How Does Cloud Gaming Work? The Engineering Behind Gaming Without a Graphics Card

Running a game not on your computer but on a server miles away, and playing it smoothly on your screen, sounds like magic. It isn’t. There’s very concrete engineering behind it, and every piece of that engineering is accountable down to the millisecond.

Here’s what we learned building Nitrova, without the hype.

The loop: input, frame, network

Cloud gaming is really one loop that never stops. You press a key, and that input travels to the server. The powerful graphics card on the server renders the frame. The frame is instantly compressed into video, sent back to you over the network, decoded and drawn on your screen. Sixty times a second, or more.

• Input: events from your keyboard, mouse or controller are sent to the server.
• Rendering: the game runs on the server’s GPU, just as it would on a regular PC.
• Compression: a hardware encoder turns each frame into video within milliseconds.
• Delivery and decoding: the video reaches you, and your device decodes and displays it.

The real enemy: latency

The total time this loop takes is the gap between pressing a key and seeing the result on screen. Nobody cares about a few seconds of buffering when watching a movie; in a game, even 100 milliseconds is noticeable.

In our own measurements, median network latency on a 12 Mbps stream came in at 9 milliseconds. When we raised the bandwidth to 15 Mbps on the same line, jitter shot up to 65 milliseconds. The lesson is clear: more bandwidth isn’t always better. Using only what the line can handle is what keeps things smooth.

Balancing bandwidth and quality

There’s a constant trade-off between image quality and smoothness. As resolution and frame rate go up, so does the data sent every second. If the line can’t carry it, frames arrive late or get dropped.

That’s why a good client picks its settings by measuring, not guessing. In Nitrova, leaving image quality on auto uses your display’s resolution, and leaving bandwidth on auto uses a value the server sets by measuring what your line can handle.

Why our own servers?

In cloud gaming, physical distance to the server translates directly into latency. The closer the servers, the shorter the loop. That’s one reason we run our own infrastructure in data centers in Istanbul and Bursa.

Not just for games

The same infrastructure works for any job that needs a graphics card. Rendering, video editing or AI training: you can use a remote RTX 3090 as if it were your own machine. For some workloads, there’s no need to stream the screen at all: your code and data stay on your computer, and only the bytes the computation needs at that moment go to the server.

Honest limits

Cloud gaming can’t replace a local PC in every situation. An unstable Wi-Fi connection will make even the best server look bad. In competitive games, a few milliseconds can make a difference. Some games’ anti-cheat systems don’t allow remote play. Better to say all this up front than to hide it.

Nitrova is currently in beta, and with every release we push those limits back a little further. We want to prove that gaming without a graphics card isn’t magic. It’s good engineering.