diff --git a/src/cpu/base.cc b/src/cpu/base.cc
index 8121307..67f8e7b 100644
--- a/src/cpu/base.cc
+++ b/src/cpu/base.cc
@@ -198,7 +198,11 @@
     for (int i = 0; i < numThreads; i++) {
         fetchStats.emplace_back(new FetchCPUStats(this, i));
         executeStats.emplace_back(new ExecuteCPUStats(this, i));
-        commitStats.emplace_back(new CommitCPUStats(this, i));
+        // create commitStat object for thread i and set ipc, cpi formulas
+        CommitCPUStats* commitStatptr = new CommitCPUStats(this, i);
+        commitStatptr->ipc = commitStatptr->numInsts / baseStats.numCycles;
+        commitStatptr->cpi = baseStats.numCycles / commitStatptr->numInsts;
+        commitStats.emplace_back(commitStatptr);
     }
 }
 
@@ -392,13 +396,28 @@
 BaseCPU::
 BaseCPUStats::BaseCPUStats(statistics::Group *parent)
     : statistics::Group(parent),
+      ADD_STAT(numInsts, statistics::units::Count::get(),
+               "Number of instructions committed (core level)"),
+      ADD_STAT(numOps, statistics::units::Count::get(),
+               "Number of ops (including micro ops) committed (core level)"),
       ADD_STAT(numCycles, statistics::units::Cycle::get(),
                "Number of cpu cycles simulated"),
+      ADD_STAT(cpi, statistics::units::Rate<
+                statistics::units::Cycle, statistics::units::Count>::get(),
+               "CPI: cycles per instruction (core level)"),
+      ADD_STAT(ipc, statistics::units::Rate<
+                statistics::units::Count, statistics::units::Cycle>::get(),
+               "IPC: instructions per cycle (core level)"),
       ADD_STAT(numWorkItemsStarted, statistics::units::Count::get(),
                "Number of work items this cpu started"),
       ADD_STAT(numWorkItemsCompleted, statistics::units::Count::get(),
                "Number of work items this cpu completed")
 {
+    cpi.precision(6);
+    cpi = numCycles / numInsts;
+
+    ipc.precision(6);
+    ipc = numInsts / numCycles;
 }
 
 void
@@ -839,6 +858,10 @@
 BaseCPU::
 FetchCPUStats::FetchCPUStats(statistics::Group *parent, int thread_id)
     : statistics::Group(parent, csprintf("fetchStats%i", thread_id).c_str()),
+    ADD_STAT(numInsts, statistics::units::Count::get(),
+             "Number of instructions fetched (thread level)"),
+    ADD_STAT(numOps, statistics::units::Count::get(),
+             "Number of ops (including micro ops) fetched (thread level)"),
     ADD_STAT(numBranches, statistics::units::Count::get(),
              "Number of branches fetched"),
     ADD_STAT(numFetchSuspends, statistics::units::Count::get(),
@@ -927,6 +950,16 @@
 BaseCPU::
 CommitCPUStats::CommitCPUStats(statistics::Group *parent, int thread_id)
     : statistics::Group(parent, csprintf("commitStats%i", thread_id).c_str()),
+    ADD_STAT(numInsts, statistics::units::Count::get(),
+             "Number of instructions committed (thread level)"),
+    ADD_STAT(numOps, statistics::units::Count::get(),
+             "Number of ops (including micro ops) committed (thread level)"),
+    ADD_STAT(cpi, statistics::units::Rate<
+                statistics::units::Cycle, statistics::units::Count>::get(),
+             "CPI: cycles per instruction (thread level)"),
+    ADD_STAT(ipc, statistics::units::Rate<
+                statistics::units::Count, statistics::units::Cycle>::get(),
+             "IPC: instructions per cycle (thread level)"),
     ADD_STAT(numMemRefs, statistics::units::Count::get(),
             "Number of memory references committed"),
     ADD_STAT(numFpInsts, statistics::units::Count::get(),
@@ -944,6 +977,9 @@
     ADD_STAT(committedControl, statistics::units::Count::get(),
              "Class of control type instructions committed")
 {
+    cpi.precision(6);
+    ipc.precision(6);
+
     committedInstType
         .init(enums::Num_OpClass)
         .flags(statistics::total | statistics::pdf | statistics::dist);
diff --git a/src/cpu/base.hh b/src/cpu/base.hh
index 5b2e97f..06fc2a3 100644
--- a/src/cpu/base.hh
+++ b/src/cpu/base.hh
@@ -633,8 +633,14 @@
     struct BaseCPUStats : public statistics::Group
     {
         BaseCPUStats(statistics::Group *parent);
+        // Number of CPU insts and ops committed at CPU core level
+        statistics::Scalar numInsts;
+        statistics::Scalar numOps;
         // Number of CPU cycles simulated
         statistics::Scalar numCycles;
+        /* CPI/IPC for total cycle counts and macro insts */
+        statistics::Formula cpi;
+        statistics::Formula ipc;
         statistics::Scalar numWorkItemsStarted;
         statistics::Scalar numWorkItemsCompleted;
     } baseStats;
@@ -683,6 +689,12 @@
     {
         FetchCPUStats(statistics::Group *parent, int thread_id);
 
