Question / Help Hunting for Slow CPU Preset

carlmmii

Active Member
I'll put this plainly. This forum is for OBS Studio support. Streamlabs OBS is a separate program, being developed by an entirely different group.

That said, I can offer my best guess as to what your problem is.

You're most likely seeing stuttering from missing frames, which are likely the result of rendering lag. The higher you push your CPU, the more you're creating an artificial bottleneck in the game, resulting in lowering your GPU usage and allowing SLOBS to actually have GPU resources for frame building.

Cap your in-game framerate. Stick with the medium preset.
 
1950x. jaycenttwo youtuber tried it with custom water cooling and im def not please with it
If money is not a problem, 2990wx is the way to go.

im sorry guys. i really dont want to waste your time. just give me a suggestion on which cpu i should get. i can get i7-8700 or i7-8086k. any suggestion is perfect. just 1080p 60 fps slow cpu preset with x62 kraken cooler. cpu with more ghz like more than 4.0 is perfect.plz
None of them will probably do the job.

If you are using dual-pc setup, you don't need as much IPC and frequency increase as you need more cores. On that PC your are only coding x.264, and that's a very demanding task that uses very well as much threads as possible in your hardware.

And for any CPU, an AIO liquid cooling sollution is not the best cooling you can get either. Go for a custom liquid cooling with as much and as big radiators as you can with proper fans.

should've mention dual pc smfh. im stressed out.
We need all this information and much more.

What are you using to capture the game in the dual PC setup? NDI?, capture card?, what capture card?.... We need a log.
 

Hothazer

New Member
If money is not a problem, 2990wx is the way to go.


None of them will probably do the job.

If you are using dual-pc setup, you don't need as much IPC and frequency increase as you need more cores. On that PC your are only coding x.264, and that's a very demanding task that uses very well as much threads as possible in your hardware.

And for any CPU, an AIO liquid cooling sollution is not the best cooling you can get either. Go for a custom liquid cooling with as much and as big radiators as you can with proper fans.


We need all this information and much more.

What are you using to capture the game in the dual PC setup? NDI?, capture card?, what capture card?.... We need a log.
ill see about that cpu. im using capture card and should i wait for the new amd cpu to come out? like 3700x it seems promising. err cant get log because i use slobs seems i used wrong community post threads
 

koala

Active Member
You don't seem to be aware of some relations, if you talk about overclocking, x264 presets, cpu demand for x264 and x264 quality.
I take x264 figures from here.
  • Overclocking gives up to 10-15% more cpu power.

  • faster preset gives approx. 75% file size compared to veryfast (which means something like 25% better quality)
  • faster preset needs 58 % more cpu power than veryfast preset

  • medium preset gives approx. 80% file size compared to veryfast (yes, that's worse than faster according to this)
  • medium preset needs 150% more cpu power of veryfast preset (that's 2,5 times)

  • slow preset gives approx. 76% file size compared to veryfast (yes, that's still worse than faster according to this)
  • slow preset needs 268% more cpu power of veryfast preset (that's 3,68 times)

  • With overclocking you don't solve any issue. You increase some benchmark figures, but nothing visible except increased instability

  • With any preset beyond faster you just make your CPU work more, but don't really increase the quality. Even if the referenced report might be a bit different if it were performed for different footage, especially if other footage results in actual quality increase for presets beyond faster, this size decrease that results in quality increase would be some 1-5%
  • A quality difference of 1-5% is not barely visible, it isn't visible at all, given the fact that about 70-80% of video detail is removed in any case if you do streaming.
If you want high quality streams, focus on a smooth running machine and smooth running stream without a single lost or lagged frame. Lost or lagged frames are much worse for visual perception than highest compression performance.

With "lagged frame", I mean how constant the frame rate is.

