Performance drop when gsync and sli are both used

[spartaman64]

Quote

A few days ago, one of our readers raised the idea that G-Sync had a performance penalty associated with using it. It was a bit surprising to see — Nvidia has previously assured me that there’s no performance penalty for using G-Sync — but the reports on the Nvidia forums include users with benchmark results comparing G-Sync On to G-Sync Off. There are extensive reports and they date back for months, including posts in the Nvidia reddit community.

Quote

Both G-Sync and SLI have major timing implications. Keeping the GPU and monitor running in sync takes time. Moving data from one GPU to another and then displaying the frames rendered by that same GPU takes time. If you target a 30fps frame rate, you need to deliver a new frame every 33.3ms. If you want to run at 60fps, that means you need a new frame every 16.6ms. If you target 120fps, that means you need a new frame every 8.3ms. At a certain point, you’ve got a very limited window with which to work That’s our theory, at least, after spending some time exploring this issue. To start, let’s look at the results user octiceps posted on the Nvidia forums:

Quote

Test Setup
There’s a substantial configuration difference, I suspect, between the G-Sync monitor we have available and the hardware these enthusiasts are using. The only G-Sync monitor I have is an Acer XB280HK, and while it’s 4K capable, it supports a maximum refresh rate of 60Hz. This means that by definition, I can’t use G-Sync in tests that hit frame rates higher than 60Hz. If this is a timing issue that only occurs above that frame rate, I definitionally won’t see it.

Quote

Deus Ex: Mankind Divided
Deus Ex: Mankind Divided shows a small-but-definite performance loss when G-Sync is enabled, as shown below. Interestingly, this gap appears only in DX11 mode. When we switched to DX12, it vanished.

Quote

Far Cry 5
Because we have to keep the frame rate under 60fps at all times to see the true impact of turning G-Sync off or on, we tested FC5 in four different resolutions: 4K, 4K+1.1x supersampling, 4K+1.3x supersampling, and 4K+1.5x supersampling. The results and the performance gap between having G-Sync on versus G-Sync off are shown below:

Quote

Hitman (2016)
Hitman wasn’t listed as an impacted title, but I took it for a spin anyway. I’m glad I did. It provided a useful third example in our tests. Here are our Hitman results in both DX11 and DX12. As with Far Cry, we tested Hitman with 1.3x and 1.5x supersampling to push the frame rate completely into our monitor’s G-Sync range. This time, we had to use supersampling at 4K to start with, which is why you don’t see the static “4K” test results.

 

Nvidia should really investigate this. I also heard from some people that they are getting a performance hit on a single gpu with g sync enabled but since I didn't find any official articles on that so take those reports with a grain of salt. I wonder if free sync has the same problems because if it doesn't then it would be really ironic since g sync cost a lot more money to implement.

https://www.extremetech.com/extreme/279259-nvidia-gpu-performance-craters-when-g-sync-sli-are-used-together

[Shimejii]

I dont use Gysnc in some games because i noticed some performance loss (980ti) such as ARK: Survival Evolved, and other big openworld games. If the game isnt heavily affected, and there is no screen tearing/stutters from it off, i just leave it off. Noticed this a few years ago when testing and put in a ticket saying exactly this, and how to easily reproduce it.

[Spotty]

Just looks like another reason why SLI isn't worth it.

[Curufinwe_wins]

I haven't noticed this in 2+ years of g-sync with sli 980ti on an ultrawide, but I did see this post and haven't throughly tested it yet. I probably should test and get back with results.

 

EDIT: I don't have the games @Spotty tested, and tbh, I really really don't feel like downloading a program to let me record the average in Witcher 3, so I only have two AAA games to report here. Sorry all. I don't really feel like going through my settings... so a comparison between my results (I took pictures of them if people really doubt me, but seriously here... save me some effort plox) and someone else isn't really a valid thing, but here is what I found:

 

Driver 416.34 SLI 980ti, Overclocked 5820k, Asus PG348Q

 

FPS listed is Ave/Min

 

Dragon Age Inquisition InGameBench w/ g-sync: 98.2/53.2 w/o g-sync 98.0/56.1. So no average difference, and **maybe** a minimum difference.

