diff --git a/pkg/parcacol/querier.go b/pkg/parcacol/querier.go
index 593bebcc4ef..0cf28044b06 100644
--- a/pkg/parcacol/querier.go
+++ b/pkg/parcacol/querier.go
@@ -344,6 +344,7 @@ const (
 	ColumnPeriodSum   = "sum(" + profile.ColumnPeriod + ")"
 	ColumnValueCount  = "count(" + profile.ColumnValue + ")"
 	ColumnValueSum    = "sum(" + profile.ColumnValue + ")"
+	ColumnValueFirst  = "first(" + profile.ColumnValue + ")"
 )
 
 func (q *Querier) queryRangeDelta(ctx context.Context, filterExpr logicalplan.Expr, step time.Duration, sampleTypeUnit string) ([]*pb.MetricsSeries, error) {
@@ -531,13 +532,22 @@ func (q *Querier) queryRangeNonDelta(ctx context.Context, filterExpr logicalplan
 		Filter(filterExpr).
 		Aggregate(
 			[]logicalplan.Expr{
-				logicalplan.Sum(logicalplan.Col(profile.ColumnValue)),
+				logicalplan.Sum(logicalplan.Col(profile.ColumnValue)).Alias(ColumnValueFirst),
 			},
 			[]logicalplan.Expr{
 				logicalplan.DynCol(profile.ColumnLabels),
 				logicalplan.Col(profile.ColumnTimestamp),
 			},
 		).
+		Aggregate(
+			[]logicalplan.Expr{
+				logicalplan.Take(logicalplan.Col(profile.ColumnValue), 1).Alias(ColumnValueFirst),
+			},
+			[]logicalplan.Expr{
+				logicalplan.DynCol(profile.ColumnLabels),
+				logicalplan.Duration(1000 * time.Millisecond),
+			},
+		).
 		Execute(ctx, func(ctx context.Context, r arrow.Record) error {
 			r.Retain()
 			records = append(records, r)
@@ -561,7 +571,7 @@ func (q *Querier) queryRangeNonDelta(ctx context.Context, filterExpr logicalplan
 	// Add necessary columns and their found value is false by default.
 	columnIndices := map[string]columnIndex{
 		profile.ColumnTimestamp: {},
-		ColumnValueSum:          {},
+		ColumnValueFirst:        {},
 	}
 	labelColumnIndices := []int{}
 	labelSet := labels.Labels{}
@@ -623,17 +633,17 @@ func (q *Querier) queryRangeNonDelta(ctx context.Context, filterExpr logicalplan
 			}
 
 			ts := ar.Column(columnIndices[profile.ColumnTimestamp].index).(*array.Int64).Value(i)
-			value := ar.Column(columnIndices[ColumnValueSum].index).(*array.Int64).Value(i)
+
+			// value := ar.Column(columnIndices[ColumnValueFirst].index).(*array.Int64).Value(i)
+			valueList := ar.Column(columnIndices[ColumnValueFirst].index).(*array.List)
+			start, _ := valueList.ValueOffsets(i)
+			value := valueList.ListValues().(*array.Int64).Value(int(start))
 
 			// Each step bucket will only return one of the timestamps and its value.
 			// For this reason we'll take each timestamp and divide it by the step seconds.
 			// If we have seen a MetricsSample for this bucket before, we'll ignore this one.
 			// If we haven't seen one we'll add this sample to the response.
 
-			// TODO: This still queries way too much data from the underlying database.
-			// This needs to be moved to FrostDB to not even query all of this data in the first place.
-			// With a scrape interval of 10s and a query range of 1d we'd query 8640 samples and at most return 960.
-			// Even worse for a week, we'd query 60480 samples and only return 1000.
 			tsBucket := ts / 1000 / int64(step.Seconds())
 			if _, found := resSeriesBuckets[index][tsBucket]; found {
 				// We already have a MetricsSample for this timestamp bucket, ignore it.