;
     }
-    rc = mca_coll_accelerator_check_buf(rbuf);
+    rc = mca_coll_accelerator_check_buf(rbuf, &rbuf_dev);
     if (rc < 0) {
         return rc;
     }
@@ -56,7 +59,8 @@ int mca_coll_accelerator_exscan(const void *sbuf, void *rbuf, size_t count,
             if (NULL != sbuf1) free(sbuf1);
             return OMPI_ERR_OUT_OF_RESOURCE;
         }
-        mca_coll_accelerator_memcpy(rbuf1, rbuf, bufsize);
+        mca_coll_accelerator_memcpy(rbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, rbuf, rbuf_dev, bufsize,
+                                    MCA_ACCELERATOR_TRANSFER_DTOH);
         rbuf2 = rbuf; /* save away original buffer */
         rbuf = rbuf1 - gap;
     }
@@ -68,7 +72,8 @@ int mca_coll_accelerator_exscan(const void *sbuf, void *rbuf, size_t count,
     }
     if (NULL != rbuf1) {
         rbuf = rbuf2;
-        mca_coll_accelerator_memcpy(rbuf, rbuf1, bufsize);
+        mca_coll_accelerator_memcpy(rbuf, rbuf_dev, rbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, bufsize,
+                                    MCA_ACCELERATOR_TRANSFER_HTOD);
         free(rbuf1);
     }
     return rc;
diff --git a/ompi/mca/coll/accelerator/coll_accelerator_reduce.c b/ompi/mca/coll/accelerator/coll_accelerator_reduce.c
index 993271fa16b..509ea0961d4 100644
--- a/ompi/mca/coll/accelerator/coll_accelerator_reduce.c
+++ b/ompi/mca/coll/accelerator/coll_accelerator_reduce.c
@@ -6,6 +6,7 @@
  * Copyright (c) 2014-2015 NVIDIA Corporation.  All rights reserved.
  * Copyright (c) 2022      Amazon.com, Inc. or its affiliates.  All Rights reserved.
  * Copyright (c) 2024      Triad National Security, LLC. All rights reserved.
+ * Copyright (c) 2024      Advanced Micro Devices, Inc. All Rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -39,12 +40,13 @@ mca_coll_accelerator_reduce(const void *sbuf, void *rbuf, size_t count,
     int rank = ompi_comm_rank(comm);
     ptrdiff_t gap;
     char *rbuf1 = NULL, *sbuf1 = NULL, *rbuf2 = NULL;
+    int rbuf_dev, sbuf_dev;
     size_t bufsize;
     int rc;
 
     bufsize = opal_datatype_span(&dtype->super, count, &gap);
 
-    rc = mca_coll_accelerator_check_buf((void *)sbuf);
+    rc = mca_coll_accelerator_check_buf((void *)sbuf, &sbuf_dev);
     if (rc < 0) {
         return rc;
     }
@@ -53,11 +55,12 @@ mca_coll_accelerator_reduce(const void *sbuf, void *rbuf, size_t count,
         if (NULL == sbuf1) {
             return OMPI_ERR_OUT_OF_RESOURCE;
         }
-        mca_coll_accelerator_memcpy(sbuf1, sbuf, bufsize);
+        mca_coll_accelerator_memcpy(sbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, sbuf, sbuf_dev, bufsize,
+                                    MCA_ACCELERATOR_TRANSFER_DTOH);
         sbuf = sbuf1 - gap;
     }
 
-    rc = mca_coll_accelerator_check_buf(rbuf);
+    rc = mca_coll_accelerator_check_buf(rbuf, &rbuf_dev);
     if (rc < 0) {
         return rc;
     }
@@ -67,7 +70,8 @@ mca_coll_accelerator_reduce(const void *sbuf, void *rbuf, size_t count,
             if (NULL != sbuf1) free(sbuf1);
             return OMPI_ERR_OUT_OF_RESOURCE;
         }
-        mca_coll_accelerator_memcpy(rbuf1, rbuf, bufsize);
+        mca_coll_accelerator_memcpy(rbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, rbuf, rbuf_dev, bufsize,
+                                    MCA_ACCELERATOR_TRANSFER_DTOH);
         rbuf2 = rbuf; /* save away original buffer */
         rbuf = rbuf1 - gap;
     }
@@ -80,7 +84,8 @@ mca_coll_accelerator_reduce(const void *sbuf, void *rbuf, size_t count,
     }
     if (NULL != rbuf1) {
         rbuf = rbuf2;
-        mca_coll_accelerator_memcpy(rbuf, rbuf1, bufsize);
+        mca_coll_accelerator_memcpy(rbuf, rbuf_dev, rbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, bufsize,
+                                    MCA_ACCELERATOR_TRANSFER_HTOD);
         free(rbuf1);
     }
     return rc;
@@ -94,12 +99,13 @@ mca_coll_accelerator_reduce_local(const void *sbuf, void *rbuf, size_t count,
 {
     ptrdiff_t gap;
     char *rbuf1 = NULL, *sbuf1 = NULL, *rbuf2 = NULL;
+    int sbuf_dev, rbuf_dev;
     size_t bufsize;
     int rc;
 
     bufsize = opal_datatype_span(&dtype->super, count, &gap);
 
