diff --git a/diskann-disk/src/search/pq/pq_scratch.rs b/diskann-disk/src/search/pq/pq_scratch.rs
index e90af5212..1df666e18 100644
--- a/diskann-disk/src/search/pq/pq_scratch.rs
+++ b/diskann-disk/src/search/pq/pq_scratch.rs
@@ -78,6 +78,12 @@ impl PQScratch {
         self.query_scratch.copy_from_slice(&query[..dim]);
         Ok(())
     }
+
+    /// Return the largest number of PQ vectors whose distances can be computed using this
+    /// scratch data structure.
+    pub(crate) fn max_vectors(&self) -> usize {
+        self.aligned_dist_scratch.len()
+    }
 }
 
 #[cfg(test)]
@@ -112,6 +118,8 @@ mod tests {
             0
         );
 
+        assert_eq!(pq_scratch.max_vectors(), graph_degree);
+
         // Test set() method
         let query: Vec<f32> = (1..=dim).map(|i| i as f32).collect();
         pq_scratch.set(&query).unwrap();
diff --git a/diskann-disk/src/search/provider/disk_provider.rs b/diskann-disk/src/search/provider/disk_provider.rs
index 1dd518781..da54a58db 100644
--- a/diskann-disk/src/search/provider/disk_provider.rs
+++ b/diskann-disk/src/search/provider/disk_provider.rs
@@ -924,21 +924,42 @@ where
         let mut accessor = strategy
             .search_accessor(provider, &DefaultContext)
             .into_ann_result()?;
-        let computer = accessor.build_query_computer(query).into_ann_result()?;
+
+        // Derive the batch size from the scratch data structure. Providing too many vectors
+        // will panic.
+        let batch_size = accessor.scratch.pq_scratch.max_vectors();
+
+        // This check should always hold since `graph_degree` comes from
+        // `diskann::graph::Config` and is forced to be non-zero. But this is defensive
+        // against misconfiguration.
+        if batch_size == 0 {
+            return Err(ANNError::message(
+                diskann::ANNErrorKind::IndexError,
+                "pq scratch must support at least one vector",
+            ));
+        }
+
+        let mut id_buffer = Vec::with_capacity(batch_size);
 
         let mut best = NeighborPriorityQueue::new(neighbors_before_reranking);
         let mut cmps = 0u32;
 
-        let num_points = provider.num_points as u32;
-        for id in 0..num_points {
-            if vector_filter(&id) {
-                let element = accessor.get_element(id).await.into_ann_result()?;
-                let dist = computer.evaluate_similarity(element);
-                best.insert(Neighbor::new(id, dist));
-                cmps += 1;
+        let mut iter = (0..provider.num_points as u32).filter(vector_filter);
+        loop {
+            id_buffer.clear();
+            id_buffer.extend(iter.by_ref().take(batch_size));
+
+            if id_buffer.is_empty() {
+                break;
             }
+
+            accessor.pq_distances(&id_buffer, |dist, id| best.insert(Neighbor::new(id, dist)))?;
+            cmps += id_buffer.len() as u32;
         }
 
+        // FIXME: This is a temporary bridge. We don't really need the query computer, but
+        // we do need to satisfy the trait definition until PR 1067 lands.
+        let computer = accessor.build_query_computer(query).into_ann_result()?;
         let result_count = strategy
             .default_post_processor()
             .post_process(&mut accessor, query, &computer, best.iter(), output)