Latency while Cloning with 480hz OLED in Dual-PC setup

[Yukai]

Hi!
I am pursuing 0 latency Display Cloning (Duplicate) to a Capture Card in a Dual-PC setup with these specs:

Main System:
• Asus PG27AQDP (1440p 480hz WOLED)
• 4060 Ti 14900KF
• Ghost Spectre win10 Compact
• 2208x1242 480hz using SRE (1:1 pixel ratio non-scaled 16:9 23" res with black borders)

Secondary:
• AverMedia Live Gamer Ultra 2.1 GC553G2
• 2208x1242 240hz using CRU

VRR: Off
Scaling: Off
MPO: Off
Exclusive Full Screen with Hardware: Internal/Legacy Flip in all games


I am experiencing added latency on main gaming OLED in the range of few ms. Originally, it made me swap from Ultra S (120hz) to Ultra 2.1 (240hz) capture card, but it didn't solve the issue. The latency persists regardless of resolution used even with the native 1440p/1332p modes.

I wanted to test:
1. 240hz - identical EDID's
2. 240hz - DSC off

However, in attempt to perfectly match timings by creating the same 1242p 240hz resolutions in CRU my OLED would not respect no scaling in NVCP and fullscreen any res to 27". And while using SRE, which works fine, I don't know what timings are being used and can only guess that it either uses CVT-RB standard or drives it's custom resolutions differently while using native 1440p EDID table.

Therefore I seek answers to these questions:
• What is the culprit behind added latency and what are the possible solutions to run 480hz? Is it related to DSC / timings mismatch?
• Is there a way to check and perfectly match timings with the resolution created through SRE?
• Is there a way to make custom resolutions created via CRU have no scaling or respect no scaling setting in NVCP?

Thanks in advance! I know there are a bunch of highly knowledgable people so detailed answers are appreciated <3

[kyube]

I am pursuing 0 latency Display Cloning (Duplicate) to a Capture Card in a Dual-PC setup with these specs:

"0 latency" where?
There is no "0 latency" as a concept.
What I assume you're trying to find is the lowest possible total system latency solution for recording / streaming (correct me if I'm wrong)

The 'fool-proof' solution for recording/streaming gameplay is pointing an external DSLR camera on your screen and recording footage that way.

Main System:
• Asus PG27AQDP (1440p 480hz WOLED)
• 4060 Ti 14900KF
• Ghost Spectre win10 Compact
• 2208x1242 480hz using SRE (1:1 pixel ratio non-scaled 16:9 23" res with black borders)

You're already on a losing battle by using a custom ISO...
What in particular made you decide on that image resolution to achieve a 23" viewport?
Image resolutions which aren't divisible by 8 should be avoided, as encoders/decoders & internal TCON boards don't play nice with certain values.
Though I cannot recall the source for the latter claim, I believe I've read that somewhere around here (by Chief or someone else)

VRR: Off
Scaling: Off
MPO: Off
Exclusive Full Screen with Hardware: Internal/Legacy Flip in all games

What specific games are you planning on recording?
HW: Legacy Flip (FSE) isn't a thing in D3D12 titles for ages now and from anecdotal conversations (no data), the quest of making it work in modern D3D11 games is a Pyrrhic victory.
Sticking to Hardware Composed: Independant Flip should be as good as FSE (emphasis on should)

I am experiencing added latency on main gaming OLED in the range of few ms.
The latency persists regardless of resolution used even with the native 1440p/1332p modes.

A “few milliseconds” is very inconcrete. How have you come to such a conclusion?
Does this added penalty you're referring to occur only when using a capture card?

However, in attempt to perfectly match timings by creating the same 1242p 240hz resolutions in CRU my OLED would not respect no scaling in NVCP and fullscreen any res to 27". And while using SRE, which works fine, I don't know what timings are being used and can only guess that it either uses CVT-RB standard or drives it's custom resolutions differently while using native 1440p EDID table.

DSC-based displays & Nvidia dGPUs disallow custom EDID overrides (colloquially referred to as 'custom resolutions'), therefore you cannot use CRU & rely on SRE at best. Source (creator of CRU): https://www.monitortests.com/forum/Thre ... tility-CRU

You could perhaps use: https://github.com/BoringBoredom/Windows-MultiTool to see the current EDID on Nvidia dGPUs, though I'm not certain how accurate the values are.

• What is the culprit behind added latency and what are the possible solutions to run 480hz? Is it related to DSC / timings mismatch?

