Blur Busters Forums

I tested in several games and I felt less input delay with the same fps, has anyone felt that too? I know it looks strange.

RES: 1366x768
Native res: 1920x1080

Because frame-time is composed by both CPU time + GPU time. So if your CPU renders in 3.3ms and the GPU in 2.5ms, you will have 300 fps (1000 *1/max(cpu, gpu)) once the buffer stabilizes but the total latency of a single frame is 3.3 + 2.5. If you can reduce the GPU render time to 0.5ms by lowering the resolution than you have the same fps because your CPU is bottlenecking but your end to end frame latency drops by 2ms.

I notice this too but my setup is old - i5 3570k with a GTX 1070

lower resolutions definitely feel smoother, probably due to the CPU (bottleneck)

There can be lower lag at the same frame rates because there is lower percentage utilization.

A heavily loaded GPU will sometimes lag more even at the same frame rate.

andrelip wrote: ↑

18 Apr 2020, 11:15

Because frame-time is composed by both CPU time + GPU time. So if your CPU renders in 3.3ms and the GPU in 2.5ms, you will have 300 fps (1000 *1/max(cpu, gpu)) once the buffer stabilizes but the total latency of a single frame is 3.3 + 2.5.

It can math out that way, or differently -- depending on how the programming is done, and how it's all multithreaded (or not!), there are different latency behaviours.

The CPU can continue on the next frame while the GPU is rendering the previous frame, due to the way things are pipelined.

Frame renders also can be overlapped somewhat (shingled in a GANTT chart). This is a much more noticeable behavior in SLI operation -- you have potentially double the frame rate, but the frame time is still equal to a single GPU (time period of two frame intervals), because of this parallelism.

Chief Blur Buster wrote: ↑

18 Apr 2020, 20:40

There can be lower lag at the same frame rates because there is lower percentage utilization.

A heavily loaded GPU will sometimes lag more even at the same frame rate.

andrelip wrote: ↑

18 Apr 2020, 11:15

Because frame-time is composed by both CPU time + GPU time. So if your CPU renders in 3.3ms and the GPU in 2.5ms, you will have 300 fps (1000 *1/max(cpu, gpu)) once the buffer stabilizes but the total latency of a single frame is 3.3 + 2.5.

It can math out that way, or differently -- depending on how the programming is done, and how it's all multithreaded (or not!), there are different latency behaviours.

The CPU can continue on the next frame while the GPU is rendering the previous frame, due to the way things are pipelined.

Frame renders also can be overlapped somewhat (shingled in a GANTT chart). This is a much more noticeable behavior in SLI operation -- you have potentially double the frame rate, but the frame time is still equal to a single GPU (time period of two frame intervals), because of this parallelism.

The minimum latency of a rendered frame is CPU time + GPU time. Even if you have a queue and the CPU do not idle while the GPU is rendering, all you can do is to avoid additional latency.

With that in mind, that's why many pros feel that a lower resolution is better. Even the F1 2019 winner plays on low (David Tonizza). Playing in low reduces the GPU time by a lot and directly reduces the latency in systems, even if the CPU is the bottleneck.

In theory, you can insert frames that will not be rendered; this way, you can skip GPU rendering entirely and have a lower time. It would be good to improve latency, input capture, better tick rate timing, and so on. It will maximize sampling from inputs, but the frame that you will be seeing is still lagging by the CPU + GPU time.

----

Imagine having a perfect 8 ms in CPU and 8 in GPU
With no queue, you will have 16ms and 60fps. The minimum latency is 16ms. 50% utilization on both CPU and GPU

With the queue, you will have 8ms and 120fps. The minimum latency is 16ms (CPU start to GPU rendered). 100% utilization on both CPU and GPU. The frames will be delivered each 8 ms, but they have 16 ms in their whole lifecycle.

----

Many people noticed that GPU utilization affects input latency,
this one is a perfect example: https://www.youtube.com/watch?v=8ZRuFaFZh5M @ 13:50

But they don't seem to know the cause, and the reason is:

Lower GPU utilization means faster render times (the GPU is idling more).
With lower GPU render time, lower latency.

When GPU reaches 100%, it means that the GPU is slower than the CPU, and it will start queueing frames until it feels the buffer. Worst case scenario.

ey3 wrote: ↑

16 Apr 2020, 14:39

I tested in several games and I felt less input delay with the same fps, has anyone felt that too? I know it looks strange.

RES: 1366x768
Native res: 1920x1080

What people fail to understand is that frames need to be processed on the GPU into the display, which is where GPU clocks/GPU memory clock and things such as smaller res frame lower latency at same framerates.

Higher resolutions require more processing in the GPU, but not really any more processing in the CPU. When you're GPU limited, you get some buffering while waiting for the GPU to finish the previous frame, that one major source of lag, in addition to the extra time the GPU spends processing the frame.

Also, while it probably depends on how the given game calculates mouse input, couldn't mouse DPI start to feel faster at a lower-than-native res when scaling as well? Because I know it feels slower with a higher-than-native res when downsampling.

E.g. short of actual processing improvements on the performance-side due to running in lower resolutions (especially in more demanding games), this could also potentially (psychologically) contribute to the overall impression of "smoother" or more responsive gaming, at least to a certain degree in certain scenarios.

jorimt wrote: ↑

30 Jun 2020, 06:28

Also, while it probably depends on how the given game calculates mouse input, couldn't mouse DPI start to feel faster at a lower-than-native res when scaling as well? Because I know it feels slower with a higher-than-native res when downsampling.

E.g. short of actual processing improvements on the performance-side due to running in lower resolutions (especially in more demanding games), this could also potentially (psychologically) contribute to the overall impression of "smoother" or more responsive gaming, at least to a certain degree in certain scenarios.

In CS:GO the mouse exactly the same if you have the same aspect ratio (i.e. 4:3 or 16:9) and thus a lower resolution wouldn't affect the mouse directly. If you use a stretched resolution though then that's another story.

Regardless, plenty of people say that a lower reslution feels "better", even with black bars. I would believe this is due to the slight increase in frames but I dunno. At the same time a non-native resolution should add slight input lag so theoretically, a lower resolution should be less responsive? (maybe the input lag addition is so small it doesn't matter).

senny22 wrote: ↑

30 Jun 2020, 09:07

In CS:GO the mouse exactly the same if you have the same aspect ratio (i.e. 4:3 or 16:9) and thus a lower resolution wouldn't affect the mouse directly.

Correct, which is why I was sure to add "it probably depends on how the given game calculates mouse input."

The bulk of the difference is likely due to the aforementioned increase in higher possible performance (even at the same average framerates; tighter frametimes, less CPU/GPU wait/faster processing, etc), but I was noting the contribution of DPI/sensitivity differences when downscaling for thoroughness sake, regardless.