nvidia driver frame latency issues

[piesquared]

This seems to be a taboo subject that has gone ignored. Here's one example:

http://alienbabeltech.com/abt/viewtopic.php?f=6&t=27699#p81461


1920x1080 for the GTX 680


HD 7970 GHz at 1920x1080 (60Hz)


The rest of the tests are in the link. I'm curious why websites are focusing on the perception that this is an isolated AMD occurrence.

 

[SirPauly]

Imho,

Awareness to both may create smoother experiences from both - the bigger picture. With more sites that investigate, offer reviews, articles, create discussion and go beyond just frame-rate is welcomed.

 

[raghu78]

its no coincidence that techreport left out BF3 in their fall GPU article "HD 7950 vs GTX 660 Ti". the timing of that article right in the middle of the holiday shopping season and ignoring a top AAA title like BF3 which is still very popular on multiplayer servers on the internet and which has had mumerous expansion packs like Back to Karkand, Close Quarters, Armored Kill, Aftermath and the upcoming Endgame (march 2013) clearly points to the fact that TR was trying to show that this was a AMD only issue which its not.

 

[RussianSensation]

Ever since Catalyst 12.11s, HD7970Ghz edition has owned GTX680 in real world gameplay in BF3: http://www.youtube.com/watch?v=qR3ewLMbywY

There are plenty of games where NV cards also exhibited more stuter. In FC3, GTX690 is much more stuttery compared to HD7970 GE in CF.
http://videos.pcgameshardware.de/video/12558/Far-Cry-3-Mikroruckeln-in-Slow-Motion

The issue of frame times has plagued GPUs for many generations. Even Fermi cards had it but people ignored it because it was understood that in some games AMD's cards had better frame times and in others NV did. It is true that NV also took steps to minimize frame times by moving from Fermi to Kepler due to hardware frame metering but it still doesn't mean that their cards are smoother in most games either as many attempted to claim by generalizing SLI vs. CF and extrapolating that to single GPUs. I think we have to look at it on a game-by-game GPU vs. GPU basis. Which actually makes sense since you buy a GPU based on the games you intend to play

Once the TR articles came out, the author pointed out issues in some titles that AMD cards have, but completely omitted focusing on any games where NV cards had these issues. Due to their random selection of games which happened to include games that are more prone to stutter on AMD cards, it appeared as if only AMD cards had frame times latency issues. Then the fanboys came out of the woodwork claiming AMD's cards are less smooth "in most games" when in reality both AMD and NV cards suffered from frame times issues depending on the game you played. If you included games like Sleeping Dogs, Hitman Absolution, Sniper Elite V2, Dirt Showdown, BF3, Arma II, Metro 2033, Risen 2, etc. then we might as well have made a sweeping conclusion that all GTX600 cards suffer from frame latency time issues. That's not correct either since the choice of games you include can dramatically change the overall conclusion. The other thing TR did by misleading readers is pitting a stock HD7950 V2 against a GTX660Ti but then people generalized this to GTX680 to HD7970GE and all other NV vs. AMD comparisons. This is flat out incorrect.

It's one thing to discuss performance trends in recent titles, or specific demanding titles, and it's completely another to make generalized claims across all HD7000 GPUs for most games as TR insinuated. The positive thing is AMD was aware of some of these frame times issues in Skyrim, etc. Hopefully soon enough they will rewrite the memory management subsystem of HD7000 cards and improve frame times.

The point is it's great that TR encouraged GPU makers to focus more on frame times for smoother gameplay, even if it meant a reduction in average FPS, and also for readers they pointed out that FPS is not the only measurement we should be paying attention to. However, the one-sided view they took towards AMD suffering from this problem was simply poor journalism.

 

[SirPauly]

its no coincidence that techreport left out BF3 in their fall GPU article "HD 7950 vs GTX 660 Ti". the timing of that article right in the middle of the holiday shopping season and ignoring a top AAA title like BF3 which is still very popular on multiplayer servers on the internet and which has had mumerous expansion packs like Back to Karkand, Close Quarters, Armored Kill, Aftermath and the upcoming Endgame (march 2013) clearly points to the fact that TR was trying to show that this was a AMD only issue which its not.

