diff --git a/Demos3/BasicGpuDemo/BasicGpuDemo.cpp b/Demos3/BasicGpuDemo/BasicGpuDemo.cpp
index 73dc48cc0..eff53d1ea 100644
--- a/Demos3/BasicGpuDemo/BasicGpuDemo.cpp
+++ b/Demos3/BasicGpuDemo/BasicGpuDemo.cpp
@@ -15,9 +15,9 @@ subject to the following restrictions:
 
 
 ///create 125 (5x5x5) dynamic object
-#define ARRAY_SIZE_X 30
+#define ARRAY_SIZE_X 25
 #define ARRAY_SIZE_Y 20
-#define ARRAY_SIZE_Z 30
+#define ARRAY_SIZE_Z 25
 
 //maximum number of objects (and allow user to shoot additional boxes)
 #define MAX_PROXIES (ARRAY_SIZE_X*ARRAY_SIZE_Y*ARRAY_SIZE_Z + 1024)
@@ -164,7 +164,7 @@ void BasicGpuDemo::exitCL()
 BasicGpuDemo::BasicGpuDemo()
 {
 	m_clData = new btInternalData;
-	setCameraDistance(btScalar(SCALING*120.));
+	setCameraDistance(btScalar(SCALING*60.));
 	this->setAzi(45);
 	this->setEle(45);
 
@@ -181,8 +181,8 @@ BasicGpuDemo::~BasicGpuDemo()
 void	BasicGpuDemo::initPhysics()
 {
 	setTexturing(true);
-	setShadows(true);
-
+	setShadows(false);//too slow with many objects
+	
 	
 	///collision configuration contains default setup for memory, collision setup
 	m_collisionConfiguration = 0;
diff --git a/src/Bullet3OpenCL/RigidBody/b3GpuRigidBodyPipeline.cpp b/src/Bullet3OpenCL/RigidBody/b3GpuRigidBodyPipeline.cpp
index 4f5f98812..dcdc0123f 100644
--- a/src/Bullet3OpenCL/RigidBody/b3GpuRigidBodyPipeline.cpp
+++ b/src/Bullet3OpenCL/RigidBody/b3GpuRigidBodyPipeline.cpp
@@ -19,7 +19,7 @@
 #define B3_RIGIDBODY_UPDATEAABB_PATH "src/Bullet3OpenCL/RigidBody/kernels/updateAabbsKernel.cl"
 
 bool useDbvt = false;
-bool useBullet2CpuSolver = false;//false;
+bool useBullet2CpuSolver = true;
 bool dumpContactStats = false;
 
 #ifdef TEST_OTHER_GPU_SOLVER
@@ -200,7 +200,8 @@ void	b3GpuRigidBodyPipeline::stepSimulation(float deltaTime)
 	b3OpenCLArray<b3Contact4> gpuContacts(m_data->m_context,m_data->m_queue,0,true);
 	gpuContacts.setFromOpenCLBuffer(m_data->m_narrowphase->getContactsGpu(),m_data->m_narrowphase->getNumContactsGpu());
 
-	if (useBullet2CpuSolver)
+	int numJoints = m_data->m_joints.size();
+	if (useBullet2CpuSolver && numJoints)
 	{
 
 		b3AlignedObjectArray<b3RigidBodyCL> hostBodies;
@@ -210,7 +211,6 @@ void	b3GpuRigidBodyPipeline::stepSimulation(float deltaTime)
 		b3AlignedObjectArray<b3Contact4> hostContacts;
 		gpuContacts.copyToHost(hostContacts);
 		{
-			int numJoints = m_data->m_joints.size();
 			b3TypedConstraint** joints = numJoints? &m_data->m_joints[0] : 0;
 			b3Contact4* contacts = numContacts? &hostContacts[0]: 0;
 //			m_data->m_solver->solveContacts(m_data->m_narrowphase->getNumBodiesGpu(),&hostBodies[0],&hostInertias[0],numContacts,contacts,numJoints, joints);