diff --git a/src/Cuda/xpmclient.cpp b/src/Cuda/xpmclient.cpp
index eac7b6d..14a3886 100644
--- a/src/Cuda/xpmclient.cpp
+++ b/src/Cuda/xpmclient.cpp
@@ -24,6 +24,7 @@
 #include "prime.h"
 #include <openssl/bn.h>
 #include <openssl/sha.h>
+#include <thread>
 
 void _blkmk_bin2hex(char *out, void *data, size_t datasz) {
   unsigned char *datac = (unsigned char *)data;
@@ -287,6 +288,7 @@ void PrimeMiner::Mining(GetBlockTemplateContext* gbp, SubmitContext* submit) {
   cudaBuffer<uint32_t> primeBuf[maxHashPrimorial];
   cudaBuffer<uint32_t> primeBuf2[maxHashPrimorial];
   
+  printf("-----+----------------------test print at line 291\n");
   CUevent sieveEvent;
   CUDA_SAFE_CALL(cuEventCreate(&sieveEvent, CU_EVENT_BLOCKING_SYNC));
   
@@ -700,7 +702,6 @@ void PrimeMiner::Mining(GetBlockTemplateContext* gbp, SubmitContext* submit) {
           work.multiplier[0] = buffer[3];
           std::reverse_copy(buffer+4, buffer+4+buffer[3], work.multiplier+1);
           submit->submitBlock(workTemplate, work, dataId);
-          
           LOG_F(1, "GPU %d found share: %d-ch type %d", mID, chainlength, candi.type+1);
           if(isblock)
             LOG_F(1, "GPU %d found BLOCK!", mID);
@@ -815,6 +816,21 @@ void initCmdLineOptions(option *options)
   options[clOptionLast] = {0, 0, 0, 0};
 }
 
+struct MineContext {
+  GetBlockTemplateContext *gbp;
+  SubmitContext *submit;
+  PrimeMiner* miner;
+  CUDADeviceInfo *device;
+  CUmodule *mod;
+};
+
+void* InvokeMining(void *arg)
+{
+  MineContext *ctx = (MineContext*)arg;
+  ctx->miner->Initialize(ctx->device->context, ctx->device->device, *(ctx->mod));
+  ctx->miner->Mining(ctx->gbp, ctx->submit);
+}
+
 int main(int argc, char **argv) {
   srand(time(0));  
   blkmk_sha256_impl = sha256;
@@ -885,8 +901,8 @@ int main(int argc, char **argv) {
     exit(1);
   }
   printf("block sum is %d\n", gThreadsNum);
-  GetBlockTemplateContext* getblock = new GetBlockTemplateContext(0, gUrl, gUserName, gPassword, gWallet, 4, gThreadsNum, extraNonce);
-  getblock->run();
+  GetBlockTemplateContext getblock(0, gUrl, gUserName, gPassword, gWallet, 4, gThreadsNum, extraNonce);
+  getblock.run();
   blktemplate_t *workTemplate = 0;
   unsigned int dataId;
   bool hasChanged;
@@ -896,8 +912,6 @@ int main(int argc, char **argv) {
   //}
   //printf("blocktemplate_rwrqwrqw %ld\n", (long int)workTemplate);
   //printf("block height %d\n", workTemplate->height);
-  
-  SubmitContext *submit = new SubmitContext(0, gUrl, gUserName, gPassword);
 
   {
     int np = sizeof(gPrimes)/sizeof(unsigned);
@@ -920,7 +934,7 @@ int main(int argc, char **argv) {
   std::map<int,int> mDeviceMap;
   std::map<int,int> mDeviceMapRev;
 
-  for (unsigned i = 0; i < devicesNum; i++) {
+  for (unsigned i = 0; i < 1; i++) {
       char name[128];
       CUDADeviceInfo info;
       mDeviceMap[i] = gpus.size();
@@ -988,10 +1002,15 @@ int main(int argc, char **argv) {
   }
 
   unsigned int sievePerRound = 5;
+  MineContext *mineCtx = new MineContext[gpus.size()];
   for(unsigned i = 0; i < gpus.size(); ++i) {
-      PrimeMiner* miner = new PrimeMiner(i, gpus.size(), sievePerRound, depth, clKernelLSize);
-      miner->Initialize(gpus[i].context, gpus[i].device, modules[i]);
-      miner->Mining(getblock, submit);
+      pthread_t thread;
+      mineCtx[i].gbp = &getblock;
+      mineCtx[i].submit = new SubmitContext(0, gUrl, gUserName, gPassword);
+      mineCtx[i].miner = new PrimeMiner(i, gpus.size(), sievePerRound, depth, clKernelLSize);
+      mineCtx[i].device = &gpus[i];
+      mineCtx[i].mod = &modules[i];
+      pthread_create(&thread, 0, InvokeMining, &mineCtx[i]);
   }
 
   return 0;