So, you're also implying that TechReport's goal was to be disingenuous and a calculated move to give nVidia a competitive advantage?

 

[raghu78]

The point is it's great that TR brought up a point that GPU makers should focus on frame times for smoother gameplay, even if it means a reduction in average FPS, and also pointed out that FPS is not the only measurement we should be paying attention to. However, the one-sided view towards AMD suffering from this problem was simply poor journalism.

thats very well said. :thumbsup:

 

[piesquared]

So, you're also implying that TechReport's goal was to be disingenuous and a calculated move to give nVidia a competitive advantage?

Is there any other explanation? techreport has been doing these tests for what, around 4 months now? Every one of them have been headlines focusing on AMD's frame latency issues, with nary a whisper of nv's driver latency problems.

 

[SirPauly]

I am grateful for "smooth" awareness because I felt focus was needed, specifically with multi-GPU.

This "poor" journalism has helped with more awareness:

http://techreport.com/blog/24247/as-the-second-turns-further-developments

 

[BallaTheFeared]

To be fair they've only shown what they've been testing, since there are many games and many different game engines it's unreasonable to believe it's possible to test them all. These are people not super humans capable of dredging through countless games, in countless scenarios on many different systems.

The video put up by the Sweden(?) website that showed the 7970 looking awful in BL2 and Skyrim, also showed it looking smoother than the 680 in BF3. Oh it seems RS linked it.

Now when you watch that video, can you see the difference? I can, I can see the 680 isn't as smooth.

Same can be said of BL2, http://www.youtube.com/watch?v=W7kuXuuSAFk

As well as Skyrim: http://www.youtube.com/watch?v=CqTmx1V47WE

In both those videos I see a direct correlation to the frametime graphs and the real world output. In both those videos is quite obvious the 7970 GHz is having issues delivering a smooth frame experience.

I heard Nvidia's newest beta has addressed the problem in BF3, however I haven't seen any reviewers test it yet. So far apoppin and people on GeForce seemed to have been the only ones really talking about it.

However the key in these frametime tests is that they expose the fallacy of relentlessly linking fps charts, as FPS tells us absolutely nothing about the overall smoothness of the game. The only way to know if the fps are representative of actual smoothness is to run a frametime test, as such the fps charts themselves are invalid unless they can be backed up with frametimes, which makes them pointless. The overall hope here is that through knowledge we can garner a better experience for ourselves as gamers, because in the end no matter whose "side" you're on neither company is on "yours".

 

[BrightCandle]

The BF3 problems seemed to get introduced around October/November (can't remember the exact release) and there is no doubt its a big problem. More so than the 7000 series one because turning down settings to reduce the GPU load doesn't improve things. At least with the 7000 series problem you can turn settings down and it normally improves or goes away, that is not true with the 680 problem in BF3.

But its also an isolated case where nvidia clearly introduced a problem and has resolved it a few months later. My test on the latest beta shows its gone. The 7000 series seems to be struggling with a large list of modern games, and it has done ever since it was released. Back in the early days of the card it had issues with new games, it was heavily dependent on getting a new driver release and profile before the games would run well. Its stuttered for well over a year.

So there is a difference in the persistence and wide aspect of the problem between the two manufacturers. But awareness is the key, we need everyone testing for smoothness instead of FPS average so that more games and rigs get tested so that both manufacturers get the message that this matters. TechReport managed to finally get AMD to take it seriously and test and fix the problem. Something we know Nvidia has made statements about in the past and has been doing things about this for some years already. It was good journalism in that it focused on the company that needed to change not the one that had made a single mistake. To me this isn't an NVidia v AMD issue, its about changing 2 decades of graphics performance testing to focus on something much closer to how humans perceive motion. It might be true today that AMD has problems but they will fix it of that I have no doubt.

