diff --git a/ggml/src/ggml-vulkan/ggml-vulkan-push-constants.h b/ggml/src/ggml-vulkan/ggml-vulkan-push-constants.h
index 8446e313c..f066d1064 100644
--- a/ggml/src/ggml-vulkan/ggml-vulkan-push-constants.h
+++ b/ggml/src/ggml-vulkan/ggml-vulkan-push-constants.h
@@ -742,6 +742,10 @@ struct vk_op_conv2d_push_constants {
     // init_fastdiv_values constants for dividing by OW, OW*OH
     uint32_t OWmp;   uint32_t OWL;
     uint32_t OWOHmp; uint32_t OWOHL;
+
+    uint32_t knl_offset;
+    uint32_t src_offset;
+    uint32_t dst_offset;
 };
 
 template <> inline void init_pushconst_fastdiv(vk_op_conv2d_push_constants &p) {
@@ -777,6 +781,10 @@ struct vk_op_conv3d_push_constants {
     uint32_t OWmp;     uint32_t OWL;
     uint32_t OWOHmp;   uint32_t OWOHL;
     uint32_t OWOHODmp; uint32_t OWOHODL;
+
+    uint32_t knl_offset;
+    uint32_t src_offset;
+    uint32_t dst_offset;
 };
 
 template <> inline void init_pushconst_fastdiv(vk_op_conv3d_push_constants &p) {
@@ -1032,6 +1040,24 @@ template <> inline void init_pushconst_tensor_offsets(ggml_backend_vk_context *
     GGML_UNUSED(src3);
 }
 
+template <> inline void init_pushconst_tensor_offsets(ggml_backend_vk_context * ctx, vk_op_conv2d_push_constants &p, const ggml_tensor * src0, const ggml_tensor * src1, const ggml_tensor * src2, const ggml_tensor * src3, ggml_tensor * dst) {
+    p.knl_offset = get_misalign_bytes(ctx, src0) / ggml_type_size(src0->type);
+    p.src_offset = get_misalign_bytes(ctx, src1) / ggml_type_size(src1->type);
+    p.dst_offset = get_misalign_bytes(ctx, dst)  / ggml_type_size(dst->type);
+
+    GGML_UNUSED(src2);
+    GGML_UNUSED(src3);
+}
+
+template <> inline void init_pushconst_tensor_offsets(ggml_backend_vk_context * ctx, vk_op_conv3d_push_constants &p, const ggml_tensor * src0, const ggml_tensor * src1, const ggml_tensor * src2, const ggml_tensor * src3, ggml_tensor * dst) {
+    p.knl_offset = get_misalign_bytes(ctx, src0) / ggml_type_size(src0->type);
+    p.src_offset = get_misalign_bytes(ctx, src1) / ggml_type_size(src1->type);
+    p.dst_offset = get_misalign_bytes(ctx, dst)  / ggml_type_size(dst->type);
+
+    GGML_UNUSED(src2);
+    GGML_UNUSED(src3);
+}
+
 template <> inline void init_pushconst_tensor_offsets(ggml_backend_vk_context * ctx, vk_op_im2col_3d_push_constants &p, const ggml_tensor * src0, const ggml_tensor * src1, const ggml_tensor * src2, const ggml_tensor * src3, ggml_tensor * dst) {
     const uint32_t a_offset = get_misalign_bytes(ctx, src1) / ggml_type_size(src1->type);
     const uint32_t d_offset = get_misalign_bytes(ctx, dst) / ggml_type_size(dst->type);
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp b/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp
index c64004cdc..5ed15a258 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp
@@ -62,6 +62,11 @@ layout(push_constant) uniform parameter {
     // fastdiv helper values
     uint32_t OWmp;   uint32_t OWL;
     uint32_t OWOHmp; uint32_t OWOHL;
+
+    // element offsets for misaligned buffer bindings
+    uint32_t knl_offset;
+    uint32_t src_offset;
+    uint32_t dst_offset;
 }
 
