diff --git a/include/ofi_atomic_queue.h b/include/ofi_atomic_queue.h
index 8ffecbe2667..9d5eb70de6e 100644
--- a/include/ofi_atomic_queue.h
+++ b/include/ofi_atomic_queue.h
@@ -94,19 +94,28 @@ extern "C" {
 
 #define OFI_CACHE_LINE_SIZE (64)
 
+/*
+ * Base address of atomic queue must be cache line aligned to maximize atomic
+ * value perforamnce benefits
+ */
 #define OFI_DECLARE_ATOMIC_Q(entrytype, name)			\
 struct name ## _entry {						\
 	ofi_atomic64_t	seq;					\
 	bool		noop;					\
 	entrytype	buf;					\
-};								\
+} __attribute__((__aligned__(64)));				\
+								\
 struct name {							\
-	int		size;					\
-	int		size_mask;				\
 	ofi_atomic64_t	write_pos;				\
-	char		pad0[OFI_CACHE_LINE_SIZE];		\
+	uint8_t		pad0[OFI_CACHE_LINE_SIZE -		\
+			     sizeof(ofi_atomic64_t)];		\
 	ofi_atomic64_t	read_pos;				\
-	char		pad1[OFI_CACHE_LINE_SIZE];		\
+	uint8_t		pad1[OFI_CACHE_LINE_SIZE -		\
+			     sizeof(ofi_atomic64_t)];		\
+	int		size;					\
+	int		size_mask;				\
+	uint8_t		pad2[OFI_CACHE_LINE_SIZE -		\
+			     (sizeof(int) * 2)];		\
 	struct name ## _entry entry[];				\
 } __attribute__((__aligned__(64)));				\
 								\
@@ -114,6 +123,7 @@ static inline void name ## _init(struct name *aq, size_t size)	\
 {								\
 	size_t i;						\
 	assert(size == roundup_power_of_two(size));		\
+	assert(!((uintptr_t) aq % OFI_CACHE_LINE_SIZE));	\
 	aq->size = size;					\
 	aq->size_mask = aq->size - 1;				\
 	ofi_atomic_initialize64(&aq->write_pos, 0);		\
diff --git a/prov/shm/src/smr_util.c b/prov/shm/src/smr_util.c
index 190787ac935..16f157b5e80 100644
--- a/prov/shm/src/smr_util.c
+++ b/prov/shm/src/smr_util.c
@@ -108,8 +108,8 @@ size_t smr_calculate_size_offsets(size_t tx_count, size_t rx_count,
 	tx_size = roundup_power_of_two(tx_count);
 	rx_size = roundup_power_of_two(rx_count);
 
-	/* Align cmd_queue offset to 128-bit boundary. */
-	cmd_queue_offset = ofi_get_aligned_size(sizeof(struct smr_region), 16);
+	/* Align cmd_queue offset to cache line */
+	cmd_queue_offset = ofi_get_aligned_size(sizeof(struct smr_region), 64);
 	resp_queue_offset = cmd_queue_offset + sizeof(struct smr_cmd_queue) +
 			    sizeof(struct smr_cmd_queue_entry) * rx_size;
 	inject_pool_offset = resp_queue_offset + sizeof(struct smr_resp_queue) +