diff --git a/cpu.cpp b/cpu.cpp
index 103dc01..2cdc883 100644
--- a/cpu.cpp
+++ b/cpu.cpp
@@ -18,6 +18,12 @@
 
 #define IS_0(x, wm) ((wm) == wm_byte ? ((x) & 0xff) == 0 : (x) == 0)
 
+// see https://retrocomputing.stackexchange.com/questions/6960/what-was-the-clock-speed-and-ips-for-the-original-pdp-11
+constexpr const double pdp11_clock_cycle = 150;  // ns, for the 11/70
+constexpr const double pdp11_MHz = 1000.0 / pdp11_clock_cycle;
+constexpr const double pdp11_avg_cycles_per_instruction = (1 + 5) / 2.0;
+constexpr const double pdp11_estimated_mips = pdp11_MHz / pdp11_avg_cycles_per_instruction;
+
 cpu::cpu(bus *const b, std::atomic_uint32_t *const event) : b(b), event(event)
 {
 	reset();
@@ -97,15 +103,15 @@ std::tuple<double, double, uint64_t, uint32_t, double> cpu::get_mips_rel_speed(c
 
         double mips = t_diff ? instr_count / double(t_diff) : 0;
 
-        // see https://retrocomputing.stackexchange.com/questions/6960/what-was-the-clock-speed-and-ips-for-the-original-pdp-11
-        constexpr double pdp11_clock_cycle = 150;  // ns, for the 11/70
-        constexpr double pdp11_MHz = 1000.0 / pdp11_clock_cycle;
-        constexpr double pdp11_avg_cycles_per_instruction = (1 + 5) / 2.0;
-        constexpr double pdp11_estimated_mips = pdp11_MHz / pdp11_avg_cycles_per_instruction;
-
 	return { mips, mips * 100 / pdp11_estimated_mips, instr_count, t_diff, wait_time };
 }
 
+uint32_t cpu::get_effective_run_time(const uint64_t instruction_count) const
+{
+	// division is to go from ns to ms
+	return instruction_count * pdp11_clock_cycle / 1000000l;
+}
+
 void cpu::add_to_stack_trace(const uint16_t p)
 {
 	auto da = disassemble(p);
diff --git a/cpu.h b/cpu.h
index 9c5c40f..56e37eb 100644
--- a/cpu.h
+++ b/cpu.h
@@ -133,6 +133,8 @@ public:
 	uint64_t get_instructions_executed_count() const;
 	uint64_t get_wait_time() const { return wait_time; }
 	std::tuple<double, double, uint64_t, uint32_t, double> get_mips_rel_speed(const std::optional<uint64_t> & instruction_count, const std::optional<uint64_t> & t_diff_1s) const;
+	// how many ms would've really passed when executing `instruction_count` instructions
+	uint32_t get_effective_run_time(const uint64_t instruction_count) const;
 
 	bool get_debug() const { return debug_mode; }
 	void set_debug(const bool d) { debug_mode = d; stacktrace.clear(); }
diff --git a/kw11-l.cpp b/kw11-l.cpp
index 2d23054..6114ee7 100644
--- a/kw11-l.cpp
+++ b/kw11-l.cpp
@@ -68,22 +68,26 @@ void kw11_l::operator()()
 
 	TRACE("Starting KW11-L thread");
 
+	uint64_t prev_cycle_count = b->getCpu()->get_instructions_executed_count();
+
 	while(!stop_flag) {
 		if (*cnsl->get_running_flag()) {
-			set_lf_crs_b7();
- 
-			if (get_lf_crs() & 64)
-				b->getCpu()->queue_interrupt(6, 0100);
+			myusleep(1000000 / 100);  // 100 Hz
 
-			// TODO: depending on cpu cycles processed
-#if defined(ESP32)
-			myusleep(1000000 / 20);  // 50ms
-#else
-			myusleep(1000000 / 50);  // 20ms
-#endif
+			uint64_t current_cycle_count = b->getCpu()->get_instructions_executed_count();
+			uint32_t took_ms = b->getCpu()->get_effective_run_time(current_cycle_count - prev_cycle_count);
+
+			if (took_ms >= 1000 / 50) {
+				set_lf_crs_b7();
+
+				if (get_lf_crs() & 64)
+					b->getCpu()->queue_interrupt(6, 0100);
+
+				prev_cycle_count = current_cycle_count;
+			}
 		}
 		else {
-			myusleep(1000000 / 10);  // 100ms
+			myusleep(1000000 / 10);  // 10 Hz
 		}
 	}