  • A 60 fps stream where the GPU is producing one frame every 16.6 ms and this frame is encoded, transmitted and displayed to the viewer every 16.6 ms, appears absolutely smooth to the viewer.
  • A 60 fps stream, where the GPU is producing 80, 100, 200, 300 or 400 fps and OBS has always to take one of the intermediate frames that are approx. 16.6 ms away, but not exactly 16.6 ms, doesn't appear absolutely smooth to the viewer. This is because the actual frame rate of the GPU is not dividable by 60. Such a video will always appear somewhat stuttery. Not really bad, but you can tell a difference between such a video and a really smooth video.
  • A 60 fps stream, where the GPU is overloaded and producing less than 60 fps, has always some kind of stutter
It's also not really constant fps what is needed, it's also important that frames are shown at exactly the point in time they were produced. If a game produces 5 frames with 15 ms, then 16 ms, then 20 ms, then 15 ms in between, which is 16,6 ms average, the corresponding video isn't smooth if the video player displays the 5 frames every 16.6 ms. Instead, it is smooth if the video player shows the frames with the same delay between the frames as they were produced, i. e. with 15 ms, 16 ms, 20 ms, 15 ms. To achieve this, modern container formats contain time stamps for every frame. The issue with this is that there is no time stamp within the GPU when the frame was produced. It may be that a frame was produced 4 ms before it is scanned by OBS, so the time stamp OBS puts in the video is 4 ms off. This is ok, if every frame would be 4 ms off, but in high load situations this delay varies, so the resulting time stamps stutter.

To avoid such situations, make sure you don't load your system fully, so there is leeway for every process to perform its work exactly when it is supposed to. Don't fully load the encoder, don't fully load the CPU, don't fully load the GPU. So: don't use the most CPU demanding preset you think your CPU can handle!
 

TryHD

Member
@koala I'm sorry that I have to say it but you are wrong, simply because your assumptions about CRF are wrong, CRF can vary greatly with different presets, even the x264 devs repeat this always. So for this case here the values from http://blogs.motokado.com/yoshi/2011/06/25/comparison-of-x264-presets/ are useless because they don't include SSIM or PSNR

Your part from "lagged frame" is true on the other hand, till the part with the CPU demanding preset, of course you should go for best preset you can achive while maintaining good frametimes, everything else would be a waste of money.
 

Hothazer

New Member
You don't seem to be aware of some relations, if you talk about overclocking, x264 presets, cpu demand for x264 and x264 quality.
I take x264 figures from here.
  • Overclocking gives up to 10-15% more cpu power.

  • faster preset gives approx. 75% file size compared to veryfast (which means something like 25% better quality)
  • faster preset needs 58 % more cpu power than veryfast preset

  • medium preset gives approx. 80% file size compared to veryfast (yes, that's worse than faster according to this)
  • medium preset needs 150% more cpu power of veryfast preset (that's 2,5 times)

  • slow preset gives approx. 76% file size compared to veryfast (yes, that's still worse than faster according to this)
  • slow preset needs 268% more cpu power of veryfast preset (that's 3,68 times)

  • With overclocking you don't solve any issue. You increase some benchmark figures, but nothing visible except increased instability

  • With any preset beyond faster you just make your CPU work more, but don't really increase the quality. Even if the referenced report might be a bit different if it were performed for different footage, especially if other footage results in actual quality increase for presets beyond faster, this size decrease that results in quality increase would be some 1-5%
  • A quality difference of 1-5% is not barely visible, it isn't visible at all, given the fact that about 70-80% of video detail is removed in any case if you do streaming.
If you want high quality streams, focus on a smooth running machine and smooth running stream without a single lost or lagged frame. Lost or lagged frames are much worse for visual perception than highest compression performance.

With "lagged frame", I mean how constant the frame rate is.

  • A 60 fps stream where the GPU is producing one frame every 16.6 ms and this frame is encoded, transmitted and displayed to the viewer every 16.6 ms, appears absolutely smooth to the viewer.
  • A 60 fps stream, where the GPU is producing 80, 100, 200, 300 or 400 fps and OBS has always to take one of the intermediate frames that are approx. 16.6 ms away, but not exactly 16.6 ms, doesn't appear absolutely smooth to the viewer. This is because the actual frame rate of the GPU is not dividable by 60. Such a video will always appear somewhat stuttery. Not really bad, but you can tell a difference between such a video and a really smooth video.
  • A 60 fps stream, where the GPU is overloaded and producing less than 60 fps, has always some kind of stutter
It's also not really constant fps what is needed, it's also important that frames are shown at exactly the point in time they were produced. If a game produces 5 frames with 15 ms, then 16 ms, then 20 ms, then 15 ms in between, which is 16,6 ms average, the corresponding video isn't smooth if the video player displays the 5 frames every 16.6 ms. Instead, it is smooth if the video player shows the frames with the same delay between the frames as they were produced, i. e. with 15 ms, 16 ms, 20 ms, 15 ms. To achieve this, modern container formats contain time stamps for every frame. The issue with this is that there is no time stamp within the GPU when the frame was produced. It may be that a frame was produced 4 ms before it is scanned by OBS, so the time stamp OBS puts in the video is 4 ms off. This is ok, if every frame would be 4 ms off, but in high load situations this delay varies, so the resulting time stamps stutter.