Shadow of Mordor InGameBench w/ g-sync: 119.10/60.27 w/o g-sync 118.67/60.92. No difference in average or minimum.

 

If someone has a game they want me to test, and wants to throw it upon me... feel free... lol

 

EDIT:EDIT: 

Second update. Found how to change the average interval in rtss. Changed it to 100 seconds. Tested W3 (on the random save file I loaded up) stable 67 fps G-sync on, 69 fps G-sync off. Tested AC:Syndicate (same thing, most recent save file skybox), 70 FPS on, 69 FPS off. 

 

Still have yet to see any statistically signficant difference yet with 980tis, in 4 games including one they explicitly mentioned. But these were pascal cards they tested, so maybe Maxwell isn't affected. IDK.

[Dabombinable]

TBH I still think that Nvidia's implementation of SLI is kind of bad since resources have to be mirrored across both cards.

[Spotty]

Decided to see if Gsync has any performance hit for single card systems.

 

Hypothesis

Does enabling Gsync hurt gaming performance in single card systems?

Testing Methodology

Run 3 benchmarks using the in game benchmark tool with Gsync Off and repeat the test 3 times with Gsync On.
Unless otherwise stated, all tests were conducted using the in-game benchmark tool and in-game benchmark results.

Final scores were taken from the Average FPS of each run, that was then averaged between the 3 runs to give a final average. In the case of GTA V, Average FPS of all scenes were averaged to give a single average score for each benchmark run.

 

System

CPU: Intel i7 6700k Stock (@4.0Ghz)

Motherboard: Gigabyte Z170x Gaming 5

RAM: 2x8GB 3000MHz G.Skill Ripjaws 5 (15-15-15-30)

GPU: Gigabyte Aorus Extreme GTX 1080ti (GV-N108TAORUS-X-11GD) [Driver Version: 416.34]

PSU: Corsair HX750W

Case: BeQuiet SilentBase 800

Cooler: Corsair H100i AIO

SSD: Samsung 840 500GB | HDD: Seagate Ironwolf 8TB + 2x Seagate Ironwolf 6TB |

Monitor: Acer Predator XB271HU 2560x1440p 144hz Gsync monitor.

 

 

Games Tested
GTA V - 2560x1440 Very High* settings (*See attached benchmark.txt files in additional notes for full graphical settings)
Deux Ex Mankind Divided DX11 Mode - 2560x1440 Very High Preset, 2x MSAA
Deux Ex Mankind Divided DX12 Mode - 2560x1440 Very High Preset, 2x MSAA

Metro Last Light Redux - 2560x1440 High Quality, SSAA On, Texture Filtering AF 4X, Motion Blur OFF, Tesselation High, Vsync Off, Advanced Phsyx On, Scene 1

 

Why the games were chosen
GTA V, Deus Ex DX11, and Deus Ex DX12 were all used in the extremetech article and were games that I owned for testing. Metro Last Light Redux benchmark was used as an additional test that I like to use due to its demanding nature.

 

 

Results
 

Quote

 

 

 

 

 

Conclusion
From the tests conducted on the specified system it appears that there is no difference between in game performance with Gsync enabled and Gsync disabled on single card systems.

 

 

Additional Notes (Full benchmark results included)

Spoiler

Limitations

• This only looks at the Average FPS for each test. Not calculated is the effect on minimum and maximum FPS, as well as frametime consistency.
• This only tests single card systems and does not compare results to SLI enabled multi-gpu systems. (If anyone wants to give me another 1080ti I'll be happy to test SLI as well )
• Only 3 games were chosen, with both DX11 and DX12 mode being tested in only one game.

