The short answer
If your PC has any hardware encoder (an NVIDIA card, an AMD card, or Intel graphics with Quick Sync), use it, and make it lighter than OBS makes it by default:
- Encoder: NVENC, AMF/AMD, QuickSync, or VAAPI on Linux. Not x264.
- Output resolution 1280x720 (or 1280x720 from a 1920x1080 canvas), 30 fps. 60 fps doubles every per-frame cost.
- Color format NV12. It is the default, and the hardware encoders only stay on their fast path with it.
- On NVENC, step the preset down from P5 and turn off Look-ahead and the two-pass Multipass mode. OBS's defaults are tuned for quality, not for a weak machine (details below).
- On Windows, Game Capture with "Limit capture framerate" ticked.
If you only have the CPU (x264):
- CPU Usage Preset
superfastorultrafastinstead of the defaultveryfast. - 1280x720 or lower, 30 fps.
- Keep NV12. Anything else makes OBS convert the picture on the CPU before x264 even starts.
On our test laptop (a 15 W Intel CPU, no graphics card), x264 at OBS's default veryfast used 34% of the CPU at 720p30, superfast 23% and ultrafast 18%. The laptop's Intel graphics encoder used 12%. None of them skipped a frame. The full numbers are in the test section.
Everything else in this guide is about why, and about the settings people change that do less than they think.
First, find out what is actually overloaded
OBS does three separate jobs every frame, and each has its own counter. Open Docks → Stats while you stream or record and look at these three lines:
| Stats line | What it counts | What it points to |
|---|---|---|
| Frames missed due to rendering lag | Frames OBS could not draw in time | Your GPU, or a slow source in the scene |
| Skipped frames due to encoding lag | Frames the encoder could not accept in time | The encoder (your CPU for x264) |
| Dropped frames (Network) | Frames thrown away because the upload fell behind | Your connection, not your PC |
The Stats dock turns a line yellow above 1% and red above 5% (OBSBasicStats.cpp).
"Encoding overloaded" means only one of them. The yellow "Encoding overloaded! Consider turning down video settings..." message in the status bar is triggered by the skipped-frames counter and nothing else. Once a second OBS checks whether more than 10 frames were skipped since the last check and whether at least 0.1% of all frames so far were skipped (OBSBasicStatusBar.cpp). It does not look at CPU usage at all, which is why you can see it with the CPU at 40%. Rendering lag has no warning in the status bar, so a GPU problem can go unnoticed until you open the Stats dock.
If it says rendering lag
OBS draws the whole scene once per frame on a fixed clock. When a frame takes longer than its slot, OBS does not slow down. It counts the slots it missed and sends the one frame it did draw to the encoder that many times (obs-video.c). So rendering lag shows up as repeated frames in the stream, and with x264 the encoder still has to encode every copy, while the PC is already behind.
What helps:
- Lower the frame rate first. Every per-frame step (drawing, scaling, color conversion, encoding) runs once per frame, so 30 instead of 60 fps halves all of it.
- Find the expensive source. Sources are updated inside the same render loop. On Windows, Window Capture in BitBlt mode copies the window with GDI from inside that loop (window-capture.c), so one slow window can make every frame late. Hide sources one at a time and watch the counter.
- Limit Game Capture's frame rate. "Limit capture framerate" is off by default, and then the game copies frames for OBS at up to twice OBS's frame rate. The source has a comment about a game running at 900 FPS that "will dramatically reduce performance" (game-capture.c). Tick it.
- Close Studio Mode and the Multiview. In Studio Mode the preview draws a second scene every frame (OBSBasic_Preview.cpp). The Multiview draws every scene in its grid again (Multiview.cpp).
Hiding the preview helps less than you'd think. OBS renders the scene every frame whether the preview is visible or not (obs-video.c). Disabling the preview skips only drawing the finished image into the OBS window (obs-display.c).
On our test laptop, Studio Mode made no measurable difference with a one-source scene: 1.2 ms per frame with it on or off. The second render costs what your scene costs, so it adds up with many sources, browser sources and filters.