DSC's latency overhead was discussed here: viewtopic.php?t=12730&sid=b845a72d94703 ... 30#p118897
It's likely not the added latency you're referring to.
'Timing mismatch' is also likely not the issue.

• Is there a way to check and perfectly match timings with the resolution created through SRE?

Refer to the tool I've linked above

• Is there a way to make custom resolutions created via CRU have no scaling or respect no scaling setting in NVCP?

You're on a DSC-based display, that's already a fruitless endeavour.

[Yukai]

Hi! Thanks for your reply

"0 latency" where?
There is no "0 latency" as a concept.
What I assume you're trying to find is the lowest possible total system latency solution for recording / streaming (correct me if I'm wrong)

My bad, that's exactly right. I've assumed that I can dumb down my goal for simplicity without stating the obvious

You're already on a losing battle by using a custom ISO...

Ghost Spectre Compact is the least intrusive custom OS out there. It doesn't remove anything that could potentionally compromise Windows functionability and only applies known to be safe changes to settings on install. It's proven to not cause any issues in any games and was recommended to me by one of the most known overclockers in Europe (madness777), who strongly oposses most of the custom OS builds himself

What in particular made you decide on that image resolution to achieve a 23" viewport?
Image resolutions which aren't divisible by 8 should be avoided, as encoders/decoders & internal TCON boards don't play nice with certain values.
Though I cannot recall the source for the latter claim, I believe I've read that somewhere around here (by Chief or someone else)

I achieve this size by using No Scaling option in NVCP, which results in slightly lower than 23" size on 27(26.5)" 1440p OLED. I can make the res appear not only thanks to SRE, but also by connecting a capture card spoofed with this resolution through CRU. In such case, I assume NVIDIA drivers recognize that my monitor is perfectly compatible with 1242p480hz. UPD: Just figured that I can create a custom 16:9 resolution with DSC turned off and after turning it back on it will support 360/480hz DSC modes while using 1440p timings.

What specific games are you planning on recording?
HW: Legacy Flip (FSE) isn't a thing in D3D12 titles for ages now and from anecdotal conversations (no data), the quest of making it work in modern D3D11 games is a Pyrrhic victory.

I've mentioned it to specifically avoid this question since it is not relevant to the issue, but Legacy Flip was used in Black Desert Online just for another reference in my testing. The other games that used Independent Flip and the ones that I plan to record are Fortnite and Overwatch. In all games fps/frametimes were exactly the same compared to single Display mode and I haven't experienced any other issues which most likely means that the issue lies within the image output pipeline, not in the game's performance/settings

A “few milliseconds” is very inconcrete. How have you come to such a conclusion?
Does this added penalty you're referring to occur only when using a capture card?

Yes, it occurs only when actively using Cloning with a capture card. Does not occur in the Extended mode with a capture card or just having the capture card connected. Regarding how I've figured the rough latency estimate: believe it or not, and I mean it with no disrespect, using the sense of delay and between my nerves and perceived visual changes with my eyes. I know, many people are too old to believe such an outrageous claims, but when you've used to compete in tournaments for a living for years, especially in the high apm timings-based game like Fortnite, you can even notice 2-3ms difference in ping dev in the moments, let alone consistently apparent visual delay. Additionally, I have tested a bunch of different monitors in the past, therefore my experience with cloning on 480hz OLED feels very similar to using a ~360hz TN/IPS panel in terms of delay. However, I am not too naive to rely only on my senses, especially when it comes to fine tuning PC latency, so I will order OSLTT soon to troubleshoot the issue further and compare things like:
- DSC on vs Off,
- Passthrough vs 240/240 cloning vs 480/240 cloning,
- 480/240 cloning vs NVENC vs NDI

Thanks for all the info you've provided for me! I'll keep looking into this

[kyube]

Ghost Spectre Compact is the least intrusive custom OS out there. It doesn't remove anything that could potentionally compromise Windows functionability and only applies known to be safe changes to settings on install. It's proven to not cause any issues in any games and was recommended to me by one of the most known overclockers in Europe (madness777), who strongly oposses most of the custom OS builds himself

Argumentum ad verecundiam — argument from authority
I fail to see how a “overclocker” (a person who punches in numbers and hopes that the computer is able to sustain the specific combination of clock rate, voltage & temperature given a specific 2D/3D workload) can be used as some form of arguement of whether a custom ISO (a piece of software) is good or not?