To avoid such situations, make sure you don't load your system fully, so there is leeway for every process to perform its work exactly when it is supposed to. Don't fully load the encoder, don't fully load the CPU, don't fully load the GPU. So: don't use the most CPU demanding preset you think your CPU can handle!
so it would have to be a most powerful cpu to go slow preset with stability. my friend is going i9-9900k since it has more cores and threads with his build so i might try out his build to see if i can reach slow preset. what do do you think?
 

BK-Morpheus

Active Member
waste of money, especially, when the Threadripper 1950x was not able to satisfy your needs. Also, as I said, the visual difference between fast and slow is almost invisible.
 

TryHD

Member
so it would have to be a most powerful cpu to go slow preset with stability. my friend is going i9-9900k since it has more cores and threads with his build so i might try out his build to see if i can reach slow preset. what do do you think?
I would say that a i9 9940 (9960 would be also good to have some cores spare for OBS scenes and so) would be the best bet for 1080p and setting the threads to 27 via x264 options, because you don't want to go beyond that. See http://forum.doom9.org/showthread.php?p=1646307#post1646307 why
 
Last edited:

Hothazer

New Member
I would say that a i9 9940 (9960 would be also good to have some cores spare for OBS scenes and so) would be the best bet for 1080p and setting the threads to 27 via x264 options, because you don't want to go beyond that. See http://forum.doom9.org/showthread.php?p=1646307#post1646307 why
But wait tho. I’m confused. Take a look at Tsm_Myth stream quality. His cpu preset is on slow and no issue at all. He is using i7-7800x with 6 core. How much is he overclocking it to get that quality? Wouldn’t i9-9900k would be safe bet with 2 more cores and more clock speed available compared to his cpu? Same as dr.Lupo with i7-4820 i think. He doing slow preset also with no stutter. I’m not arguing. Just making our observation perfect
 

TryHD

Member
Have they posted anywhere what exact settings they use? I don't see how they would do 1080p 60 fps on slow with a six core or even less without overriding parts of the presets via x264 options.
 

Hothazer

New Member
I know that what setting they use in stream setting by using twitch analyzer. It tells you what bitrate and FPS and preset they use. I definitely know they don’t use a lot of cores except drlupo. He using same threadripper as i am and his perfectly fine. But i don’t think he using the threadripper anymore to be honest
 

TryHD

Member
I know that what setting they use in stream setting by using twitch analyzer. It tells you what bitrate and FPS and preset they use. I definitely know they don’t use a lot of cores except drlupo. He using same threadripper as i am and his perfectly fine. But i don’t think he using the threadripper anymore to be honest
The preset is just a esimation like said on the page,it looks only at ref frames, set them higher ends with a higher preset estimation, so that doesn't help here sorry. See https://r-1.ch/analyzer/results/test_720.072479
 

DEDRICK

Member
How are you sending the signal to the stream PC via capture card. Are you running OBS on the gaming PC to project it, or are you simply duplicating your main monitor?
 

Hothazer

New Member
How are you sending the signal to the stream PC via capture card. Are you running OBS on the gaming PC to project it, or are you simply duplicating your main monitor?
I’m using elgatoHD Pro with a second pc to capture my gaming pc
 

TryHD

Member
For your current setup, what keyframe interval do you use currently? changing from the recommend 2 secs to 4 secs should increase the quality already a bit. Maybe that is already enough for you to get to your wanted quality?
 

Hothazer

New Member
For your current setup, what keyframe interval do you use currently? changing from the recommend 2 secs to 4 secs should increase the quality already a bit. Maybe that is already enough for you to get to your wanted quality?
I have it on 2. I can try 4 when i get home
 

Hothazer

New Member
For your current setup, what keyframe interval do you use currently? changing from the recommend 2 secs to 4 secs should increase the quality already a bit. Maybe that is already enough for you to get to your wanted quality?
just tried the thread=27. it clear but it not helping with fast motion. i just got in a extreme build fight with 11 players and it just freezes when fighting. im just gonna go for i9-9900k
 
Top