diff --git a/gpu-opt/Makefile b/gpu-opt/Makefile
index dbf5263..2a9668e 100644
--- a/gpu-opt/Makefile
+++ b/gpu-opt/Makefile
@@ -1,5 +1,5 @@
 main : main.cu util.h
-	nvcc -o main -std=c++11 main.cu
+	nvcc -o main -std=c++11 -O3 main.cu
 
 clean :
 	rm main
diff --git a/gpu-opt/main.cu b/gpu-opt/main.cu
index 791752d..78d26fb 100644
--- a/gpu-opt/main.cu
+++ b/gpu-opt/main.cu
@@ -9,6 +9,8 @@
 #include "util.h"
 #include "groups.h"
 
+__constant__ Rel rels[128];
+
 struct Row {
     int rel;
 
@@ -38,15 +40,12 @@ std::ostream &operator<<(std::ostream &o, const Row &r) {
 struct Solver {
     int ngens;
     int *cosets;
-    Rel *rels;
     
     Solver(int ngens,
-           thrust::device_vector<int> &cosets,
-           thrust::device_vector<Rel> &rels)
+           thrust::device_vector<int> &cosets)
         : ngens(ngens),
-           cosets(thrust::raw_pointer_cast(cosets.data())),
-           rels(thrust::raw_pointer_cast(rels.data())) {
-        }
+           cosets(thrust::raw_pointer_cast(cosets.data())) {
+    }
     
     __device__
     void operator()(Row &r) {
@@ -95,13 +94,11 @@ struct CosetInitializer {
 
 // this creates rows for cosets by index of each relation table
 struct RowGen {
-    Rel *rels;
-
     int coset;
 
-    RowGen(int coset, thrust::device_vector<Rel> &rels) 
-        : coset(coset),
-          rels(thrust::raw_pointer_cast(rels.data())) {}
+    RowGen(int coset) 
+        : coset(coset) {
+    }
 
     __device__
     Row operator()(int rel) {
@@ -130,7 +127,8 @@ bool add_coset(
         int ngens,
         int *coset,
         int *hint,
-        thrust::device_vector<int> &cosets) {
+        thrust::device_vector<int> &dcosets) {
+    thrust::host_vector<int> cosets = dcosets;
     *coset = cosets.size() / ngens;
 
     // todo: this part especially.
@@ -142,10 +140,10 @@ bool add_coset(
     int from = *hint / ngens;
     int gen = *hint % ngens;
     
-    add_row(ngens, cosets);
+    add_row(ngens, dcosets);
     
-    cosets[*hint] = *coset;
-    cosets[*coset * ngens + gen] = from;
+    dcosets[*hint] = *coset;
+    dcosets[*coset * ngens + gen] = from;
 
     return false;
 }
@@ -153,23 +151,23 @@ bool add_coset(
 // add a row for each relation table for some coset
 void gen_rows(
         int coset,
-        thrust::device_vector<Rel> &rels,
+        int nrels,
         thrust::device_vector<Row> &rows) {
-    rows.resize(rows.size() + rels.size());
+    rows.resize(rows.size() + nrels);
 
     thrust::counting_iterator<int> counter(0);
     thrust::transform(
             thrust::device,
-            counter, counter + rels.size(),
-            rows.end() - rels.size(),
-            RowGen(coset, rels));
+            counter, counter + nrels,
+            rows.end() - nrels,
+            RowGen(coset));
 }
 
 // do everything. data is implicitly passed to the device via device_vector.
 thrust::device_vector<int> solve(
         int ngens,
-        thrust::device_vector<int> subs,
-        thrust::device_vector<Rel> rels) {
+        int nrels,
+        thrust::device_vector<int> subs) {
     
     thrust::device_vector<int> cosets;
     thrust::device_vector<Row> rows;
@@ -182,7 +180,7 @@ thrust::device_vector<int> solve(
             CosetInitializer(cosets));
 
     // generate initial relation table rows for coset 0
-    gen_rows(0, rels, rows);
+    gen_rows(0, nrels, rows);
 
     // these keep track of what progress has been made
     int coset = 0;
@@ -191,7 +189,7 @@ thrust::device_vector<int> solve(
     // will break out later
     while (true) {
         // create a solver and apply it until nothing is being learned
-        Solver solve(ngens, cosets, rels);
+        Solver solve(ngens, cosets);
         thrust::for_each(
                 thrust::device,
                 rows.begin(), rows.end(),
@@ -205,7 +203,7 @@ thrust::device_vector<int> solve(
         if (done) break;
 
         // generate relation table rows for new coset
-        gen_rows(coset, rels, rows);
+        gen_rows(coset, nrels, rows);
 
         // move completed rows to the end of the list and remove.
         auto cut = thrust::partition(
@@ -224,8 +222,10 @@ int main(int argc, const char* argv[]) {
     cox = proc_args(argc, argv);
     std::vector<int> subs = {};
 
+    cudaMemcpyToSymbol(rels, cox.rels.data(), cox.rels.size() * sizeof(Rel));
+
     auto s = std::chrono::system_clock::now();
-    thrust::host_vector<int> cosets = solve(cox.ngens, subs, cox.rels);
+    thrust::host_vector<int> cosets = solve(cox.ngens, cox.rels.size(), subs);
     auto e = std::chrono::system_clock::now();
 
     std::chrono::duration<float> diff = e - s;
diff --git a/gpu-slo/Makefile b/gpu-slo/Makefile
index 0d927b1..950fd34 100644
--- a/gpu-slo/Makefile
+++ b/gpu-slo/Makefile
@@ -1,5 +1,5 @@
 main : main.cu util.h
-	nvcc -o main -std=c++11 main.cu
+	nvcc -o main -std=c++11 -O3 main.cu
 
 clean :
 	rm main