You'd be surprised how many miniscule issues can be traced back to users using these type of custom ISOs, in which the user is not aware of each specific step taken in creating it.

NTLite or https://github.com/valleyofdoom/PC-Tuning (running .reg file after installation) is a safer approach, if you'd ask me.

I achieve this size by using No Scaling option in NVCP, which results in slightly lower than 23" size on 27(26.5)" 1440p OLED. I can make the res appear not only thanks to SRE, but also by connecting a capture card spoofed with this resolution through CRU. In such case, I assume NVIDIA drivers recognize that my monitor is perfectly compatible with 1242p480hz. UPD: Just figured that I can create a custom 16:9 resolution with DSC turned off and after turning it back on it will support 360/480hz DSC modes while using 1440p timings.

My question lies more in why you've chosen that particular resolution?
Why not resolutions such as:
2304x1296 (giving you a ~24.525" viewport)
2048x1152 (giving you a 21.6"viewport)
Both of these resolutions are a 16:9 aspect ratio & divisible by 8 in both width & height.

Yes, it occurs only when actively using Cloning with a capture card. Does not occur in the Extended mode with a capture card or just having the capture card connected. Regarding how I've figured the rough latency estimate: believe it or not, and I mean it with no disrespect, using the sense of delay and between my nerves and perceived visual changes with my eyes. I know, many people are too old to believe such an outrageous claims, but when you've used to compete in tournaments for a living for years, especially in the high apm timings-based game like Fortnite, you can even notice 2-3ms difference in ping dev in the moments, let alone consistently apparent visual delay. Additionally, I have tested a bunch of different monitors in the past, therefore my experience with cloning on 480hz OLED feels very similar to using a ~360hz TN/IPS panel in terms of delay.
However, I am not too naive to rely only on my senses, especially when it comes to fine tuning PC latency, so I will order OSLTT soon to troubleshoot the issue further and compare things like:
- DSC on vs Off,
- Passthrough vs 240/240 cloning vs 480/240 cloning,
- 480/240 cloning vs NVENC vs NDI

I'd suggest trying out a relatively stock ISO & a resolution divisible by 8.
Though, I'm somewhat ignorant & clueless as to what the best solution to this is problem is.

[Chief Blur Buster]

As far as I know, no video capture solution provides a beam raced API (software gaining access to the top of a refresh cycle, even while the video capture cable is still transmitting the bottom part of a refresh cycle)

It's impossible to have zero latency video capture without beam racing the raster.

I'd love to see a video capture card that is capable of this.

You can optimize though to just maybe 1 refresh cycle of latency (with difficulty) but you've got other overheads to worry about (codecs etc). A higher refresh rate = less video capture latency.

So get the highest Hz you can afford. 240Hz video capture of 60fps = 1/240sec frame capture instead of 1/60sec frame capture. Pixels transmit one at a time, top to bottom. You don't want to wait 1/60sec from the first through last pixel in a serialized video cable transmission. Higher Hz = faster transmission of frames even at lower game frame rates.

[Yukai]

As far as I know, no video capture solution provides a beam raced API (software gaining access to the top of a refresh cycle, even while the video capture cable is still transmitting the bottom part of a refresh cycle)

It's impossible to have zero latency video capture without beam racing the raster.

Hey, thanks for your reply!

Not sure if you have meant that or was my description not clear enough, but my issue is in regards to main-monitor latency during Display Cloning specifically, not the capture latency

[Yukai]

My question lies more in why you've chosen that particular resolution?
Why not resolutions such as:
2304x1296 (giving you a ~24.525" viewport)
2048x1152 (giving you a 21.6"viewport)
Both of these resolutions are a 16:9 aspect ratio & divisible by 8 in both width & height.

I see now, sorry for misunderstanding. I have chosen it as a smallest comfortable resolution to play Fortnite with for better performance. I've also tested 1440p, 1332p and 1278p resolutions Single Display vs Cloning using a capture card with identical results

[penguin1234567]

If you manage to find a solution or a way to minimise the added input delay completely I would love to know. I'm running a similar setup. I really wish there was a way to clone it without a performance hit to the Main PC. Even if it resulted in a delay to the content sent to the stream PC, it would be worth it. There is such a noticeable difference with the 2nd PC connected vs when its not.

I tried capping the refresh rate at 60hz on the capture card using a custom EDID as I thought potentially less bandwidth may reduce latency but I need to do proper testing in order to see if it helps.