Why isn't 4k 4times slower than 1080p?

[Light-Yagami]

I've wondered for quite some time why games's fps don't scale according to resolution. In games I've watched gameplay of and benchmarks, most of the time the number is around 2.5-3x performance decrease. So if 1080p was sth like 140fps, 4k would be around 50-60 mark. 

Why is that?

 

 

[Vandorlot]

This is because goign form 1080 to 4k the gpu starts working more instead of the processor.

[kerradeph]

Not everything being computed is just the exact pixels so there's some overhead and then there's processing stuff for the UHD/1080p. 

[TVwazhere]

5 minutes ago, kerradeph said:

Not everything being computed is just the exact pixels so there's some overhead and then there's processing stuff for the UHD/1080p. 

Anti aliasing for example, 4k benchmarks usually dont run as high AA settings by default

[You_are_a_cunt]

computing scaling is never linear

It's the same reason when looking at multi-core or multi-CPU computing. 4 identical cores will never perform 4 times better than a single core. 2 "identical" CPUs will not perform twice as good as a single CPU.

The way a scene is rendered isn't linear either. Some elements (can be a single pixel or a cluster of pixels) don't require the same amount of computing as others.

I think (and don't quote me on this) there is also some approximation going on (to relieve some computing stress)

[WereCat]

Because some graphical settings, geometry, etc... have the same-ish performance impact regardless of resolution.

[Light-Yagami]

4 minutes ago, WereCat said:

Because some graphical settings, geometry, etc... have the same-ish performance impact regardless of resolution.

Same fps percentage drop regardless of resolution? is that what you mean?

[Light-Yagami]

6 minutes ago, revsilverspine said:

computing scaling is never linear

It's the same reason when looking at multi-core or multi-CPU computing. 4 identical cores will never perform 4 times better than a single core. 2 "identical" CPUs will not perform twice as good as a single CPU.

The way a scene is rendered isn't linear either. Some elements (can be a single pixel or a cluster of pixels) don't require the same amount of computing as others.

I think (and don't quote me on this) there is also some approximation going on (to relieve some computing stress)

Thanks for info.. I've noticed yeah.. How much multicore CPUs perform in terms of performance per core scaling heavily depends on program optimisation and on how cores work together.. it that makes any sense.. 

 

[hconverse02]

Diminishing return? Also game perdormance scaling isn't 4:1, even though the resolution is. 

[You_are_a_cunt]

4 minutes ago, Light-Yagami said:

Thanks for info.. I've noticed yeah.. How much multicore CPUs perform in terms of performance per core scaling heavily depends on program optimisation and on how cores work together.. it that makes any sense.. 

 

I'm not sure if you can select which cores remain active on a multi-core CPU, but if you can you'd probably find that each of them perform differently, at least in theory.

While the differences might be insignificant, they should be there.

[Coaxialgamer]

Not everything a cpu does is related to putting the actual pixels on the  screen. 

A lot of computing work is done by the gpu, most of which doesn't get harder at 4k than it is at 1080p.

GPGPU physics, for example, don't need more ressources at higher resolutions. 

 

You might also have bottlenecks that shift from some components to others. For example, a cpu bottleneck at 1080p could easily turn into a gpu bottleneck at 4k.

 

Anti-aliasing is something else entirely. Unless you are talking about SSAA, most anti aliasing techniques find ways to reduce intensity by not actually rendering the entire scene at a higher resolution, but only smoothing out edges, for example. 

[PlayStation 2]

Because that's not how it'd work. Drawing raw pixels in general isn't that hard. It's what needs to be done that'd sap performance.

[Glenwing]

At least in my tests, it basically is:

Spoiler

I get around 25-30% performance at 4K in most games.

 

2560×1440 is 1.77 times as many pixels as 1080p so you would expect 1/1.77 or 56% as many frames per second, and that's pretty close to what I get too.

 

 

 

 

 

 

Perhaps other reviewers use different graphics settings when they test at 4K. Also bear in mind that high-end graphics cards may encounter CPU bottlenecks at lower resolutions which would prevent 1080p from being 4 times higher than 4K results.

[Mira Yurizaki]

Some of it depends on the work being done. All of the geometry stuff like positioning and manipulating primitives are done independent of resolution. Everything after the rasterization step is what's ultimately affected by resolution.

 

It also depends on how fast the CPU can pump out draw commands to the GPU. At lower resolutions, the GPU is able to accept more of them in a given time frame. At 4K if the GPU can only pump out 60 FPS tops, that means the CPU is sending out 60 draw commands in a second. At 1080p, the CPU would need to send 240 draw commands in a second, which it may not be able to do after processing the logic

[Fahim Foysal]

On 9/19/2017 at 6:13 PM, Light-Yagami said:

I've wondered for quite some time why games's fps don't scale according to resolution. In games I've watched gameplay of and benchmarks, most of the time the number is around 2.5-3x performance decrease. So if 1080p was sth like 140fps, 4k would be around 50-60 mark. 

Why is that?

 

 

I dont know for sure but  if i had to guess then maybe cause game physics and stuff uses the same amount of resources regardless of resolution

[xsimplyjosh]

The thing is 1080p is more reliant on CPU, whereas 4k was more reliant on graphics performance. assuming 60fps each.

Also graphics scaling is never linear as far as I know.

[porina]

On 19/09/2017 at 1:25 PM, revsilverspine said:

computing scaling is never linear

It's the same reason when looking at multi-core or multi-CPU computing. 4 identical cores will never perform 4 times better than a single core. 2 "identical" CPUs will not perform twice as good as a single CPU.

While I agree with the sentiment, "never" is too strong a term for it. There are lots of tasks that are trivially parallel and DO scale linearly with more resource.

[You_are_a_cunt]

1 hour ago, porina said:

While I agree with the sentiment, "never" is too strong a term for it. There are lots of tasks that are trivially parallel and DO scale linearly with more resource.

fair point. I was probably thinking in terms of resulta generated by a benchmark