 

 

Full Benchmark Results

 

All test results included in .zip archive: G sync tests.zip

 

Gsync spreadsheet full.ods

Gsync spreadsheet 1.xlsx

 

 

GTA V (Full Graphical Settings listed inside)
Gsync On - Benchmark-18-10-24-12-41-29.txt

Gsync On - Benchmark-18-10-24-12-46-03.txt

Gsync On - Benchmark-18-10-24-12-50-40.txt

Gysnc Off - Benchmark-18-10-24-12-56-11.txt

Gsync Off - Benchmark-18-10-24-13-02-26.txt

Gsync Off - Benchmark-18-10-24-13-06-58.txt

 

 

 

Metro Last Light Redux

Gsync On - 2018.10.24 14-06-39.zip

Quote

Gsync On Files (You will need to download all files and place within a folder together and open the Metro_report.html file, or download the .zip above)

Preset 0 - Run 2.svg

Preset 0 - Run 2.csv

Preset 0 - Run 1.svg

Preset 0 - Run 1.csv

Preset 0 - Run 0.svg

Preset 0 - Run 0.csv

metro_report.html

 

Gsync Off - 2018.10.24 13-54-33.zip

Quote

Gsync Off Files (You will need to download all files and place within a folder together and open the Metro_report.html file, or download the .zip above)

Preset 0 - Run 2.svg

Preset 0 - Run 2.csv

Preset 0 - Run 1.svg

Preset 0 - Run 1.csv

Preset 0 - Run 0.svg

Preset 0 - Run 0.csv

metro_report.html

 

 

Deus Ex Mankind Divided (images)

Spoiler

DX11 Gysnc On test 1

DX11 Gsync On test 2

DX11 Gsync On test 3

DX11 Gsync Off test 1

DX11 Gsync Off test 2

DX11 Gsync Off test 3

 

 

 

DX12 Gsync On test 1

DX12 Gsync On test 2

DX12 Gsync On test 3

DX12 Gsync Off test 1

DX12 Gsync Off test 2

DX12 Gsync Off test 3

 

 

 

 

@GabenJr Maybe LTT can do a video on Gsync with SLI vs Single Card to see if there is a performance hit, and if so how much. Could also do the same for AMD side with Freesync and crossfire. (I don't have a couple of spare 1080tis and Vega 64s lying around to test with )

[Arika]

SLI has never really been worth it for games. At most i would have thought that g-sync would get weird results trying to sync to two different GPUs but to have a fairly large frame drop seems like a huge bug.

[mr moose]

News just in, adding extra workload to an already unstable setup leads to performance losses.

 

Mind you, I would like to see testing from a single system comparing Gsync on and off in both single and SLI configurations.   If SLI with Gsync on still performs better than a single card with Gsync on then what do people have to complain about?

 

 

[i_build_nanosuits]

55 minutes ago, Spotty said:

Decided to see if Gsync has any performance hit for single card systems.

 

Hypothesis

Does enabling Gsync hurt gaming performance in single card systems?

Testing Methodology

Run 3 benchmarks using the in game benchmark tool with Gsync Off and repeat the test 3 times with Gsync On.
Unless otherwise stated, all tests were conducted using the in-game benchmark tool and in-game benchmark results.

Final scores were taken from the Average FPS of each run, that was then averaged between the 3 runs to give a final average. In the case of GTA V, Average FPS of all scenes were averaged to give a single average score for each benchmark run.

 

System

CPU: Intel i7 6700k Stock (@4.0Ghz)

Motherboard: Gigabyte Z170x Gaming 5

RAM: 2x8GB 3000MHz G.Skill Ripjaws 5 (15-15-15-30)

GPU: Gigabyte Aorus Extreme GTX 1080ti (GV-N108TAORUS-X-11GD) [Driver Version: 416.34]

PSU: Corsair HX750W

Case: BeQuiet SilentBase 800

Cooler: Corsair H100i AIO

SSD: Samsung 840 500GB | HDD: Seagate Ironwolf 8TB + 2x Seagate Ironwolf 6TB |

Monitor: Acer Predator XB271HU 2560x1440p 144hz Gsync monitor.

