diff --git a/src/backend/common/interfaces/IWorker.h b/src/backend/common/interfaces/IWorker.h
index dcc925dcd..979d998ff 100644
--- a/src/backend/common/interfaces/IWorker.h
+++ b/src/backend/common/interfaces/IWorker.h
@@ -35,6 +35,7 @@ namespace xmrig {
 
 class VirtualMemory;
 class Job;
+class Config;
 
 
 class IWorker
@@ -48,7 +49,7 @@ public:
     virtual size_t intensity() const                          = 0;
     virtual uint64_t rawHashes() const                        = 0;
     virtual void getHashrateData(uint64_t&, uint64_t&) const  = 0;
-    virtual void start()                                      = 0;
+    virtual void start(Config*)                               = 0;
     virtual void jobEarlyNotification(const Job&)             = 0;
 
 #   ifdef XMRIG_FEATURE_BENCHMARK
diff --git a/src/backend/cpu/CpuBackend.cpp b/src/backend/cpu/CpuBackend.cpp
index ecce64fe1..a847567ae 100644
--- a/src/backend/cpu/CpuBackend.cpp
+++ b/src/backend/cpu/CpuBackend.cpp
@@ -390,7 +390,7 @@ void xmrig::CpuBackend::start(IWorker *worker, bool ready)
     mutex.unlock();
 
     if (ready) {
-        worker->start();
+        worker->start(d_ptr->controller->config());
     }
 }
 
diff --git a/src/backend/cpu/CpuWorker.cpp b/src/backend/cpu/CpuWorker.cpp
index fd1a5932b..4c7781c3c 100644
--- a/src/backend/cpu/CpuWorker.cpp
+++ b/src/backend/cpu/CpuWorker.cpp
@@ -30,6 +30,7 @@
 
 #include "backend/cpu/CpuWorker.h"
 #include "base/tools/Chrono.h"
+#include "core/config/Config.h"
 #include "core/Miner.h"
 #include "crypto/cn/CnCtx.h"
 #include "crypto/cn/CryptoNight_test.h"
@@ -198,7 +199,7 @@ bool xmrig::CpuWorker<N>::selfTest()
 
 
 template<size_t N>
-void xmrig::CpuWorker<N>::start()
+void xmrig::CpuWorker<N>::start(xmrig::Config* config)
 {
     while (Nonce::sequence(Nonce::CPU) > 0) {
         if (Nonce::isPaused()) {
@@ -219,6 +220,10 @@ void xmrig::CpuWorker<N>::start()
         alignas(16) uint64_t tempHash[8] = {};
 #       endif
 
+#       ifdef XMRIG_FEATURE_BENCHMARK
+        const size_t benchThreads = config->cpu().threads().get(m_job.currentJob().algorithm()).count();
+#       endif
+
         while (!Nonce::isOutdated(Nonce::CPU, m_job.sequence())) {
             const Job &job = m_job.currentJob();
 
@@ -231,6 +236,29 @@ void xmrig::CpuWorker<N>::start()
                 current_job_nonces[i] = *m_job.nonce(i);
             }
 
+#           ifdef XMRIG_FEATURE_BENCHMARK
+            if (m_benchSize) {
+                bool done = true;
+                for (size_t i = 0; i < N; ++i) {
+                    if (current_job_nonces[i] < m_benchSize) {
+                        done = false;
+                        break;
+                    }
+                }
+
+                // Stop benchmark when all hashes have been counted
+                if (done) {
+                    m_benchDoneTime = Chrono::steadyMSecs();
+                    return;
+                }
+
+                // Make each hash dependent on the previous one in single thread benchmark to prevent cheating with multiple threads
+                if (benchThreads == 1) {
+                    *(uint64_t*)(m_job.blob()) ^= m_benchData;
+                }
+            }
+#           endif
+
             bool valid = true;
 
 #           ifdef XMRIG_ALGO_RANDOMX
@@ -274,10 +302,6 @@ void xmrig::CpuWorker<N>::start()
                         if (current_job_nonces[i] < m_benchSize) {
                             m_benchData ^= value;
                         }
-                        else {
-                            m_benchDoneTime = Chrono::steadyMSecs();
-                            return;
-                        }
                     }
                     else
 #                   endif
diff --git a/src/backend/cpu/CpuWorker.h b/src/backend/cpu/CpuWorker.h
index 087156c24..a43837812 100644
--- a/src/backend/cpu/CpuWorker.h
+++ b/src/backend/cpu/CpuWorker.h
@@ -43,6 +43,7 @@ namespace xmrig {
 
 
 class RxVm;
+class Config;
 
 
 template<size_t N>
@@ -56,7 +57,7 @@ public:
 
