diff --git a/src/main/java/org/apache/sysds/runtime/transform/encode/ColumnEncoder.java b/src/main/java/org/apache/sysds/runtime/transform/encode/ColumnEncoder.java
index d82d0a9b443..f69da6f6457 100644
--- a/src/main/java/org/apache/sysds/runtime/transform/encode/ColumnEncoder.java
+++ b/src/main/java/org/apache/sysds/runtime/transform/encode/ColumnEncoder.java
@@ -119,13 +119,29 @@ public MatrixBlock apply(CacheBlock in, MatrixBlock out, int outputCol, int rowS
 	protected abstract double[] getCodeCol(CacheBlock in, int startInd, int blkSize);
 
 
-	protected void applySparse(CacheBlock in, MatrixBlock out, int outputCol, int rowStart, int blk){
+	/*protected void applySparse(CacheBlock in, MatrixBlock out, int outputCol, int rowStart, int blk){
 		int index = _colID - 1;
 		for(int r = rowStart; r < getEndIndex(in.getNumRows(), rowStart, blk); r++) {
 			SparseRowVector row = (SparseRowVector) out.getSparseBlock().get(r);
 			row.values()[index] = getCode(in, r);
 			row.indexes()[index] = outputCol;
 		}
+	}*/
+
+	protected void applySparse(CacheBlock in, MatrixBlock out, int outputCol, int rowStart, int blk){
+		int index = _colID - 1;
+		// Apply loop tiling to exploit CPU caches
+		double[] codes = getCodeCol(in, rowStart, blk);
+		int rowEnd = getEndIndex(in.getNumRows(), rowStart, blk);
+		int B = 32; //tile size
+		for(int i = rowStart; i < rowEnd; i+=B) {
+			int lim = Math.min(i+B, rowEnd);
+			for (int ii=i; ii<lim; ii++) {
+				SparseRowVector row = (SparseRowVector) out.getSparseBlock().get(ii);
+				row.values()[index] = codes[ii-rowStart];
+				row.indexes()[index] = outputCol;
+			}
+		}
 	}
 
 	/*protected void applyDense(CacheBlock in, MatrixBlock out, int outputCol, int rowStart, int blk){
diff --git a/src/main/java/org/apache/sysds/runtime/transform/encode/ColumnEncoderDummycode.java b/src/main/java/org/apache/sysds/runtime/transform/encode/ColumnEncoderDummycode.java
index 27fe6fdd4a4..a5670ae8cc4 100644
--- a/src/main/java/org/apache/sysds/runtime/transform/encode/ColumnEncoderDummycode.java
+++ b/src/main/java/org/apache/sysds/runtime/transform/encode/ColumnEncoderDummycode.java
@@ -125,19 +125,28 @@ protected void applyDense(CacheBlock in, MatrixBlock out, int outputCol, int row
 			throw new DMLRuntimeException("ColumnEncoderDummycode called with: " + in.getClass().getSimpleName() +
 					" and not MatrixBlock");
 		}
-		for(int i = rowStart; i < getEndIndex(in.getNumRows(), rowStart, blk); i++) {
-			// Using outputCol here as index since we have a MatrixBlock as input where dummycoding could have been
-			// applied in a previous encoder
-			double val = in.getDouble(i, outputCol);
-			if(Double.isNaN(val)){
-				// 0 if NaN
-				out.quickSetValue(i, outputCol, 0);
-				continue;
+		int rowEnd = getEndIndex(in.getNumRows(), rowStart, blk);
+		double vals[] = new double[rowEnd -rowStart];
+		for (int i=rowStart; i<rowEnd; i++)
+			vals[i-rowStart] = in.getDouble(i, outputCol);
+
+		// Using outputCol here as index since we have a MatrixBlock as input where 
+		// dummycoding might have been applied in a previous encoder
+		int B = 32;
+		for(int i=rowStart; i<rowEnd; i+=B) {
+			// Apply loop tiling to exploit CPU caches
+			int lim = Math.min(i+B, rowEnd);
+			for (int ii=i; ii<lim; ii++) {
+				double val = vals[ii-rowStart];
+				if(Double.isNaN(val)) {
+					out.quickSetValue(ii, outputCol, 0); //0 if NaN
+					continue;
+				}
+				int nCol = outputCol + (int) val - 1;
+				if(nCol != outputCol)
+					out.quickSetValue(ii, outputCol, 0);
+				out.quickSetValue(ii, nCol, 1);
 			}
-			int nCol = outputCol + (int) val - 1;
-			if(nCol != outputCol)
-				out.quickSetValue(i, outputCol, 0);
-			out.quickSetValue(i, nCol, 1);
 		}
 	}
 
diff --git a/src/main/java/org/apache/sysds/runtime/transform/encode/ColumnEncoderPassThrough.java b/src/main/java/org/apache/sysds/runtime/transform/encode/ColumnEncoderPassThrough.java
index e249159c095..57405b04203 100644
--- a/src/main/java/org/apache/sysds/runtime/transform/encode/ColumnEncoderPassThrough.java
+++ b/src/main/java/org/apache/sysds/runtime/transform/encode/ColumnEncoderPassThrough.java
@@ -78,16 +78,23 @@ protected double[] getCodeCol(CacheBlock in, int startInd, int blkSize) {
 	protected void applySparse(CacheBlock in, MatrixBlock out, int outputCol, int rowStart, int blk){
 		Set<Integer> sparseRowsWZeros = null;
 		int index = _colID - 1;
-		for(int r = rowStart; r < getEndIndex(in.getNumRows(), rowStart, blk); r++) {
-			double v = getCode(in, r);
-			SparseRowVector row = (SparseRowVector) out.getSparseBlock().get(r);
-			if(v == 0) {
-				if(sparseRowsWZeros == null)
-					sparseRowsWZeros = new HashSet<>();
-				sparseRowsWZeros.add(r);
+		// Apply loop tiling to exploit CPU caches
+		double[] codes = getCodeCol(in, rowStart, blk);
+		int rowEnd = getEndIndex(in.getNumRows(), rowStart, blk);
+		int B = 32; //tile size
+		for(int i = rowStart; i < rowEnd; i+=B) {
+			int lim = Math.min(i+B, rowEnd);
+			for (int ii=i; ii<lim; ii++) {
+				double v = codes[ii-rowStart];
+				SparseRowVector row = (SparseRowVector) out.getSparseBlock().get(ii);
+				if(v == 0) {
+					if(sparseRowsWZeros == null)
+						sparseRowsWZeros = new HashSet<>();
+					sparseRowsWZeros.add(ii);
+				}
+				row.values()[index] = v;
+				row.indexes()[index] = outputCol;
 			}
-			row.values()[index] = v;
-			row.indexes()[index] = outputCol;
 		}
 		if(sparseRowsWZeros != null){
 			addSparseRowsWZeros(sparseRowsWZeros);