 

[boxleitnerb]

There are plenty of games where NV cards also exhibited more stuter. For instance in FC3, GTX690 is much more stuttery compared to HD7970 GE in CF.
http://videos.pcgameshardware.de/video/12558/Far-Cry-3-Mikroruckeln-in-Slow-Motion

Not really comparable since
a) different fps
b) broken SLI profile (there was an update after this review)

With the new SLI profile, it plays very well. What other plenty of games are you speaking of? Where are frametime measurements in Sleeping Dogs, Hitman Absolution, Sniper Elite V2, Dirt Showdown, BF3, Arma II, Metro 2033, Risen 2 etc that you mentioned? No offense, but if you claim something, you better back it up by facts.

 

[AnandThenMan]

If there is one thing RussianSensation does, is back up his arguments with facts.

I am grateful for "smooth" awareness because I felt focus was needed, specifically with multi-GPU.

This "poor" journalism has helped with more awareness:

http://techreport.com/blog/24247/as-the-second-turns-further-developments

How about we get some awareness without the obvious cherry picking and ignoring of the same "awareness" regardless of brand.

 

[boxleitnerb]

Usually yes. But in this case, where is the data?

 

[badb0y]

Not really comparable since
a) different fps
b) broken SLI profile (there was an update after this review)

With the new SLI profile, it plays very well. What other plenty of games are you speaking of? Where are frametime measurements in Sleeping Dogs, Hitman Absolution, Sniper Elite V2, Dirt Showdown, BF3, Arma II, Metro 2033, Risen 2 etc that you mentioned? No offense, but if you claim something, you better back it up by facts.

I hope you have some Vaseline handy lol.

inb4giantRSpost

 

[Tweak155]

Can someone expand on what impact "frame latency issues" has on user experience? What would the user notice / not notice?

 

[GlacierFreeze]

Can someone expand on what impact "frame latency issues" has on user experience? What would the user notice / not notice?

Game play would have a noticeable stutter/jerkiness (aka not smooth) regardless of high FPS.

 

[f1sherman]

If you could explain their defiance on testing any issues NVidia has (and of course making it a headline and bringing it to the attention of the community) with frame latency in games, i'd be interested in reading it.

[edit]

Instead of derailing this thread, i'll make a new one.

Latency is a whole another issue that AMD has yet to address:

If AMD wanted to remove frame latency they failed miserably[...]

http://benchmark3d.com/microstutter-case-study-skyrim-exam-day

This seems to be a taboo subject that has gone ignored. Here's one example:


The rest of the tests are in the link. I'm curious why websites are focusing on the perception that this is an isolated AMD occurrence.

Looks like Nvidia fixed this even before you have had the chance to make your EQUALIZER thread:

310.90 drivers in BF3

313.95 drivers in BF3

Far Cry 3 310.90 drivers

Far Cry 3 313.95 drivers

https://forums.geforce.com/default/...3-95-are-fantastic-for-sli-and-multi-screens/

 

[Obsoleet]

Is there any other explanation? techreport has been doing these tests for what, around 4 months now? Every one of them have been headlines focusing on AMD's frame latency issues, with nary a whisper of nv's driver latency problems.

I would agree if they weren't using frame latency issues as a reason for wholeheartedly recommending AMD over Intel.
I posted this is another thread on APU comparisons (http://forums.anandtech.com/showthread.php?t=2288404&highlight=)

http://techreport.com/review/22932/amd-a10-4600m-trinity-apu-reviewed/12

Yep. Trinity definitely leads the pack here. The average FPS figures might fool you into thinking Ivy Bridge is almost as fast, but a glance at our frame latency curve will show otherwise. The Ivy system's frame times rise sharply for the last 10% or so of frames.

I actually trust TR more than AT for being unbiased. Reading AT for over 10 years, IMO I think AT stinks more of pro-Intel and pro-Nvidia bias if anything (doesn't just pertain to video cards).
I haven't actively searched for TR investigations into Nvidia frame latency, if they haven't done anything I'd find that strange. After all, it's just going to gather more page hits.