 

 

Games Tested
GTA V - 2560x1440 Very High* settings (*See attached benchmark.txt files in additional notes for full graphical settings)
Deux Ex Mankind Divided DX11 Mode - 2560x1440 Very High Preset, 2x MSAA
Deux Ex Mankind Divided DX12 Mode - 2560x1440 Very High Preset, 2x MSAA

Metro Last Light Redux - 2560x1440 High Quality, SSAA On, Texture Filtering AF 4X, Motion Blur OFF, Tesselation High, Vsync Off, Advanced Phsyx On, Scene 1

 

Why the games were chosen
GTA V, Deus Ex DX11, and Deus Ex DX12 were all used in the extremetech article and were games that I owned for testing. Metro Last Light Redux benchmark was used as an additional test that I like to use due to its demanding nature.

 

 

Results
 

 

 

 

 

Conclusion
From the tests conducted on the specified system it appears that there is no difference between in game performance with Gsync enabled and Gsync disabled on single card systems.

 

 

Additional Notes (Full benchmark results included)

  Reveal hidden contents

Limitations

• This only looks at the Average FPS for each test. Not calculated is the effect on minimum and maximum FPS, as well as frametime consistency.
• This only tests single card systems and does not compare results to SLI enabled multi-gpu systems. (If anyone wants to give me another 1080ti I'll be happy to test SLI as well )
• Only 3 games were chosen, with both DX11 and DX12 mode being tested in only one game.

 

 

Full Benchmark Results

 

All test results included in .zip archive: G sync tests.zip

 

Gsync spreadsheet full.ods

Gsync spreadsheet 1.xlsx

 

 

GTA V (Full Graphical Settings listed inside)
Gsync On - Benchmark-18-10-24-12-41-29.txt

Gsync On - Benchmark-18-10-24-12-46-03.txt

Gsync On - Benchmark-18-10-24-12-50-40.txt

Gysnc Off - Benchmark-18-10-24-12-56-11.txt

Gsync Off - Benchmark-18-10-24-13-02-26.txt

Gsync Off - Benchmark-18-10-24-13-06-58.txt

 

 

 

Metro Last Light Redux

Gsync On - 2018.10.24 14-06-39.zip

 

Gsync Off - 2018.10.24 13-54-33.zip

 

 

Deus Ex Mankind Divided (images)

  Reveal hidden contents

DX11 Gysnc On test 1

DX11 Gsync On test 2

DX11 Gsync On test 3

DX11 Gsync Off test 1

DX11 Gsync Off test 2

DX11 Gsync Off test 3

 

 

 

DX12 Gsync On test 1

DX12 Gsync On test 2

DX12 Gsync On test 3

DX12 Gsync Off test 1

DX12 Gsync Off test 2

DX12 Gsync Off test 3

 

 

 

 

@GabenJr Maybe LTT can do a video on Gsync with SLI vs Single Card to see if there is a performance hit, and if so how much. Could also do the same for AMD side with Freesync and crossfire. (I don't have a couple of spare 1080tis and Vega 64s lying around to test with )

that's cool and all...but you missed the part where it said ''In SLI''

you tested with a single card mate

[Spotty]

9 minutes ago, i_build_nanosuits said:

that's cool and all...but you missed the part where it said ''In SLI''

you tested with a single card mate

No, I didn't. You missed the part where I clearly labelled it as testing single card performance and even noted in the limitations that multi GPU set ups were not tested, nor were they the aim of the test.

 

1 hour ago, Spotty said:

Decided to see if Gsync has any performance hit for single card systems.

 

Hypothesis

Does enabling Gsync hurt gaming performance in single card systems?