-    rc = mca_coll_accelerator_check_buf((void *)sbuf);
+    rc = mca_coll_accelerator_check_buf((void *)sbuf, &sbuf_dev);
     if (rc < 0) {
         return rc;
     }
@@ -109,11 +115,12 @@ mca_coll_accelerator_reduce_local(const void *sbuf, void *rbuf, size_t count,
         if (NULL == sbuf1) {
             return OMPI_ERR_OUT_OF_RESOURCE;
         }
-        mca_coll_accelerator_memcpy(sbuf1, sbuf, bufsize);
+        mca_coll_accelerator_memcpy(sbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, sbuf, sbuf_dev, bufsize,
+                                    MCA_ACCELERATOR_TRANSFER_DTOH);
         sbuf = sbuf1 - gap;
     }
 
-    rc = mca_coll_accelerator_check_buf(rbuf);
+    rc = mca_coll_accelerator_check_buf(rbuf, &rbuf_dev);
     if (rc < 0) {
         return rc;
     }
@@ -124,7 +131,8 @@ mca_coll_accelerator_reduce_local(const void *sbuf, void *rbuf, size_t count,
             if (NULL != sbuf1) free(sbuf1);
             return OMPI_ERR_OUT_OF_RESOURCE;
         }
-        mca_coll_accelerator_memcpy(rbuf1, rbuf, bufsize);
+        mca_coll_accelerator_memcpy(rbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, rbuf, rbuf_dev, bufsize,
+                                    MCA_ACCELERATOR_TRANSFER_DTOH);
         rbuf2 = rbuf; /* save away original buffer */
         rbuf = rbuf1 - gap;
     }
@@ -137,7 +145,8 @@ mca_coll_accelerator_reduce_local(const void *sbuf, void *rbuf, size_t count,
     }
     if (NULL != rbuf1) {
         rbuf = rbuf2;
-        mca_coll_accelerator_memcpy(rbuf, rbuf1, bufsize);
+        mca_coll_accelerator_memcpy(rbuf, rbuf_dev, rbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, bufsize,
+                                    MCA_ACCELERATOR_TRANSFER_HTOD);
         free(rbuf1);
     }
     return rc;
diff --git a/ompi/mca/coll/accelerator/coll_accelerator_reduce_scatter_block.c b/ompi/mca/coll/accelerator/coll_accelerator_reduce_scatter_block.c
index 9dc27b61601..5f0fd61914f 100644
--- a/ompi/mca/coll/accelerator/coll_accelerator_reduce_scatter_block.c
+++ b/ompi/mca/coll/accelerator/coll_accelerator_reduce_scatter_block.c
@@ -5,6 +5,7 @@
  * Copyright (c) 2014-2015 NVIDIA Corporation.  All rights reserved.
  * Copyright (c) 2022      Amazon.com, Inc. or its affiliates.  All Rights reserved.
  * Copyright (c) 2024      Triad National Security, LLC. All rights reserved.
+ * Copyright (c) 2024      Advanced Micro Devices, Inc. All Rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -41,13 +42,14 @@ mca_coll_accelerator_reduce_scatter_block(const void *sbuf, void *rbuf, size_t r
     mca_coll_accelerator_module_t *s = (mca_coll_accelerator_module_t*) module;
     ptrdiff_t gap;
     char *rbuf1 = NULL, *sbuf1 = NULL, *rbuf2 = NULL;
+    int sbuf_dev, rbuf_dev;
     size_t sbufsize, rbufsize;
     int rc;
 
     rbufsize = opal_datatype_span(&dtype->super, rcount, &gap);
 