 

[Granseth]

Latency is a whole another issue that AMD has yet to address:

If AMD wanted to remove frame latency they failed miserably[...]

http://benchmark3d.com/microstutter-case-study-skyrim-exam-day



Looks like Nvidia fixed this even before you have had the chance to make your EQUALIZER thread:
(...)

Have you read the the article you are referencing? And at the same time saying nVidia has fix latency.

As far as I can se both AMD and nVidia has made huge improvements but not removed latency issues?

 

[f1sherman]

Have you read the the article you are referencing?

Yup. Read and understood.

And at the same time saying nVidia has fix latency.

Is that a question? Can you rephrase it please?

 

[Granseth]

Yup. Read and understood.

lets complete the sentence you quoted then: ... and removed most of the microstutter.

So the only reason to post half a sentence was to troll.

 

[f1sherman]

lets complete the sentence you quoted then: ... and removed most of the microstutter.

So the only reason to post half a sentence was to troll.

Sorry?

I quoted only the part relevant for what I was addressing (latency),
which I clearly indicated with [...]. I also gave a link to full article.

 

[RussianSensation]

Where are frametime measurements in Sleeping Dogs, Hitman Absolution, Sniper Elite V2, Dirt Showdown, BF3, Arma II, Metro 2033, Risen 2 etc that you mentioned? No offense, but if you claim something, you better back it up by facts.

Is this news to you? HD7970GE handily GTX680 in most of those games regarding FPS and smoothness. We've been over this for months and you keep denying it. The FPS are so much lower on the GTX680 in those titles (except BF3 where GTX680 is close on avg), there isn't even any point in doing frame time measurements. Go check. Do you really need frame times in games like Dirt Showdown or Sleeping Dogs to know GTX680 is a slideshow under those fps? Just like I wouldn't recommend someone interested in the best BL2 or AC3 gaming experience to get a HD7970GE, I wouldn't recommend someone get a GTX680 over HD7970GE for any of the titles listed above. Heck, gamer's own experiences back this up. The first 4 titles I listed are all compute heavy and run much slower on a 680. GTX680 chokes in Arma 2 with AA and Metro 2033 has been show to be much faster on AMD cards, including HD7970GE vs. 680. This is news to you again? BF3 frame times can already be found in this thread. My point is I could have picked 10 games and written a TR piece where GTX680 would have consistently exhibited inferior frame times based on my game selection. I would never go and generalize that GTX680's drivers are completely broken. Similarly I can find 10 games where GTX680 would have smoother frame times. Same story with HD6970 vs. GTX570, etc.

Buy the GPU based on the games you play the most. Don't generalize that HD7970GE has worse frame times than GTX680 in 100s of games like you have been doing.

<snip>

Excellent post Balla! :thumbsup:

I wouldn't say FPS measurements are completely irrelevant though. Frame times are giving us additional information. With frame times and FPS, I'd prefer a gaming experience at 45-60 fps over a 30 fps one. Both are useful measurements.

It seems there is just no escape from having some games simply performing much better on NV (BL2, AC3, Project CARS), and others performing better on AMD. The best way is to accept this and spend time reseraching which games run smoother based on your gaming prefences. The alternative is just to go all out for 2 Titans

 

[96Firebird]

^Well that was a disappointment...

 

[SirPauly]

If there is one thing RussianSensation does, is back up his arguments with facts.

How about we get some awareness without the obvious cherry picking and ignoring of the same "awareness" regardless of brand.

I'm for any awareness -- didn't have a problem with the OP's nVidia example -- because with investigations, reviews, articles and discussions that go beyond frame-rate may translate into smoother products from AMD and nVidia..

AMD is improving as their drivers are maturing as is nVidia; and a new smooth V-sync feature for SLi:

http://nvidia.custhelp.com/app/answers/detail/a_id/3283