 

1 hour ago, Spotty said:

Limitations

• This only tests single card systems and does not compare results to SLI enabled multi-gpu systems. (If anyone wants to give me another 1080ti I'll be happy to test SLI as well )

 

The purpose of my testing was to see if single card systems were affected by gsync being enabled. I use gsync when I'm gaming, so after reading the article - and noting that the articles results were for systems running sli - I was curious to see what, if any, effect gsync would have on my single card system.

[Curufinwe_wins]

33 minutes ago, Spotty said:

No, I didn't. You missed the part where I clearly labelled it as testing single card performance and even noted in the limitations that multi GPU set ups were not tested, nor were they the aim of the test.

 

The purpose of my testing was to see if single card systems were affected by gsync being enabled. I use gsync when I'm gaming, so after reading the article - and noting that the articles results were for systems running sli - I was curious to see what, if any, effect gsync would have on my single card system.

 

55 minutes ago, mr moose said:

News just in, adding extra workload to an already unstable setup leads to performance losses.

 

Mind you, I would like to see testing from a single system comparing Gsync on and off in both single and SLI configurations.   If SLI with Gsync on still performs better than a single card with Gsync on then what do people have to complain about?

 

 

 

Updated with my very limited testing. No difference at all for me.

[Spotty]

3 hours ago, Curufinwe_wins said:

: I don't have the games @Spotty tested, and tbh, I really really don't feel like downloading a program to let me record the average in Witcher 3, so I only have two AAA games to report here.

My internet's down at the moment so I can't do much testing, but I have shadow of Mordor and I might have Dragon Age (can't remember), so if anyone's interested I could perform those tests with a single card as well, though judging by your results and my results in other games I doubt there is much point..

 

I may even have some keys for Metro and Deus ex thanks to humble bundles so when my internet is back I'll check if I have any spare keys for you (but don't hold your breath, I normally gift games I already have pretty quickly).

Edit: Just checked, don't have any game keys for those games or others that have decent benchmarks, sorry.

Edited October 24, 2018 by Spotty

[Curufinwe_wins]

1 minute ago, Spotty said:

My internet's down at the moment so I can't do much testing, but I have shadow of Mordor and I might have Dragon Age (can't remember), so if anyone's interested I could perform those tests with a single card as well, though judging by your results and my results in other games I doubt there is much point..

 

I may even have some keys for Metro and Deus ex thanks to humble bundles so when my internet is back I'll check if I have any spare keys for you (but don't hold your breath, I normally gift games I already have pretty quickly).

Yeah, I can't see any immediate difference in W3, AC-Syndicate, or SW Battlefront, but without hard average data, I also don't feel comfortable claiming there is none.

[Curufinwe_wins]

18 minutes ago, Spotty said:

My internet's down at the moment so I can't do much testing, but I have shadow of Mordor and I might have Dragon Age (can't remember), so if anyone's interested I could perform those tests with a single card as well, though judging by your results and my results in other games I doubt there is much point..

 

I may even have some keys for Metro and Deus ex thanks to humble bundles so when my internet is back I'll check if I have any spare keys for you (but don't hold your breath, I normally gift games I already have pretty quickly).

Second update. Found how to change the average interval in rtss. Changed it to 100 seconds. Tested W3 (on the random save file I loaded up) stable 67 fps G-sync on, 69 fps G-sync off. Tested AC:Syndicate, 70 FPS on, 69 FPS off. 

 

Still have yet to see any statistically signficant difference yet with 980tis, in 4 games including one they explicitly mentioned. But these were pascal cards they tested, so maybe Maxwell isn't affected. IDK.

[Spotty]

7 minutes ago, Curufinwe_wins said:

Second update. Found how to change the average interval in rtss. Changed it to 100 seconds. Tested W3 (on the random save file I loaded up) stable 67 fps G-sync on, 69 fps G-sync off. Tested AC:Syndicate, 70 FPS on, 69 FPS off. 

That's within margin of error in my opinion.

