diff --git a/src/cpu/base.cc b/src/cpu/base.cc
index d2c0a78d44..1d293397e5 100644
--- a/src/cpu/base.cc
+++ b/src/cpu/base.cc
@@ -191,6 +191,11 @@ BaseCPU::BaseCPU(const Params &p, bool is_checker)
     modelResetPort.onChange([this](const bool &new_val) {
         setReset(new_val);
     });
+    // create a stat group object for each thread on this core
+    fetchStats.reserve(numThreads);
+    for (int i = 0; i < numThreads; i++) {
+        fetchStats.emplace_back(new FetchCPUStats(this, i));
+    }
 }
 
 void
@@ -827,4 +832,18 @@ BaseCPU::GlobalStats::GlobalStats(statistics::Group *parent)
     hostOpRate = simOps / hostSeconds;
 }
 
+BaseCPU::
+FetchCPUStats::FetchCPUStats(statistics::Group *parent, int thread_id)
+    : statistics::Group(parent, csprintf("fetchStats%i", thread_id).c_str()),
+    ADD_STAT(numBranches, statistics::units::Count::get(),
+             "Number of branches fetched"),
+    ADD_STAT(numFetchSuspends, statistics::units::Count::get(),
+             "Number of times Execute suspended instruction fetching")
+
+{
+    numBranches
+        .prereq(numBranches);
+
+}
+
 } // namespace gem5
diff --git a/src/cpu/base.hh b/src/cpu/base.hh
index 084d9b9305..d6e5d38838 100644
--- a/src/cpu/base.hh
+++ b/src/cpu/base.hh
@@ -43,6 +43,7 @@
 #define __CPU_BASE_HH__
 
 #include <vector>
+#include <memory>
 
 #include "arch/generic/interrupts.hh"
 #include "base/statistics.hh"
@@ -676,6 +677,21 @@ class BaseCPU : public ClockedObject
     const Cycles pwrGatingLatency;
     const bool powerGatingOnIdle;
     EventFunctionWrapper enterPwrGatingEvent;
+
+  public:
+    struct FetchCPUStats : public statistics::Group
+    {
+        FetchCPUStats(statistics::Group *parent, int thread_id);
+
+        /* Total number of branches fetched */
+        statistics::Scalar numBranches;
+
+        /* Number of times fetch was asked to suspend by Execute */
+        statistics::Scalar numFetchSuspends;
+
+    };
+
+    std::vector<std::unique_ptr<FetchCPUStats>> fetchStats;
 };
 
 } // namespace gem5
diff --git a/src/cpu/minor/execute.cc b/src/cpu/minor/execute.cc
index 5eaaf5804e..323ae2982b 100644
--- a/src/cpu/minor/execute.cc
+++ b/src/cpu/minor/execute.cc
@@ -1054,7 +1054,7 @@ Execute::commitInst(MinorDynInstPtr inst, bool early_memory_issue,
             DPRINTF(MinorInterrupt, "Suspending thread: %d from Execute"
                 " inst: %s\n", thread_id, *inst);
 
-            cpu.stats.numFetchSuspends++;
+            cpu.fetchStats[thread_id]->numFetchSuspends++;
 
             updateBranchData(thread_id, BranchData::SuspendThread, inst,
                 resume_pc, branch);
diff --git a/src/cpu/minor/stats.cc b/src/cpu/minor/stats.cc
index 64d4c475e0..e9ca562c16 100644
--- a/src/cpu/minor/stats.cc
+++ b/src/cpu/minor/stats.cc
@@ -52,8 +52,6 @@ MinorStats::MinorStats(BaseCPU *base_cpu)
     ADD_STAT(numDiscardedOps, statistics::units::Count::get(),
              "Number of ops (including micro ops) which were discarded before "
              "commit"),
-    ADD_STAT(numFetchSuspends, statistics::units::Count::get(),
-             "Number of times Execute suspended instruction fetching"),
     ADD_STAT(quiesceCycles, statistics::units::Cycle::get(),
              "Total number of cycles that CPU has spent quiesced or waiting "
              "for an interrupt"),
diff --git a/src/cpu/minor/stats.hh b/src/cpu/minor/stats.hh
index 1ab81f4407..524d20f85d 100644
--- a/src/cpu/minor/stats.hh
+++ b/src/cpu/minor/stats.hh
@@ -68,9 +68,6 @@ struct MinorStats : public statistics::Group
     /** Number of ops discarded before committing */
     statistics::Scalar numDiscardedOps;
 
-    /** Number of times fetch was asked to suspend by Execute */
-    statistics::Scalar numFetchSuspends;
-
     /** Number of cycles in quiescent state */
     statistics::Scalar quiesceCycles;
 
diff --git a/src/cpu/simple/base.cc b/src/cpu/simple/base.cc
index 768f63ede5..b2a11fd84b 100644
--- a/src/cpu/simple/base.cc
+++ b/src/cpu/simple/base.cc
@@ -396,7 +396,7 @@ BaseSimpleCPU::postExecute()
     }
 
     if (curStaticInst->isControl()) {
-        ++t_info.execContextStats.numBranches;
+        ++fetchStats[t_info.thread->threadId()]->numBranches;
     }
 
     /* Power model statistics */
diff --git a/src/cpu/simple/exec_context.hh b/src/cpu/simple/exec_context.hh
index 0f20763f28..d4bb017481 100644
--- a/src/cpu/simple/exec_context.hh
+++ b/src/cpu/simple/exec_context.hh
@@ -152,8 +152,6 @@ class SimpleExecContext : public ExecContext
                        "ICache total stall cycles"),
               ADD_STAT(dcacheStallCycles, statistics::units::Cycle::get(),
                        "DCache total stall cycles"),
-              ADD_STAT(numBranches, statistics::units::Count::get(),
-                       "Number of branches fetched"),
               ADD_STAT(numPredictedBranches, statistics::units::Count::get(),
                        "Number of branches predicted as taken"),
               ADD_STAT(numBranchMispred, statistics::units::Count::get(),
@@ -203,9 +201,6 @@ class SimpleExecContext : public ExecContext
             numIdleCycles = idleFraction * cpu->baseStats.numCycles;
             numBusyCycles = notIdleFraction * cpu->baseStats.numCycles;
 
-            numBranches
-                .prereq(numBranches);
-
             numPredictedBranches
                 .prereq(numPredictedBranches);
 
@@ -297,8 +292,6 @@ class SimpleExecContext : public ExecContext
         statistics::Scalar dcacheStallCycles;
 
         /// @{
-        /// Total number of branches fetched
-        statistics::Scalar numBranches;
         /// Number of branches predicted as taken
         statistics::Scalar numPredictedBranches;
         /// Number of misprediced branches