G-Sync's 1ms Polling Rate: My Findings & Questions

Talk about NVIDIA G-SYNC, a variable refresh rate (VRR) technology. G-SYNC eliminates stutters, tearing, and reduces input lag.

Re: G-Sync's 1ms Polling Rate: My Findings & Questions

Postby jorimt » 01 Dec 2016, 19:45

kandor1978 wrote:It seems when you have gsync on and vsync off, as long as you dont hit the limit of your screen there is no tearing
But if you use a frame limiter that limits it to anything that will be hit then you get tearing
I have tested this with Battlefield 1
I get terrible tearing now that I set the fps to 100
If I let it be free there is no tearing whatsoever at any fps

That's an interesting observation, and it's possible the tearing is being lessened without an fps limiter in the same range, as the fps isn't being throttled by an outside source, and instead by the GPU. However, there's still going to be tearing both in that upper range just below the max refresh rate, as well as during frametime spikes/asset loads inside the minimum refresh range.

I did another test in Overwatch, raising my settings enough to lower my fps in the 130-144 range, and I still saw tearing without an fps cap. There was much less, and on a much smaller area of the bottom of the screen, but it was still there.

It's all about frametime variances.

Again, G-Sync wasn't originally designed for the G-Sync + V-Sync off combo. If you want to avoid the tearing for now, G-Sync + V-Sync on + fps limit is your only option.
Win 10 64-bit • Acer Predator XB271HU • ASRock Z87 Extreme4 • EVGA SuperNOVA 750W G2 • i7-4770k @4.2GHz • Hyper 212 Evo • EVGA GTX 1080 FTW GAMING ACX 3.0 • 16GB G.SKILL Sniper DDR3 @1866MHz • 256GB Samsung 850 Pro • 5TB Western Digital Black 7200 RPM
User avatar
jorimt
 
Posts: 171
Joined: 04 Nov 2016, 10:44

Re: G-Sync's 1ms Polling Rate: My Findings & Questions

Postby kandor1978 » 01 Dec 2016, 21:17

Yea

Back on using vsync on anyway now

Also did some more tests, and I thought well if fps fluctuates with gsync it will be ok anyway
But did run fraps benchmark several sessions I run in Battlefield 1 and found out:
2016-12-01 21:46:05 - bf1
Frames: 92263 - Time: 606968ms - Avg: 152.006 - Min: 109 - Max: 196

So as I might not have gotten to the heaviest parts in BF for whatever reason I did fps limit it through their engine to 100fps just
And game runs so smooth, never seen it run this smooth before and its much easier to hit people.
Tried to do same in csgo, but when fraps is running csgo refuses to start
I would like to do the same in csgo, as I have pretty bad stutter at some points in that game
Any ideas how to measure this?

Best,
Kandor
kandor1978
 
Posts: 44
Joined: 09 Aug 2016, 20:02

Re: G-Sync's 1ms Polling Rate: My Findings & Questions

Postby mminedune » 02 Dec 2016, 07:22

Just a heads up BF1 and other BF games use triple buffering when vsync is enabled. You can actually see difference with it on and off when panning around its visibly clearer when its disabled even in gsync range.

You can turn on and off on the fly to test. Open game console run RenderDevice.TripleBufferingEnable 0 to turn off.
Make fast pans around with mouse and you can see it makes screen look almost slightly blurry like LCD with higher pixel response time.

So yea if you use Gsync on Vsync on recommend disable triple buffering in the game.
I just add to config file for game so its always off.
mminedune
 
Posts: 30
Joined: 01 Feb 2014, 08:12

Re: G-Sync's 1ms Polling Rate: My Findings & Questions

Postby kandor1978 » 02 Dec 2016, 07:44

Yea noticed about tripple buffering already before and have added it to my user.cfg file to disable on all games

But thanks for caring to write it anyway :)

Best,
Kandor
kandor1978
 
Posts: 44
Joined: 09 Aug 2016, 20:02

Re: G-Sync's 1ms Polling Rate: My Findings & Questions

Postby jorimt » 02 Dec 2016, 20:49

kandor1978 wrote:Tried to do same in csgo, but when fraps is running csgo refuses to start
I would like to do the same in csgo, as I have pretty bad stutter at some points in that game
Any ideas how to measure this?

You can measure frametime fluctuations via MSI Afterburner.

It has built-in graphs for multiple performance operations. With it running in the background, you can simply play a couple rounds of CS:GO, and then inspect the frametime graph for spikes.
Win 10 64-bit • Acer Predator XB271HU • ASRock Z87 Extreme4 • EVGA SuperNOVA 750W G2 • i7-4770k @4.2GHz • Hyper 212 Evo • EVGA GTX 1080 FTW GAMING ACX 3.0 • 16GB G.SKILL Sniper DDR3 @1866MHz • 256GB Samsung 850 Pro • 5TB Western Digital Black 7200 RPM
User avatar
jorimt
 
Posts: 171
Joined: 04 Nov 2016, 10:44

Re: G-Sync's 1ms Polling Rate: My Findings & Questions

Postby lexlazootin » 12 Dec 2016, 09:50

I tested the G-Sync + V-Sync / G-Sync no V-Sync combo myself and they look pretty much the same in terms of input lag.