 protected:
     bool selfTest() override;
-    void start() override;
+    void start(Config*) override;
 
     inline const VirtualMemory *memory() const override { return m_memory; }
     inline size_t intensity() const override            { return N; }
diff --git a/src/backend/cuda/CudaBackend.cpp b/src/backend/cuda/CudaBackend.cpp
index ab5b22e53..f347aacfd 100644
--- a/src/backend/cuda/CudaBackend.cpp
+++ b/src/backend/cuda/CudaBackend.cpp
@@ -481,7 +481,7 @@ void xmrig::CudaBackend::start(IWorker *worker, bool ready)
     mutex.unlock();
 
     if (ready) {
-        worker->start();
+        worker->start(d_ptr->controller->config());
     }
 }
 
diff --git a/src/backend/cuda/CudaWorker.cpp b/src/backend/cuda/CudaWorker.cpp
index 7bb8844c1..9e34e8500 100644
--- a/src/backend/cuda/CudaWorker.cpp
+++ b/src/backend/cuda/CudaWorker.cpp
@@ -145,7 +145,7 @@ size_t xmrig::CudaWorker::intensity() const
 }
 
 
-void xmrig::CudaWorker::start()
+void xmrig::CudaWorker::start(xmrig::Config*)
 {
     while (Nonce::sequence(Nonce::CUDA) > 0) {
         if (!isReady()) {
diff --git a/src/backend/cuda/CudaWorker.h b/src/backend/cuda/CudaWorker.h
index e82e34254..9ccdc519b 100644
--- a/src/backend/cuda/CudaWorker.h
+++ b/src/backend/cuda/CudaWorker.h
@@ -39,6 +39,7 @@ namespace xmrig {
 
 
 class ICudaRunner;
+class Config;
 
 
 class CudaWorker : public Worker
@@ -58,7 +59,7 @@ public:
 protected:
     bool selfTest() override;
     size_t intensity() const override;
-    void start() override;
+    void start(Config*) override;
 
 private:
     bool consumeJob();
diff --git a/src/backend/opencl/OclBackend.cpp b/src/backend/opencl/OclBackend.cpp
index 7b99700b5..04bc2cdb5 100644
--- a/src/backend/opencl/OclBackend.cpp
+++ b/src/backend/opencl/OclBackend.cpp
@@ -463,7 +463,7 @@ void xmrig::OclBackend::start(IWorker *worker, bool ready)
     mutex.unlock();
 
     if (ready) {
-        worker->start();
+        worker->start(d_ptr->controller->config());
     }
 }
 
diff --git a/src/backend/opencl/OclWorker.cpp b/src/backend/opencl/OclWorker.cpp
index ac493ab55..ca6cfaccf 100644
--- a/src/backend/opencl/OclWorker.cpp
+++ b/src/backend/opencl/OclWorker.cpp
@@ -163,7 +163,7 @@ size_t xmrig::OclWorker::intensity() const
 }
 
 
-void xmrig::OclWorker::start()
+void xmrig::OclWorker::start(xmrig::Config*)
 {
     cl_uint results[0x100];
 
diff --git a/src/backend/opencl/OclWorker.h b/src/backend/opencl/OclWorker.h
index 403a0765d..d43ccf561 100644
--- a/src/backend/opencl/OclWorker.h
+++ b/src/backend/opencl/OclWorker.h
@@ -40,6 +40,7 @@ namespace xmrig {
 
 class IOclRunner;
 class Job;
+class Config;
 
 
 class OclWorker : public Worker
@@ -59,7 +60,7 @@ public:
 protected:
     bool selfTest() override;
     size_t intensity() const override;
-    void start() override;
+    void start(Config*) override;
 
 private:
     bool consumeJob();
diff --git a/src/crypto/randomx/jit_compiler_x86.cpp b/src/crypto/randomx/jit_compiler_x86.cpp
index 88c1ce4ef..182de1fa6 100644
--- a/src/crypto/randomx/jit_compiler_x86.cpp
+++ b/src/crypto/randomx/jit_compiler_x86.cpp
@@ -297,7 +297,7 @@ namespace randomx {
 	}
 
 	void JitCompilerX86::generateProgramPrologue(Program& prog, ProgramConfiguration& pcfg) {
-		codePos = ((uint8_t*)randomx_program_prologue_first_load) - ((uint8_t*)randomx_program_prologue);
+		codePos = ADDR(randomx_program_prologue_first_load) - ADDR(randomx_program_prologue);
 		code[codePos + 2] = 0xc0 + pcfg.readReg0;
 		code[codePos + 5] = 0xc0 + pcfg.readReg1;
 		*(uint32_t*)(code + codePos + 10) = RandomX_CurrentConfig.ScratchpadL3Mask64_Calculated;