diff --git a/cconv.c b/cconv.c
index 0de2f5c..0fca764 100644
--- a/cconv.c
+++ b/cconv.c
@@ -219,7 +219,6 @@
 	o->unified_rw_rep = le32_to_cpu(top->unified_rw_rep);
 	o->gtod_reduce = le32_to_cpu(top->gtod_reduce);
 	o->gtod_cpu = le32_to_cpu(top->gtod_cpu);
-	o->gtod_offload = le32_to_cpu(top->gtod_offload);
 	o->clocksource = le32_to_cpu(top->clocksource);
 	o->no_stall = le32_to_cpu(top->no_stall);
 	o->trim_percentage = le32_to_cpu(top->trim_percentage);
@@ -381,7 +380,6 @@
 	top->unified_rw_rep = cpu_to_le32(o->unified_rw_rep);
 	top->gtod_reduce = cpu_to_le32(o->gtod_reduce);
 	top->gtod_cpu = cpu_to_le32(o->gtod_cpu);
-	top->gtod_offload = cpu_to_le32(o->gtod_offload);
 	top->clocksource = cpu_to_le32(o->clocksource);
 	top->no_stall = cpu_to_le32(o->no_stall);
 	top->trim_percentage = cpu_to_le32(o->trim_percentage);
diff --git a/gettime-thread.c b/gettime-thread.c
index 95f35c3..73632d0 100644
--- a/gettime-thread.c
+++ b/gettime-thread.c
@@ -8,11 +8,14 @@
 
 struct timeval *fio_tv = NULL;
 int fio_gtod_offload = 0;
-int fio_gtod_cpu = -1;
+static os_cpu_mask_t fio_gtod_cpumask;
 static pthread_t gtod_thread;
 
 void fio_gtod_init(void)
 {
+	if (fio_tv)
+		return;
+
 	fio_tv = smalloc(sizeof(struct timeval));
 	if (!fio_tv)
 		log_err("fio: smalloc pool exhausted\n");
@@ -31,10 +34,16 @@
 	}
 }
 
+struct gtod_cpu_data {
+	struct fio_mutex *mutex;
+	unsigned int cpu;
+};
+
 static void *gtod_thread_main(void *data)
 {
 	struct fio_mutex *mutex = data;
 
+	fio_setaffinity(gettid(), fio_gtod_cpumask);
 	fio_mutex_up(mutex);
 
 	/*
@@ -84,4 +93,7 @@
 	return ret;
 }
 
-
+void fio_gtod_set_cpu(unsigned int cpu)
+{
+	fio_cpu_set(&fio_gtod_cpumask, cpu);
+}
diff --git a/gettime.h b/gettime.h
index f541228..b775ef3 100644
--- a/gettime.h
+++ b/gettime.h
@@ -38,4 +38,6 @@
 	return 1;
 }
 
+extern void fio_gtod_set_cpu(unsigned int cpu);
+
 #endif
diff --git a/init.c b/init.c
index f606087..427768c 100644
--- a/init.c
+++ b/init.c
@@ -758,6 +758,12 @@
 		ret = 1;
 	}
 
+	if (fio_option_is_set(o, gtod_cpu)) {
+		fio_gtod_init();
+		fio_gtod_set_cpu(o->gtod_cpu);
+		fio_gtod_offload = 1;
+	}
+
 	return ret;
 }
 
@@ -2385,12 +2391,6 @@
 		return 0;
 	}
 
-	if (def_thread.o.gtod_offload) {
-		fio_gtod_init();
-		fio_gtod_offload = 1;
-		fio_gtod_cpu = def_thread.o.gtod_cpu;
-	}
-
 	if (output_format == FIO_OUTPUT_NORMAL)
 		log_info("%s\n", fio_version_string);
 
diff --git a/options.c b/options.c
index a01b2d9..80a7047 100644
--- a/options.c
+++ b/options.c
@@ -1079,16 +1079,6 @@
 	return 0;
 }
 
-static int str_gtod_cpu_cb(void *data, long long *il)
-{
-	struct thread_data *td = data;
-	int val = *il;
-
-	td->o.gtod_cpu = val;
-	td->o.gtod_offload = 1;
-	return 0;
-}
-
 static int str_size_cb(void *data, unsigned long long *__val)
 {
 	struct thread_data *td = data;
@@ -3404,7 +3394,7 @@
 		.name	= "gtod_cpu",
 		.lname	= "Dedicated gettimeofday() CPU",
 		.type	= FIO_OPT_INT,
-		.cb	= str_gtod_cpu_cb,
+		.off1	= td_var_offset(gtod_cpu),
 		.help	= "Set up dedicated gettimeofday() thread on this CPU",
 		.verify	= gtod_cpu_verify,
 		.category = FIO_OPT_C_GENERAL,
diff --git a/server.h b/server.h
index cd3f999..dc5a69e 100644
--- a/server.h
+++ b/server.h
@@ -38,7 +38,7 @@
 };
 
 enum {
-	FIO_SERVER_VER			= 41,
+	FIO_SERVER_VER			= 42,
 
 	FIO_SERVER_MAX_FRAGMENT_PDU	= 1024,
 	FIO_SERVER_MAX_CMD_MB		= 2048,
diff --git a/thread_options.h b/thread_options.h
index 530dd9a..611f8e7 100644
--- a/thread_options.h
+++ b/thread_options.h
@@ -195,7 +195,6 @@
 	unsigned int unified_rw_rep;
 	unsigned int gtod_reduce;
 	unsigned int gtod_cpu;
-	unsigned int gtod_offload;
 	enum fio_cs clocksource;
 	unsigned int no_stall;
 	unsigned int trim_percentage;
@@ -419,7 +418,6 @@
 	uint32_t unified_rw_rep;
 	uint32_t gtod_reduce;
 	uint32_t gtod_cpu;
-	uint32_t gtod_offload;
 	uint32_t clocksource;
 	uint32_t no_stall;
 	uint32_t trim_percentage;
@@ -428,6 +426,7 @@
 	uint64_t trim_backlog;
 	uint32_t clat_percentiles;
 	uint32_t percentile_precision;
+	uint32_t pad2;
 	fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
 
 	uint8_t read_iolog_file[FIO_TOP_STR_MAX];
@@ -480,10 +479,10 @@
 	uint64_t number_ios;
 
 	uint32_t sync_file_range;
-	uint32_t pad2;
 
 	uint64_t latency_target;
 	uint64_t latency_window;
+	uint32_t pad3;
 	fio_fp64_t latency_percentile;
 } __attribute__((packed));