If it says encoding lag
With x264, encoding happens on one dedicated thread that takes frames from a small queue of six (obs.c). When x264 takes longer per frame than the frame rate allows, the queue fills and new frames are skipped (video-io.c). That makes the fix simple: make each frame cheaper to encode, or hand the work to a hardware encoder.
x264: the preset is the setting
OBS starts from x264's own preset (default veryfast), then puts its bitrate and size settings on top (obs-x264.c). The preset decides almost everything about how hard x264 works, so moving from veryfast to superfast or ultrafast is the biggest single change you can make on a weak CPU. In Simple output mode the list is ultrafast, superfast, veryfast, faster and fast (OBSBasicSettings.cpp).
Measured at 720p30 on our test laptop: faster 51% CPU, veryfast 34%, superfast 23%, ultrafast 18%. Each step down took a quarter to a third off. At 1080p30, veryfast needed 62% and started skipping frames (0.2%).
Lowering the bitrate doesn't fix encoding overloaded. What fills the queue is the time per frame, and that comes from the preset, the resolution and the frame rate. The bitrate fixes network problems, not this one.
Two Simple-mode details that catch people out. The "Software (x264 low CPU usage preset, increases file size)" option only changes the recording encoder, not the stream (SimpleOutput.cpp). And if you record at a different quality than you stream, OBS runs a second x264 encode at the same time. Recording quality "Same as stream" reuses the stream's encoder (SimpleOutput.cpp).
Hardware encoders: lighter than x264, but OBS's defaults aren't light
A hardware encoder takes the frame straight from the GPU, so the CPU no longer copies every frame back to system memory and encodes it (obs-video.c). That is why it is the first thing to try. But the defaults OBS gives each one are tuned for quality:
- NVENC defaults to preset P5 ("Slow (Good Quality)"), tune High Quality, two-pass quarter-resolution Multipass, adaptive quantization on, Look-ahead on wherever the card supports it, and 2 B-frames (nvenc-properties.c). OBS's own tooltips say Look-ahead and adaptive quantization cost extra GPU. On an old or entry-level card, try P1 to P3, Multipass off and Look-ahead off.
- AMD (AMF) defaults to the Quality preset with 2 B-frames (texture-amf.cpp). It is the only encoder in OBS that checks itself: when it is created, if the card can't sustain the requested resolution and frame rate at that preset, it steps down to Balanced or Speed on its own (texture-amf.cpp).
- QuickSync defaults to the middle target usage (TU4, "balanced") with 3 B-frames (obs-qsv11.c). TU7 ("speed") is the light end.
- VAAPI on Linux defaults to no B-frames (obs-ffmpeg-vaapi.c), which is already the light end.
On our test laptop's Intel UHD 620, VAAPI used 12% CPU at 720p30, 11% at 1080p30 and 14% at 720p60, with no skipped frames in any of them. That is less than x264 needed on its lightest preset at 720p30. Turning on 2 B-frames made no measurable difference (11% either way).
If a hardware encoder fails outright, OBS doesn't fall back to x264. The stream stops with "Encoding error" (obs-encoder.c).
Resolution, frame rate and the scaling filter
"Base (Canvas) Resolution" is what OBS draws, "Output (Scaled) Resolution" is what it encodes. The scaling between them runs on the GPU, once, and every encoder shares it (obs-video.c). A few things the code does that the settings dialog doesn't tell you:
- The default filter is Bicubic (OBSBasic.cpp), not Bilinear. Per output pixel, Bilinear is 1 texture read, Bicubic 9, Lanczos 25 (bicubic, lanczos). Area gets more expensive the bigger the downscale.
- Below half size, your choice is ignored. If the output is less than half the canvas in both width and height (1920x1080 down to 960x540 or smaller), OBS always uses a cheap 8-sample bilinear filter, whatever you picked (obs-video.c).
- Matching sizes skip the step entirely. If output and canvas match, there is no scaling pass (obs-video.c). If they are within 16 pixels, it uses plain bilinear.
