diff --git a/apps/benchmarks/benchmark/fconv3d.bmark b/apps/benchmarks/benchmark/fconv3d.bmark
index b2af37e3e..9e3186a03 100644
--- a/apps/benchmarks/benchmark/fconv3d.bmark
+++ b/apps/benchmarks/benchmark/fconv3d.bmark
@@ -43,7 +43,11 @@ void warm_caches(uint64_t heat) {
   volatile double buf;
 
   for (uint64_t k = 0; k < heat; ++k)
+#ifndef SCALAR
     fconv3d_CHx7x7(o, i, f, M, N, CH, F);
+#else
+    fconv3d_CVxFxF(o, i, f, M, N, CH, F);
+#endif
 // The following artificial warming ensures, with a larger cache,
 // not to experience any cache misses
 #ifdef AD_HOC_WARMING
@@ -64,7 +68,11 @@ int main() {
   HW_CNT_READY;
   start_timer();
   if (F == 7)
+#ifndef SCALAR
     fconv3d_CHx7x7(o, i, f, M, N, CH, F);
+#else
+    fconv3d_CVxFxF(o, i, f, M, N, CH, F);
+#endif
   else {
     printf("Error: the filter size is different from 7.\n");
     return -1;
diff --git a/apps/fconv3d/fconv3d.h b/apps/fconv3d/fconv3d.h
index 26b24d659..15a2d4986 100644
--- a/apps/fconv3d/fconv3d.h
+++ b/apps/fconv3d/fconv3d.h
@@ -22,6 +22,9 @@
 #include <stdint.h>
 #include <stdio.h>
 
+double* conv3d_CHxFxF_scalar(double *o, double *i, double *f, int64_t M, int64_t N,
+                    int64_t C, int64_t F);
+
 void fconv3d_CHx7x7(double *o, double *i, double *f, int64_t M, int64_t N,
                     int64_t C, int64_t F);
 
diff --git a/apps/fconv3d/fconv3d_3x7x7.c b/apps/fconv3d/fconv3d_3x7x7.c
index d4da1b9e9..c418a2f14 100644
--- a/apps/fconv3d/fconv3d_3x7x7.c
+++ b/apps/fconv3d/fconv3d_3x7x7.c
@@ -52,6 +52,41 @@
 
 extern int64_t event_trigger;
 
+// a - 2D matrix (as a 1D array), w - kernel
+double* conv3d_CHxFxF_scalar(double *o, double *i, double *f, int64_t M, int64_t N,
+                    int64_t C, int64_t F) {
+    double acc;
+    int i;
+    int j;
+    int k1, k2;
+    int l1, l2;
+    int t1, t2;
+    int ch;
+
+    for(i = 0; i < (M + 2*F - 1); i++)
+    {
+        t1 = i * N; // loop invariants
+        for(j = 0; j < (N + 2*F - 1); j++)
+        {
+            acc = 0.0;
+            for (ch = 0; ch < C; ch++)
+            {
+              for(k1 = F - 1, k2 = 0; k1 >= 0; k1--, k2++)
+              {
+                  t2 = k1 * F;  // loop invariants
+                  for(l1 = F - 1, l2 = 0; l1 >= 0; l1--, l2++)
+                  {
+                      acc += w[t2 + l1 + F * F * ch] * a[(i + k2) * N + (j + l2) + (M + 2*F - 1) * (N + 2*F - 1) * ch];
+                  }
+              }
+            }
+            result[t1 + j] = acc;
+        }
+    }
+
+    return result;
+}
+
 void fconv3d_CHx7x7(double *o, double *i, double *f, int64_t M, int64_t N,
                     int64_t C, int64_t F) {
 
diff --git a/apps/fconv3d/main.c b/apps/fconv3d/main.c
index 991691e29..7f94db061 100644
--- a/apps/fconv3d/main.c
+++ b/apps/fconv3d/main.c
@@ -85,7 +85,11 @@ int main() {
   // Call the main kernel, and measure cycles
   start_timer();
   if (F == 7)
+#ifndef SCALAR
     fconv3d_CHx7x7(o, i, f, M, N, CH, F);
+#else
+    fconv3d_CHxFxF_scalar(golden_o, i, f, M, N, CH, F);
+#endif
   else
     printf("Error: the filter size is different from 7.\n");
   stop_timer();