+        /* Total number of instructions fetched */
+        statistics::Scalar numInsts;
+
+        /* Total number of operations fetched */
+        statistics::Scalar numOps;
+
         /* Total number of branches fetched */
         statistics::Scalar numBranches;
 
@@ -742,6 +754,14 @@
     {
         CommitCPUStats(statistics::Group *parent, int thread_id);
 
+        /* Number of simulated instructions committed */
+        statistics::Scalar numInsts;
+        statistics::Scalar numOps;
+
+        /* CPI/IPC for total cycle counts and macro insts */
+        statistics::Formula cpi;
+        statistics::Formula ipc;
+
         /* Number of committed memory references. */
         statistics::Scalar numMemRefs;
 
diff --git a/src/cpu/minor/execute.cc b/src/cpu/minor/execute.cc
index 5c0354b..2908c22 100644
--- a/src/cpu/minor/execute.cc
+++ b/src/cpu/minor/execute.cc
@@ -871,14 +871,16 @@
     {
         thread->numInst++;
         thread->threadStats.numInsts++;
-        cpu.stats.numInsts++;
+        cpu.commitStats[inst->id.threadId]->numInsts++;
+        cpu.baseStats.numInsts++;
 
         /* Act on events related to instruction counts */
         thread->comInstEventQueue.serviceEvents(thread->numInst);
     }
     thread->numOp++;
     thread->threadStats.numOps++;
-    cpu.stats.numOps++;
+    cpu.commitStats[inst->id.threadId]->numOps++;
+    cpu.baseStats.numOps++;
     cpu.commitStats[inst->id.threadId]
         ->committedInstType[inst->staticInst->opClass()]++;
 
diff --git a/src/cpu/minor/stats.cc b/src/cpu/minor/stats.cc
index b20ce95..e31cbe9 100644
--- a/src/cpu/minor/stats.cc
+++ b/src/cpu/minor/stats.cc
@@ -45,29 +45,13 @@
 
 MinorStats::MinorStats(BaseCPU *base_cpu)
     : statistics::Group(base_cpu),
-    ADD_STAT(numInsts, statistics::units::Count::get(),
-             "Number of instructions committed"),
-    ADD_STAT(numOps, statistics::units::Count::get(),
-             "Number of ops (including micro ops) committed"),
     ADD_STAT(quiesceCycles, statistics::units::Cycle::get(),
              "Total number of cycles that CPU has spent quiesced or waiting "
-             "for an interrupt"),
-    ADD_STAT(cpi, statistics::units::Rate<
-                statistics::units::Cycle, statistics::units::Count>::get(),
-             "CPI: cycles per instruction"),
-    ADD_STAT(ipc, statistics::units::Rate<
-                statistics::units::Count, statistics::units::Cycle>::get(),
-             "IPC: instructions per cycle")
+             "for an interrupt")
 
 {
     quiesceCycles.prereq(quiesceCycles);
 
-    cpi.precision(6);
-    cpi = base_cpu->baseStats.numCycles / numInsts;
-
-    ipc.precision(6);
-    ipc = numInsts / base_cpu->baseStats.numCycles;
-
 }
 
 } // namespace minor
diff --git a/src/cpu/minor/stats.hh b/src/cpu/minor/stats.hh
index f7d5e71..98ac80f 100644
--- a/src/cpu/minor/stats.hh
+++ b/src/cpu/minor/stats.hh
@@ -59,19 +59,9 @@
 {
     MinorStats(BaseCPU *parent);
 
