What your GPU's video encoder does for a screen recording

Your GPU’s video encoder is a separate part of the graphics chip that does one job: it turns frames of pixels into compressed video, so your CPU does not have to. It is the difference between a screen recording that stutters, drops frames and spins up the fan, and one you forget is running. Focal, a screen recorder for Windows, records straight into it.

My name is Vince, and I make Focal. Building a screen recorder means pressing record more times than anyone should and listening, every time, for the laptop fan to go from a hum to a leaf blower. Here is what I learned about the part of the chip that keeps it at a hum.

Why encoding is the expensive part

A 4K screen is 3840 by 2160 pixels, a little over 8 million of them. At 60 frames a second that is about 500 million pixels every second, and at four bytes a pixel, about 2 GB of raw image a second. An hour of it, stored as is, would fill more than 7 TB.

Nobody stores that. A video encoder compares each frame with the ones around it, keeps what changed, describes the rest as “the same as before, moved a little”, and drops detail your eye will not miss. The file comes out hundreds of times smaller. That takes a great deal of arithmetic, and it has a deadline: at 60 fps each frame has to be done in about 16.7 milliseconds, or the next one arrives first. That is sixty deadlines a second, and not one of them can be moved.

Capturing the screen is cheap by comparison, because the picture already exists: Windows draws your desktop on the GPU to show it on your monitor. Encoding is where the work is. Done on the CPU, with a software encoder such as x264, it competes with the very thing you are recording: the build in your terminal, the browser with forty tabs, the game, the video call. When the CPU falls behind, a recorder can skip frames or queue them, and both show up in the video as stutter.

What the hardware encoder changes

Almost every graphics chip sold in the last ten years, from a gaming card to the graphics built into a laptop processor, has a block of silicon that does nothing but encode video. NVIDIA calls theirs NVENC. AMD’s is VCN, short for Video Core Next, and Intel’s is Quick Sync Video. Whatever the name, it runs alongside the part that draws your windows rather than taking turns with it, and it barely needs the CPU.

What that buys you:

• The CPU stays free. Encoding drops from a large share of the processor to a few percent, so the program you are recording runs almost as it would with no recorder at all.
• No dropped frames. Real-time video is what the block was made for, so it keeps pace with the screen even when the rest of the machine is busy.
• A laptop that stays cool. A block built for one job spends a fraction of the power the CPU would on the same frames, so the fan stays quiet, the battery lasts longer and your lap stops being part of the cooling system.

Here is what that looks like in Focal. Measured on 2026-10-31 on an RTX 3090 with a Ryzen 9 9950X3D, an hour of recording at 4K 60 fps with everything on used 4.8% of the CPU: 216,032 frames, none dropped, memory flat at 130 MB. That is one machine, and other machines differ. More of Focal’s numbers are on the home page.

H.264 and HEVC

These are the two formats a hardware encoder usually writes. H.264 is the one nearly everything plays: browsers, phones, editors and the sites you upload to. HEVC, also called H.265, is its successor and makes a smaller file at the same quality, but some older devices and apps still do not play it.

Recent hardware encoders handle both. Focal records in HEVC on NVIDIA GPUs and exports MP4 in either. Its exports give both codecs the same bitrate at each quality, so an HEVC export is no smaller than an H.264 one, and the Export dialog says what it costs you: fewer players can open it. My own rule is H.264 for any file that leaves my computer. No file is small enough to make up for a phone call about why it will not play.

Why frames should stay on the GPU

The screen image starts on the GPU, and the encoder is on the GPU. A recorder that copies each frame into main memory, works on it there and sends it back is moving that whole torrent, 2 GB a second, over to the CPU’s side of the machine and back again, and spending processor time on every copy. It is the video version of printing an email so you can scan it back in, once for every frame. That is where a lot of recorders lose time, and why some still drop frames on a busy machine with a hardware encoder sitting right there.

The fast route is the short one. The frame goes from the screen to the encoder without leaving the GPU, and only the compressed video, hundreds of times smaller, comes back to be written to disk. Focal records straight into the GPU’s hardware encoder, and every frame is on disk the moment it is written, so a crash or a power cut keeps what you had.

Export is encoding too

Exporting an edited recording is the same job with an extra step at the front. The recorded video is decoded back into frames, the edits are drawn onto each one (the zoom, the cursor, the captions, the background), and the result is encoded again. GPUs have a separate block for decoding as well, so all three steps can stay on the graphics chip.

Focal does it that way: decoding, compositing and encoding all run on the GPU. On the machine above, a 4K recording exported to 1080p ran 8 times faster than real time, so ten minutes of recording took about a minute and a quarter. That is not long enough to justify getting up for coffee, though I have tried.

What if your machine has no hardware encoder

Some machines have none: older desktops, a few budget graphics cards, and some virtual machines, which can hand the system inside them a plain display adapter with no video blocks at all.

In that case Focal falls back to the CPU. A modern desktop CPU still records 4K at 60 fps. A laptop CPU may not, more so on battery, when it runs slower to save power. The answer there is fewer pixels: a lower resolution or a lower frame rate. In Focal the simple way to get fewer pixels is to record one window or a region instead of the whole monitor, which for a demo is often the clearer video anyway. Nobody has finished watching a demo wishing they had seen more of the taskbar.

If you only need a quick clip, Windows has recorders of its own, and how to record your screen on Windows 11 goes through each built-in way and when you need more.

How to tell which one your machine has

Open Task Manager with Ctrl+Shift+Esc, go to the Performance tab and select your GPU. The four small graphs at the top each show one engine of the chip, and clicking the name above a graph lets you switch it to another engine. On many cards one of them is called Video Encode. Start a recording and watch it: if that graph rises while the CPU graph stays low, the hardware encoder is doing the work. If the CPU graph is the one climbing, the fan will usually tell you before Task Manager does.

Intel graphics may show no graph by that name. Intel’s support pages explain that its encoder shows up as Video Decode 1. If no GPU appears in the list at all, update the graphics driver, since Task Manager only shows GPU engines with a recent enough one.

What to ask of a recorder

Three questions cover most of it. Does it use the hardware encoder? What does it fall back to when there is none? And does it say what it measured, and on which machine? A number with no machine attached is a weather forecast with no city.

If you know OBS Studio, Focal vs OBS Studio for demo videos compares the two for that kind of recording. What Focal does beyond recording, from automatic zooms to captions, is in the Focal 1.0 announcement. The surest test is your own computer: recording with the free trial has no time limit and needs no card or account, so record ten minutes with Task Manager open and watch the Video Encode graph.