     sbufsize = rbufsize * ompi_comm_size(comm);
-    rc = mca_coll_accelerator_check_buf((void *)sbuf);
+    rc = mca_coll_accelerator_check_buf((void *)sbuf, &sbuf_dev);
     if (rc < 0) {
         return rc;
     }
@@ -56,10 +58,11 @@ mca_coll_accelerator_reduce_scatter_block(const void *sbuf, void *rbuf, size_t r
         if (NULL == sbuf1) {
             return OMPI_ERR_OUT_OF_RESOURCE;
         }
-        mca_coll_accelerator_memcpy(sbuf1, sbuf, sbufsize);
+        mca_coll_accelerator_memcpy(sbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, sbuf, sbuf_dev, sbufsize,
+                                    MCA_ACCELERATOR_TRANSFER_DTOH);
         sbuf = sbuf1 - gap;
     }
-    rc = mca_coll_accelerator_check_buf(rbuf);
+    rc = mca_coll_accelerator_check_buf(rbuf, &rbuf_dev);
     if (rc < 0) {
         return rc;
     }
@@ -69,7 +72,8 @@ mca_coll_accelerator_reduce_scatter_block(const void *sbuf, void *rbuf, size_t r
             if (NULL != sbuf1) free(sbuf1);
             return OMPI_ERR_OUT_OF_RESOURCE;
         }
-        mca_coll_accelerator_memcpy(rbuf1, rbuf, rbufsize);
+        mca_coll_accelerator_memcpy(rbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, rbuf, rbuf_dev, rbufsize,
+                                    MCA_ACCELERATOR_TRANSFER_DTOH);
         rbuf2 = rbuf; /* save away original buffer */
         rbuf = rbuf1 - gap;
     }
@@ -80,7 +84,8 @@ mca_coll_accelerator_reduce_scatter_block(const void *sbuf, void *rbuf, size_t r
     }
     if (NULL != rbuf1) {
         rbuf = rbuf2;
-        mca_coll_accelerator_memcpy(rbuf, rbuf1, rbufsize);
+        mca_coll_accelerator_memcpy(rbuf, rbuf_dev, rbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, rbufsize,
+                                    MCA_ACCELERATOR_TRANSFER_HTOD);
         free(rbuf1);
     }
     return rc;
diff --git a/ompi/mca/coll/accelerator/coll_accelerator_scan.c b/ompi/mca/coll/accelerator/coll_accelerator_scan.c
index ef7ac86971a..7018744c1c8 100644
--- a/ompi/mca/coll/accelerator/coll_accelerator_scan.c
+++ b/ompi/mca/coll/accelerator/coll_accelerator_scan.c
@@ -5,6 +5,7 @@
  * Copyright (c) 2014-2015 NVIDIA Corporation.  All rights reserved.
  * Copyright (c) 2022      Amazon.com, Inc. or its affiliates.  All Rights reserved.
  * Copyright (c) 2024      Triad National Security, LLC. All rights reserved.
+ * Copyright (c) 2024      Advanced Micro Devices, Inc. All Rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -36,11 +37,12 @@ int mca_coll_accelerator_scan(const void *sbuf, void *rbuf, size_t count,
     mca_coll_accelerator_module_t *s = (mca_coll_accelerator_module_t*) module;
     ptrdiff_t gap;
     char *rbuf1 = NULL, *sbuf1 = NULL, *rbuf2 = NULL;
+    int sbuf_dev, rbuf_dev;
     size_t bufsize;
     int rc;
 
     bufsize = opal_datatype_span(&dtype->super, count, &gap);
-    rc = mca_coll_accelerator_check_buf((void *)sbuf);
+    rc = mca_coll_accelerator_check_buf((void *)sbuf, &sbuf_dev);
     if (rc < 0) {
         return rc;
     }
@@ -49,10 +51,11 @@ int mca_coll_accelerator_scan(const void *sbuf, void *rbuf, size_t count,
         if (NULL == sbuf1) {
             return OMPI_ERR_OUT_OF_RESOURCE;
         }
-        mca_coll_accelerator_memcpy(sbuf1, sbuf, bufsize);
+        mca_coll_accelerator_memcpy(sbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, sbuf, sbuf_dev, bufsize,
+                                    MCA_ACCELERATOR_TRANSFER_DTOH);
         sbuf = sbuf1 - gap;
     }
-    rc = mca_coll_accelerator_check_buf(rbuf);
+    rc = mca_coll_accelerator_check_buf(rbuf, &rbuf_dev);
     if (rc < 0) {
         return rc;
     }
@@ -62,7 +65,8 @@ int mca_coll_accelerator_scan(const void *sbuf, void *rbuf, size_t count,
             if (NULL != sbuf1) free(sbuf1);
             return OMPI_ERR_OUT_OF_RESOURCE;
         }
-        mca_coll_accelerator_memcpy(rbuf1, rbuf, bufsize);
+        mca_coll_accelerator_memcpy(rbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, rbuf, rbuf_dev, bufsize,
+                                    MCA_ACCELERATOR_TRANSFER_DTOH);
         rbuf2 = rbuf; /* save away original buffer */
         rbuf = rbuf1 - gap;
     }
@@ -73,7 +77,8 @@ int mca_coll_accelerator_scan(const void *sbuf, void *rbuf, size_t count,
     }
     if (NULL != rbuf1) {
         rbuf = rbuf2;
-        mca_coll_accelerator_memcpy(rbuf, rbuf1, bufsize);
+        mca_coll_accelerator_memcpy(rbuf, rbuf_dev, rbuf1, MCA_ACCELERATOR_NO_DEVICE_ID, bufsize,
+                                    MCA_ACCELERATOR_TRANSFER_HTOD);
         free(rbuf1);
     }
     return rc;