Scaling is a GPU cost, not a CPU cost. It matters when rendering lag is your problem, and barely at all when encoding lag is.
Keep the color format on NV12
OBS converts the picture to the encoder's format on the GPU, as long as you use one of the YUV formats it has a converter for (obs.c). Two settings quietly move work back to the CPU:
- x264 only takes NV12, I420 or I444 (obs-x264.c). Pick anything else and OBS converts every frame on the CPU before encoding.
- NVENC's fast path needs NV12 or P010. With anything else it logs "nv12/p010 not active, falling back to non-texture encoder" and switches to a path that copies every frame through the CPU (nvenc.c).
The settings window even warns that other formats are "primarily intended for recording" (en-US.ini).
Settings that do less than people say
- Process Priority (Settings → Advanced, Windows only). It raises the priority of OBS's own process, only while you stream or record, and resets it when you stop (OBSBasic.hpp, platform-windows.cpp). It can help x264 win CPU time from the game, and it does nothing for rendering lag on the GPU. On Linux and macOS the setting doesn't exist.
- "Enable network optimizations" (Windows only). It moves sending to a separate thread with a buffer and doesn't change CPU or GPU load in any meaningful way. It is also switched off silently for
rtmps://addresses (rtmp-stream.c). - Hiding the preview: see rendering lag above. It saves drawing the OBS window, not the scene.
- Lowering the bitrate for encoding overloaded: see encoding lag above.
If the weak part is your connection
A toaster PC often comes with a toaster connection. "Dropped frames (Network)" is not packet loss. It counts frames OBS threw away because its own send queue got too far behind: once the unsent video spans more than 700 ms, OBS drops B-frames, and past 900 ms it drops every frame until the next keyframe (rtmp-stream.c, defaults at L1903-L1906). Audio is never dropped. With Twitch's 2-second keyframe interval, one drop can freeze the picture for up to two seconds.
The setting that prevents this is "Dynamically change bitrate to manage congestion" under Settings → Advanced → Network. Once the queue reaches 200 ms, well before any drop, OBS lowers the bitrate to what your connection actually managed to send over the last one to two seconds, then raises it by 10% of your setting every 4 seconds until it is back (rtmp-stream.c). Two catches: it needs stream delay off, and while it is on OBS stops dropping frames altogether, so the dropped-frames counter stays near zero even when quality dips. OBS's log literally says "Dynamic bitrate enabled. Dropped frames begone!" (rtmp-stream.c).
Tested on a real toaster
Every number in this section was measured, not estimated. The machine: a Dell Vostro 3580 office laptop from 2019. Intel Core i7-8565U (4 cores, 8 threads, 15 W), 8 GB of RAM, Intel UHD 620 graphics, Ubuntu with OBS 32.2. Each run recorded the same looping 1920x1080 gameplay clip for two and a half minutes, plugged in, on the Performance power mode. CPU, frame render time and the skipped and missed frame counters came from OBS itself through obs-websocket; clock speed and temperature came from the system.
CPU is the whole system's usage, averaged over the run. Render time is OBS's own average time to draw a frame. Skipped frames are the encoding-lag counter, the one behind "Encoding overloaded".