-    /** Number of simulated instructions */
-    statistics::Scalar numInsts;
-
-    /** Number of simulated insts and microops */
-    statistics::Scalar numOps;
-
     /** Number of cycles in quiescent state */
     statistics::Scalar quiesceCycles;
 
-    /** CPI/IPC for total cycle counts and macro insts */
-    statistics::Formula cpi;
-    statistics::Formula ipc;
-
 };
 
 } // namespace minor
diff --git a/src/cpu/simple/base.cc b/src/cpu/simple/base.cc
index 70da659..35d1490 100644
--- a/src/cpu/simple/base.cc
+++ b/src/cpu/simple/base.cc
@@ -154,10 +154,36 @@
 
     if (!curStaticInst->isMicroop() || curStaticInst->isLastMicroop()) {
         t_info.numInst++;
-        t_info.execContextStats.numInsts++;
     }
     t_info.numOp++;
-    t_info.execContextStats.numOps++;
+}
+
+void
+BaseSimpleCPU::countFetchInst()
+{
+    SimpleExecContext& t_info = *threadInfo[curThread];
+
+    if (!curStaticInst->isMicroop() || curStaticInst->isLastMicroop()) {
+        // increment thread level numInsts fetched count
+        fetchStats[t_info.thread->threadId()]->numInsts++;
+    }
+    // increment thread level numOps fetched count
+    fetchStats[t_info.thread->threadId()]->numOps++;
+}
+
+void
+BaseSimpleCPU::countCommitInst()
+{
+    SimpleExecContext& t_info = *threadInfo[curThread];
+
+    if (!curStaticInst->isMicroop() || curStaticInst->isLastMicroop()) {
+        // increment thread level and core level numInsts count
+        commitStats[t_info.thread->threadId()]->numInsts++;
+        baseStats.numInsts++;
+    }
+    // increment thread level and core level numOps count
+    commitStats[t_info.thread->threadId()]->numOps++;
+    baseStats.numOps++;
 }
 
 Counter
@@ -376,6 +402,11 @@
         if (predict_taken)
             ++t_info.execContextStats.numPredictedBranches;
     }
+
+    // increment the fetch instruction stat counters
+    if (curStaticInst) {
+        countFetchInst();
+    }
 }
 
 void
@@ -443,6 +474,9 @@
         ->committedInstType[curStaticInst->opClass()]++;
     commitStats[t_info.thread->threadId()]->updateComCtrlStats(curStaticInst);
 
+    /* increment the committed numInsts and numOps stats */
+    countCommitInst();
+
     if (FullSystem)
         traceFunctions(instAddr);
 
diff --git a/src/cpu/simple/base.hh b/src/cpu/simple/base.hh
index df5290c..46a25a0 100644
--- a/src/cpu/simple/base.hh
+++ b/src/cpu/simple/base.hh
@@ -182,6 +182,8 @@
     }
 
     void countInst();
+    void countFetchInst();
+    void countCommitInst();
     Counter totalInsts() const override;
     Counter totalOps() const override;
 
diff --git a/src/cpu/simple/exec_context.hh b/src/cpu/simple/exec_context.hh
index 42d6181..c0927fc 100644
--- a/src/cpu/simple/exec_context.hh
+++ b/src/cpu/simple/exec_context.hh
@@ -86,10 +86,6 @@
             : statistics::Group(cpu,
                            csprintf("exec_context.thread_%i",
                                     thread->threadId()).c_str()),
-              ADD_STAT(numInsts, statistics::units::Count::get(),
-                       "Number of instructions committed"),
-              ADD_STAT(numOps, statistics::units::Count::get(),
-                       "Number of ops (including micro ops) committed"),
               ADD_STAT(numMatAluAccesses, statistics::units::Count::get(),
                        "Number of matrix alu accesses"),
               ADD_STAT(numCallsReturns, statistics::units::Count::get(),
@@ -140,10 +136,6 @@
                 .prereq(numBranchMispred);
         }
 
-        // Number of simulated instructions
-        statistics::Scalar numInsts;
-        statistics::Scalar numOps;
-
         // Number of matrix alu accesses
         statistics::Scalar numMatAluAccesses;