 

 

3 hours ago, Curufinwe_wins said:

Shadow of Mordor InGameBench w/ g-sync: 119.10/60.27 w/o g-sync 118.67/60.92. No difference in average or minimum.

I don't have Dragon Age:Inquisition, but I tested Middle Earth Shadow Of Mordor. I wasn't going to bother posting the results, but I noticed that it mostly stayed above the Gsync limit of 144FPS throughout the test (averaging 178fps), so I thought it would be good to include the results to show if there is any difference with gysnc enabled when the FPS is beyond the Gsync limit (in a SINGLE CARD SYSTEM).

 

Still no difference.

Middle Earth: Shadow of Mordor - 2560x1440, Very High Preset

Quote

Gsync On  (test 1, test 2, test 3)
Average FPS: 178.87, 178.18, 177.82
Max FPS: 255.32, 267.41, 255.22
Min FPS: 93.93, 100.22, 103.46

Gsync Off (test 1, test 2, test 3)
Average FPS: 178.90, 178.72, 178.73
Max FPS: 260.28, 264.20, 261.36
Min FPS: 98.11, 99.60, 103.91

Quote

 

[leadeater]

@Spotty

I think it was Hardware Unboxed that did a big test on Gysnc performance, Gsync alone wasn't much of a difference but Gysnc + HDR was on Pascal but not Turing. HDR only I think wasn't a big difference either, it was mostly with both on and on Pascal (earlier cards likely too). Turing added better hardware support for high bit colours.

 

Edit: Found it

 

[Curufinwe_wins]

9 minutes ago, leadeater said:

@Spotty

I think it was Hardware Unboxed that did a big test on Gysnc performance, Gsync alone wasn't much of a difference but Gysnc + HDR was on Pascal but not Turing. HDR only I think wasn't a big difference either, it was mostly with both on and on Pascal (earlier cards likely too). Turing added better hardware support for high bit colours.

 

Edit: Found it

 

Interesting. And see Maxwell wouldn't have HDR support, so it wouldn't be bothered by the combination. Nice catch!

[leadeater]

1 minute ago, Curufinwe_wins said:

Interesting. And see Maxwell wouldn't have HDR support, so it wouldn't be bothered by the combination. Nice catch!

It wasn't quite the video I remembered it was, HDR test video but hey it has some Gsync data so .

[Spotty]

21 minutes ago, leadeater said:

@Spotty

I think it was Hardware Unboxed that did a big test on Gysnc performance, Gsync alone wasn't much of a difference but Gysnc + HDR was on Pascal but not Turing. HDR only I think wasn't a big difference either, it was mostly with both on and on Pascal (earlier cards likely too). Turing added better hardware support for high bit colours.

To be honest it wasn't something that I put a great deal of thought in to until I read this article. Was only after reading that I started to think about what, if any, performance drop was caused by using gsync (for my single card set up). There may still be other factors at play such as added latency, but I can't really test that...

Figured it would be easy enough for me to test it myself since it really just involves switching gsync on and off and looking at the difference in a couple of games.

 

I didn't really agree with the main articles testing methodology of using a 4k 60hz gsync panel and then enabling super sampling to force the refresh rate below 60fps. Just seemed like doing it that way was adding an unnecessary variable (supersampling) that could skew the results.

Also seems a little unrealistic to me that people with an sli Pascal setup running a 4k display would be using super sampling, (is 4k not enough!?!) At less than 60fps, but I guess people do all sorts of things...

 

Was interesting to see little difference between @Curufinwe_wins results with gsync on/off, compared to the huge variance in the main articles results... Could be a number of variables at play between those two examples however, given different generation of cards, different games tested, and different resolution/refresh rate.

 

To get the full picture we would need someone with access to a couple of Pascal cards, maybe even new rtx cards as well, to do a full range of tests with sli, without sli (single card), with gsync, without gsync, and at different resolutions and refresh rates.

Plus the same tests again for AMD/freesync.