| Settings (1080p canvas) | CPU | Render time | Skipped frames |
|---|---|---|---|
x264 faster, 720p30 | 51% | 2.4 ms | 0 |
x264 veryfast, 720p30 (OBS default) | 34% | 1.9 ms | 0 |
x264 superfast, 720p30 | 23% | 1.5 ms | 0 |
x264 ultrafast, 720p30 | 18% | 1.5 ms | 0 |
x264 veryfast, 1080p30 | 62% | 2.6 ms | 0.2% |
x264 veryfast, 720p30, I444 color | 33% | 4.6 ms | 0 |
x264 veryfast, 720p30, 10 minutes | 28% | 1.4 ms | 0 of 18,040 |
| VAAPI (Intel GPU), 720p30 | 12% | 1.2 ms | 0 |
| VAAPI, 720p30, 2 B-frames | 11% | 1.2 ms | 0 |
| VAAPI, 1080p30 (no scaling) | 11% | 1.8 ms | 0 |
| VAAPI, 720p60 | 14% | 1.4 ms | 0 |
| VAAPI, 720p30, Studio Mode on | 9% | 1.2 ms | 0 |
Fixing "Encoding overloaded" one step at a time
We also started from a setup this laptop can't handle and fixed it one change at a time, keeping each fix:
| Step | CPU | Skipped frames | Rendering lag |
|---|---|---|---|
Start: x264 veryfast, 1080p60 | 87% | 89% | 10% |
| + output 720p (still 60 fps) | 74% | 3.8% | 2.4% |
| + 30 fps | 25% | 0 | 0 |
+ preset superfast | 21% | 0 | 0 |
| Swap x264 for VAAPI (Intel GPU) | 15% | 0 | 0 |
| VAAPI back at 60 fps | 26% | 0.03% | 0.03% |
The same default settings measured 34% CPU in the main runs and 25% in the ladder an hour later, so read differences under about 10 points as noise. The skipped-frame counts are steadier: they were either zero or clearly not.
What surprised us:
- 1080p60 isn't a little too much, it is unusable. 89% of the frames were skipped. Two changes (720p, then 30 fps) took it to zero.
- Heat didn't stop it. The CPU sat at 97 to 100 °C in almost every run and still held about 2.9 GHz. Ten minutes at the default settings skipped nothing.
- The integrated GPU is the real upgrade. At 720p30, VAAPI needed 12% of the CPU where x264 needed 34% on its default preset, and it handled 60 fps with no skipped frames at all.
- I444 doesn't show up as CPU. It cost about the same CPU as NV12 but more than doubled the render time (4.6 ms against 1.9 ms), because the color conversion moves into the frame's drawing time.
What we couldn't measure here: everything Windows-only (Game Capture, Window Capture modes, Process Priority, network optimizations), NVENC, AMD and QuickSync, running on battery, and a live stream to Twitch. The scaling filters are next on our list to measure. Those parts of the guide come from the source code only. We'll add numbers when we test them.
Your VOD afterwards
A stream that held together is worth keeping. Turn on Store past broadcasts in your Twitch Creator Dashboard (Settings → Stream → VOD Settings), or Twitch won't save it and you can't make clips or Highlights from it. On a weak PC, don't add a separate local recording at a different quality just to have a copy: that is a second encode running at the same time.
If you want short clips out of the stream without scrubbing hours of it on the same laptop, that is what FlowCut does: it processes the VOD on its own servers, finds the moments, cuts them and renders vertical clips and montages, and you or your AI agent decide which ones to keep. How to clip a long VOD walks through it.
FAQ
What are the best OBS settings for a low-end PC?
A hardware encoder if you have one (NVENC, AMD, QuickSync or VAAPI) with a light preset, 1280x720 output, 30 fps and the NV12 color format. Without a hardware encoder, x264 on superfast or ultrafast at 720p30.
Why does OBS say encoding overloaded when my CPU isn't at 100%?
Because the warning isn't based on CPU usage. It appears when the encoder skipped more than 10 frames in the last second and at least 0.1% of all frames so far. A single encoding thread can fall behind while the CPU as a whole still has room.
What's the difference between rendering lag and encoding lag?
Rendering lag means OBS couldn't draw the frame in time (GPU or a slow source). Encoding lag means the encoder couldn't take the frame in time (the CPU for x264). They are separate counters in the Stats dock and need different fixes.
Does disabling the preview improve OBS performance?
A little. OBS still renders the whole scene every frame. Disabling the preview only skips drawing it in the OBS window.
Is NVENC or x264 better on a weak PC?
NVENC (or any hardware encoder) on a weak CPU, because it takes the encoding and the frame copy off the CPU. Lower its preset and turn off Look-ahead and Multipass if the GPU is also weak.