 p;
@@ -206,7 +211,7 @@ ACC_TYPE perElemOpStore(const in uint32_t r, const in uint32_t c, const in ACC_T
     uint32_t OW_idx  = NPQ_idx - N_idx * p.OH * p.OW - OH_idx * p.OW;
     uint32_t dst_idx = OW_idx + OH_idx * p.nb1 + K_idx * p.nb2 + N_idx * p.nb3;
     if (aligned != 0 || (K_idx < K && NPQ_idx < NPQ)) {
-        dst_data[dst_idx] = D_TYPE(elem);
+        dst_data[dst_idx + p.dst_offset] = D_TYPE(elem);
     }
     return elem;
 }
@@ -286,7 +291,7 @@ void main() {
             if (aligned == 0) {
                 knl_idx = min(knl_idx, K * CRS - 1);
             }
-            float    val     = knl_data[knl_idx];
+            float val = knl_data[knl_idx + p.knl_offset];
             if (aligned == 0 && (K_idx >= K || CRS_idx_a >= CRS)) {
                 val = 0.0;
             }
@@ -341,7 +346,7 @@ void main() {
             if (aligned == 0 || !hw_in_bounds || !stride_in_bounds) {
                 src_idx = min(max(src_idx, 0), p.Cin * p.N * p.W * p.H - 1);
             }
-            float val = src_data[src_idx];
+            float val = src_data[src_idx + p.src_offset];
             bool oob = false;
             if (aligned == 0 && (CRS_idx_b >= CRS || NPQ_idx >= NPQ)) {
                 oob = true;
@@ -444,7 +449,7 @@ void main() {
             uint32_t OW_idx  = NPQ_idx - N_idx * p.OH * p.OW - OH_idx * p.OW;
             uint32_t dst_idx = OW_idx + OH_idx * p.nb1 + K_idx * p.nb2 + N_idx * p.nb3;
             if (aligned != 0 || (K_idx < K && NPQ_idx < NPQ)) {
-                dst_data[dst_idx] = D_TYPE(Csh[k_local * Csh_stride + npq_thread]);
+                dst_data[dst_idx + p.dst_offset] = D_TYPE(Csh[k_local * Csh_stride + npq_thread]);
             }
         }
     }
@@ -464,7 +469,7 @@ void main() {
                 uint32_t OW_idx  = NPQ_idx - N_idx * p.OH * p.OW - OH_idx * p.OW;
                 uint32_t dst_idx = OW_idx + OH_idx * p.nb1 + K_idx * p.nb2 + N_idx * p.nb3;
                 if (aligned != 0 || (K_idx < K && NPQ_idx < NPQ)) {
-                    dst_data[dst_idx] = regC[T_ly][T_lx];
+                    dst_data[dst_idx + p.dst_offset] = regC[T_ly][T_lx];
                 }
             }
         }
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/conv3d_mm.comp b/ggml/src/ggml-vulkan/vulkan-shaders/conv3d_mm.comp
index d5ce4290b..6919861b3 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/conv3d_mm.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/conv3d_mm.comp
@@ -61,6 +61,11 @@ layout(push_constant) uniform parameter {
     uint32_t OWmp;   uint32_t OWL;
     uint32_t OWOHmp; uint32_t OWOHL;
     uint32_t OWOHODmp; uint32_t OWOHODL;
+
+    // element offsets for misaligned buffer bindings
+    uint32_t knl_offset;
+    uint32_t src_offset;
+    uint32_t dst_offset;
 }
 
 p;
@@ -214,7 +219,7 @@ ACC_TYPE perElemOpStore(const in uint32_t r, const in uint32_t c, const in ACC_T
     split_npq(NPQ_idx, N_idx, OD_idx, OH_idx, OW_idx);
     uint32_t dst_idx = OW_idx + OH_idx * p.nb1 + OD_idx * p.nb2 + (N_idx * p.OC + K_idx) * p.nb3;
     if (aligned != 0 || (K_idx < K && NPQ_idx < NPQ)) {
-        dst_data[dst_idx] = D_TYPE(elem);
+        dst_data[dst_idx + p.dst_offset] = D_TYPE(elem);
     }
     return elem;
 }
@@ -261,7 +266,7 @@ void main() {
             if (aligned == 0) {
                 knl_idx = min(knl_idx, K * CRS - 1);
             }
-            float    val     = knl_data[knl_idx];
+            float val = knl_data[knl_idx + p.knl_offset];
             if (aligned == 0 && (K_idx >= K || CRS_idx_a >= CRS)) {
                 val = 0.0;
             }
@@ -294,7 +299,7 @@ void main() {
             if (aligned == 0 || !dhw_in_bounds) {
                 src_idx = min(src_idx, p.IC * p.N * p.IW * p.IH * p.ID - 1);
             }
-            float val = src_data[src_idx];
+            float val = src_data[src_idx + p.src_offset];
             bool oob = false;
             if (aligned == 0 && (CRS_idx_b >= CRS || NPQ_idx >= NPQ)) {
                 oob = true;
@@ -393,7 +398,7 @@ void main() {
             split_npq(NPQ_idx, N_idx, OD_idx, OH_idx, OW_idx);
             uint32_t dst_idx = OW_idx + OH_idx * p.nb1 + OD_idx * p.nb2 + (N_idx * p.OC + K_idx) * p.nb3;
             if (aligned != 0 || (K_idx < K && NPQ_idx < NPQ)) {
-                dst_data[dst_idx] = D_TYPE(Csh[k_local * Csh_stride + npq_thread]);
+                dst_data[dst_idx + p.dst_offset] = D_TYPE(Csh[k_local * Csh_stride + npq_thread]);
             }
         }
     }
@@ -415,7 +420,7 @@ void main() {
                 split_npq(NPQ_idx, N_idx, OD_idx, OH_idx, OW_idx);
                 uint32_t dst_idx = OW_idx + OH_idx * p.nb1 + OD_idx * p.nb2 + (N_idx * p.OC + K_idx) * p.nb3;
                 if (aligned != 0 || (K_idx < K && NPQ_idx < NPQ)) {
-                    dst_data[dst_idx] = D_TYPE(regC[T_ly][T_lx]);
+                    dst_data[dst_idx + p.dst_offset] = D_TYPE(regC[T_ly][T_lx]);
                 }
             }
         }