diff --git a/corrscope/corrscope.py b/corrscope/corrscope.py
index f2b1f21..74b1a3a 100644
--- a/corrscope/corrscope.py
+++ b/corrscope/corrscope.py
@@ -1,12 +1,15 @@
 # -*- coding: utf-8 -*-
 import os.path
+import threading
 import time
+from concurrent.futures import ProcessPoolExecutor, Future
 from contextlib import ExitStack, contextmanager
 from enum import unique
 from fractions import Fraction
 from pathlib import Path
-from typing import Iterator, Optional, List, Callable
-from multiprocessing.pool import Pool
+from queue import Queue
+from threading import Thread
+from typing import Iterator, Optional, List, Callable, Tuple
 
 import attr
 
@@ -15,12 +18,17 @@ from corrscope.channel import Channel, ChannelConfig, DefaultLabel
 from corrscope.config import KeywordAttrs, DumpEnumAsStr, CorrError, with_units
 from corrscope.layout import LayoutConfig
 from corrscope.outputs import FFmpegOutputConfig, IOutputConfig
-from corrscope.renderer import Renderer, RendererConfig, RendererFrontend, RenderInput
+from corrscope.renderer import (
+    Renderer,
+    RendererConfig,
+    RendererFrontend,
+    RenderInput,
+    ByteBuffer,
+)
 from corrscope.triggers import (
     CorrelationTriggerConfig,
     PerFrameCache,
     SpectrumConfig,
-    MainTrigger,
 )
 from corrscope.util import pushd, coalesce
 from corrscope.wave import Wave, Flatten, FlattenOrStr
@@ -351,9 +359,132 @@ class CorrScope:
                             thread_shared.end_frame = frame + 1
                             break
 
+        # Multiprocess
+        def play_parallel():
+            ncores = len(os.sched_getaffinity(0))
+
+            abort_from_thread = threading.Event()
+            # self.arg.is_aborted() from GUI, abort_from_thread.is_set() from thread
+            is_aborted = lambda: self.arg.is_aborted() or abort_from_thread.is_set()
+
+            # Same size as ProcessPoolExecutor, so threads won't starve if they all
+            # finish a job at the same time.
+            render_to_output: "Queue[Tuple[int, Future[ByteBuffer]] | None]" = Queue(
+                ncores
+            )
+
+            def worker_create_renderer(renderer: RendererFrontend):
+                global WORKER_RENDERER
+                # TODO del self.renderer and recreate Renderer if it can't be pickled?
+                WORKER_RENDERER = renderer
+
+            # TODO https://stackoverflow.com/questions/2829329/catch-a-threads-exception-in-the-caller-thread
+            def render_thread():
+                end_frame = thread_shared.end_frame
+                prev = -1
+
+                # TODO gather trigger points from triggering threads
+                # For each frame, render each wave
+                for frame in range(begin_frame, end_frame):
+                    if is_aborted():
+                        # Only count output-displayed frames, not rendered.
+                        # # Used for FPS calculation
+                        # thread_shared.end_frame = frame
+
+                        for output in self.outputs:
+                            output.terminate()
+                        break
+
+                    time_seconds = frame / fps
+                    should_render = (frame - begin_frame) % render_subfps == ahead
+
+                    rounded = int(time_seconds)
+                    if PRINT_TIMESTAMP and rounded != prev:
+                        self.arg.progress(rounded)
+                        prev = rounded
+
+                    render_inputs = []
+                    trigger_samples = []
+                    # Get render-data from each wave.
+                    for render_wave, channel in zip(self.render_waves, self.channels):
+                        sample = round(render_wave.smp_s * time_seconds)
+
+                        # Get trigger.
+                        if not_benchmarking or benchmark_mode == BenchmarkMode.TRIGGER:
+                            cache = PerFrameCache()
+
+                            result = channel.trigger.get_trigger(sample, cache)
+                            trigger_sample = result.result
+                            freq_estimate = result.freq_estimate
+
+                        else:
+                            trigger_sample = sample
+                            freq_estimate = 0
+
+                        # Get render data.
+                        if should_render:
+                            trigger_samples.append(trigger_sample)
+                            data = channel.get_render_around(trigger_sample)
+                            render_inputs.append(RenderInput(data, freq_estimate))
+
+                    if not should_render:
+                        continue
+
+                    # blocking
+                    render_to_output.put(
+                        (
+                            frame,
+                            pool.submit(
+                                worker_render_frame, render_inputs, trigger_samples
+                            ),
+                        )
+                    )
+
+                render_to_output.put(None)
+
+            def worker_render_frame(
+                render_inputs: List[RenderInput], trigger_samples: List[int]
+            ) -> ByteBuffer:
+                global WORKER_RENDERER
+                renderer = WORKER_RENDERER
+                renderer.update_main_lines(render_inputs, trigger_samples)
+                frame_data = renderer.get_frame()
+                return frame_data
+
+            def output_thread():
+                while True:
+                    msg = render_to_output.get()  # blocking
+                    if msg is None:
+                        break
+                    frame, render_future = msg
+                    frame_data: ByteBuffer = render_future.result()
+
+                    if not_benchmarking or benchmark_mode == BenchmarkMode.OUTPUT:
+                        # Output frame
+                        for output in self.outputs:
+                            if output.write_frame(frame_data) is outputs_.Stop:
+                                abort_from_thread.set()
+                                break
+                        if is_aborted():
+                            # Outputting frame happens after most computation finished.
+                            thread_shared.end_frame = frame + 1
+                            break
+
+            with ProcessPoolExecutor(
+                ncores, initializer=worker_create_renderer, initargs=(renderer,)
+            ) as pool:
+                render_handle = Thread(target=render_thread, name="render_thread")
+                output_handle = Thread(target=output_thread, name="output_thread")
+
+                render_handle.start()
+                output_handle.start()
+
+                render_handle.join()
+                output_handle.join()
+
         with self._load_outputs():
             if self.arg.parallel:
-                pass
+                play_parallel()
             else:
                 play_impl()