Run the test above for at least 60 seconds. Here is what each
number means for this game specifically.
Red lines (packet loss)
Loss is the reason to raise, not lower, your interp. A client interpolating over 15 ms cannot ride out a gap larger than that, and the result is the hitscan-through-the-target effect people blame on the netcode.
Jitter (why lerp exists)
Interpolation exists precisely to hide jitter. Measure it here first: if your jitter is well under 15 ms, the minimum interp is realistic. If it is not, a higher cl_interp_ratio will feel better than a lower one.
Ping (latency)
Under 50 ms to a 66-tick server is comfortable. Ping and lerp add up in the total delay between your input and the server's world, which is why tuning one while ignoring the other rarely helps.
What you see in game, and which number on this page explains it.
Team Fortress 2 is one of the very few games that hands you control over your own interpolation delay, and one of the fewer still where the right value is arithmetic rather than taste. Valve servers generate roughly 66.7 world states per second and cap cl_updaterate at 66. Your lerp follows directly from that: cl_interp equals cl_interp_ratio divided by cl_updaterate, so ratio 1 gives 1/66 - 15.2 milliseconds - and ratio 2 gives 30.3.
Interpolation exists to hide jitter. Your client deliberately draws the world slightly in the past so that it always has two known positions to draw between, rather than guessing. The lerp value is how far into the past it works. Set it too low and the client reaches for a packet that has not arrived yet - and the result looks exactly like packet loss, with hitscan passing through a target that was visibly there.
This is why copying a competitive player's config is such reliably bad advice. Their 15.2 ms works because their connection delivers packets evenly. If this chart shows any measurable loss, ratio 2 is the better setting for you: riding out a gap is worth more than 15 milliseconds of theoretical advantage you cannot actually use.
One more distinction worth carrying into the game: choke and loss are different failures. Loss means packets vanished in transit, and that is what this page measures. Choke means the server did not send them because your rate settings did not permit it - a client-side configuration problem with a client-side fix. A clean chart here combined with choke in net_graph points squarely at your rate settings rather than at your ISP.
If the chart above shows packet loss or high jitter, work through
these in order.
What is the correct cl_interp for TF2?
It depends on your connection, and it is arithmetic: cl_interp_ratio divided by cl_updaterate. On Valve servers with the 66 cap, ratio 1 gives 15.2 ms and ratio 2 gives 30.3 ms. Use ratio 2 if you measure any packet loss.
What tick rate do TF2 servers run at?
Valve servers generate about 66.7 world states per second and limit cl_updaterate to 66. Community servers sometimes run higher tick rates.
Is a lower lerp always better?
No. Lerp is the buffer that hides jitter. Setting it below what your connection needs means interpolating from packets that have not arrived yet, which looks exactly like packet loss in game.
What is the difference between choke and loss?
Loss means packets vanished on the way. Choke means the server did not send them because your rate settings did not allow it. This test measures loss; net_graph in game shows both.