Image

Image

Image

Everyone like pictures :)

results (I know they are bad, you don't need to tell me):

https://docs.google.com/spreadsheets/d/ ... r2KATeB1hQ

Edit: i added some realistic results. (De_nuke 250fps~)
lexlazootin
 
Posts: 809
Joined: 16 Dec 2014, 02:57

Re: G-Sync's 1ms Polling Rate: My Findings & Questions

Postby jorimt » 12 Dec 2016, 11:33

lexlazootin wrote:I tested the G-Sync + V-Sync / G-Sync no V-Sync combo myself and they look pretty much the same in terms of input lag.

Thanks lexlazootin!

I've been doing a slew of input latency tests myself over the past week measuring middle screen (crosshair area) input latency with a 1000 fps camera and a mouse LED, and my finding have been largely the same.

I've done a total of 40 samples over 2 separate runs per scenario, and averaged out, there is virtually no input latency difference between V-Sync off, G-Sync + V-Sync off, and G-Sync + V-Sync on at the same framerate limit. I'll be posting charts and a download of the spreadsheet with all the numbers to my "G-Sync 101" thread, along with my updated findings on the G-Sync range.

I nearly have all the 144 Hz numbers completed, and will then move on to the 165, 100, and 60 Hz scenarios.
Win 10 64-bit • Acer Predator XB271HU • ASRock Z87 Extreme4 • EVGA SuperNOVA 750W G2 • i7-4770k @4.2GHz • Hyper 212 Evo • EVGA GTX 1080 FTW GAMING ACX 3.0 • 16GB G.SKILL Sniper DDR3 @1866MHz • 256GB Samsung 850 Pro • 5TB Western Digital Black 7200 RPM
User avatar
jorimt
 
Posts: 171
Joined: 04 Nov 2016, 10:44

Re: G-Sync's 1ms Polling Rate: My Findings & Questions

Postby Haste » 13 Dec 2016, 13:39

btw jorimt, whats your take on that video from LinusTechTips that has almost 800K views.
https://www.youtube.com/watch?v=MzHxhjcE0eQ

His results are really weird.

Any idea what happened there?
Monitor: Asus ROG Swift PG278Q
Haste
 
Posts: 273
Joined: 22 Dec 2013, 09:03

Re: G-Sync's 1ms Polling Rate: My Findings & Questions

Postby jorimt » 13 Dec 2016, 15:49

Haste wrote:btw jorimt, whats your take on that video from LinusTechTips that has almost 800K views.
https://www.youtube.com/watch?v=MzHxhjcE0eQ

His results are really weird.

Any idea what happened there?

My take? Plain old user error (he doesn't seem very organized, let alone experienced with input latency testing; things can get confusing real quick) and system differences that led to frametime variances.

For starters, they used a 960 fps camera. I can't verify whether or not they did the correct calculations.

Secondly, they used a game that had more variation from shot to shot than CSGO (what I'm currently testing with), which he complained he'd have to go to the trouble of using the fps_max command with -_-

Thirdly, the board with the LED was well below the crosshair area of the monitor. You want them relatively level for more accurate measurements. That, and he apparently had only four seconds to do multiple shots before the recording ran out. This isn't enough time to pace the shots properly, let alone get enough in. He did them so fast, I doubt he gave the system the time it needed to rest and reset for reaction to the left click (threads were constantly stimulated), which may have also thrown off the numbers. Thirdly, you need at least 20 samples per run, and at least 2 runs per scenario to get really accurate averages. They never specified how many they did, but I doubt it was that much.

I also found it funny he complained about it taking him 3 hours. I've been at this for a week, and am still going. I'll have 2 runs per scenario for a total of 40 samples per scenario, with what will be over 40 total scenarios.

Bottom line, I wouldn't take much stock in those results; too many unknowns.
Win 10 64-bit • Acer Predator XB271HU • ASRock Z87 Extreme4 • EVGA SuperNOVA 750W G2 • i7-4770k @4.2GHz • Hyper 212 Evo • EVGA GTX 1080 FTW GAMING ACX 3.0 • 16GB G.SKILL Sniper DDR3 @1866MHz • 256GB Samsung 850 Pro • 5TB Western Digital Black 7200 RPM
User avatar
jorimt
 
Posts: 171
Joined: 04 Nov 2016, 10:44

Re: G-Sync's 1ms Polling Rate: My Findings & Questions

Postby Sparky » 13 Dec 2016, 16:29

Haste wrote:btw jorimt, whats your take on that video from LinusTechTips that has almost 800K views.
https://www.youtube.com/watch?v=MzHxhjcE0eQ

His results are really weird.

Any idea what happened there?


What happened there is Linus is using a very tedious and time consuming method of testing, so it harmed the sample size(he had what, 4 samples per test?) He also didn't implement a proper control. (v-sync off, and a CRT as a comparison point on both GPUS, or at least SOME reference monitor that was tested on both systems). Then there's the question of how he controlled for framerate.

If you take the v-sync off 45fps results for example, Nvidia went 73 73 72 73 right? Looks super consistent right? Well no. That right there is proof the sample size is far too small, as at that framerate you get plus or minus 11ms just from the framerate alone.

Then there was the time they tried to test latency on the Steam Link: viewtopic.php?f=15&t=2831
Sparky
 
Posts: 474
Joined: 15 Jan 2014, 02:29

PreviousNext

Return to G-SYNC

Who is online

Users browsing this forum: Bing [Bot] and 2 guests