[gamegpu] Far Cry 4 performance

[amenx]

Patch 1.6 out...

Added hotkey for toggling the HUD on/off (F11)
Fixed an issue when the game loaded incorrect save from Uplay Cloud
Fixed an issue with performance drop for Crossfire Multi-GPU
Fixed an issue with broken shadows in the northern region
Fixed various Controls issues
Fixed various Graphic issues
Fixed various IGE issues
Fixed various UI issues
Fixed various Uplay/Steam issues
Memory usage reduced

 

[Freddy1765]

Still stuttering with high texture resolutions while driving?

 

[Magic Carpet]

*some progress*

A very nice article on Far Cry 4 performance has been out for a few days, over at HardOCP. The game has been applied the latest patch (1.6), and gives some ideas on what to expect from various video cards.

Overall, the latest generation Maxwell GPUs perform better in Far Cry 4. With all the Kepler GPUs we tested, GTX 780, GTX 770 and GTX 760 these all performed rather poorly. We had to sacrifice many in-game quality settings just to find a playable performance. With the GTX 970 and GTX 980 though we were able to maintain higher settings.

We also found that some AMD GPUs are even superior to Kepler GPUs. For example the AMD Radeon R9 280X and R9 270X performed a lot better than we had anticipated. Both GPUs beat the NVIDIA competition in this game. However, the opposite is true for the R9 290 and R9 290X. Those two GPUs cannot at all keep up with NVIDIA GTX 970 and GTX 980 in Far Cry 4. These are underpowered in this game and are beaten squarely in the face by the GTX 980 and GTX 970.

When it comes to the high-end GPUs, again GTX 970 and GTX 980 is the way to go. GTX 970 is probably the best value, which is able to beat out the AMD Radeon R9 290X even. Overclock a GTX 970 and it should do even better. You will be able to play at 1440p with the GTX 970, but you won't be able to max out the game. Even the GTX 980 cannot max out the graphics settings. This game will need even more powerful GPUs to max it out at 1440p, or SLI can do that as well right now.

The GTX 780 and 770 are bad for this game in terms of performance. You'd be better off with the AMD Radeon R9 280X which will offer a better experience compared to either of those video cards. Compared to the GTX 760 the AMD R9 270X would also be better.

In terms of low-end GPUs you'll actually find a bit faster performance with the R7 260X versus the GTX 750 Ti. However, both are still very slow and you'll end up sacrificing either in-game settings by a lot, or having to reduce resolution to 720p to maintain good image quality in the game. These two video cards are just too slow for this game, you'll need to aim for a better GPU to enjoy this game.

Running Far Cry 4 at 4K is very demanding, and only GeForce GTX 980 SLI allows a decent experience. It still cannot max out the game settings at 4K, but right now it comes the closest to doing so since CrossFire is broken. There is no way you'll enjoy this game at 4K with a single video card.

As for me, I finished playing the campaign yesterday at 75% on my Asus Top GTX 670 2GB (4770K @ 4.3 Ghz HT Off + 8GB RAM + Intel SSD). I had everything on Ultra (Enhanced God Rays, Soft Shadows and SMAA) at 1680x1050. I usually don't play with AA, but SMAA made the scenery look so much better, so I had to keep it for a small performance hit. For the most part I have averaged between about 45 to 60 FPS (Vsync On with Triple Buffering) with very occasional frame drops to about ~25 FPS. Have not experienced any stuttering whatsoever (maybe SSD helped with that), input lag or whatever other people have been reporting, everything was smooth and bug-free. Not a single exit to desktop, perfect. Can't say how it compares to my 270 CF as CF support still non-existant, but I agree with the article above, 270 runs fairly well. Also, the game proved to like 4 cores with HT disabled. The game ran fairly well on my other rig with X5650 equipped w/ MSI R9 290 GAMING (friend is busy testing it). That one also felt better with 6 cores @ 4.0 Ghz and HT off, but the AMD card didn't feel as smooth at times, but clearly more powerful.

Here is my GamerProfile config, if it helps anybody (%userprofile%\Documents\My Games\Far Cry 4)

<RenderProfile AntiAliasingMode="4" UseTrippleBuffering="1" VSyncMode="1" UseMotionBlur="1" SSAOLevel="3" EnableNvidiaPCSS="0" GodRaysLevel="2" FurLevel="1" TreeTessellationLevel="1" Version="1" AlphaToCoverage="1" ResolutionX="1680" ResolutionY="1050" Quality="custom" QualityEditor="editor_ps3" Fullscreen="0" Borderless="1" UseD3D11="1" WidescreenLetterbox="0" UseWidescreenFOV="1" FOVScaleFactor="1.5" AspectRatio="3" VSync="1" VSyncWindow="0" RefreshRate="60" DisableLoadingMip0="0" GPUMaxBufferedFrames="0" ShowFPS="0" Brightness="1" Contrast="1" CalibrationScreensShown="0" GammaRamp="1" AllowAsynchShaderLoading="1" SafeFrameAreaWidth="0.85" SafeFrameAreaHeight="0.85">
<CustomQuality>
<quality ResolutionX="1680" ResolutionY="1050" OtherQuality="low" EnvironmentQuality="ultrahigh" AntiPortalQuality="default" PortalQuality="default" PostFxQuality="ultrahigh" TextureQuality="ultrahigh" WaterQuality="ultrahigh" DepthPassQuality="low" VegetationQuality="ultrahigh" TerrainQuality="ultrahigh" GeometryQuality="ultrahigh" LightingQuality="high" ShadowQuality="softshadow" ShadowCinematicQuality="low" EditorQuality="default" Hdr="1" HdrFP32="0" ReflectionHdr="1" EnableVertexBinding="1" id="custom" />
</CustomQuality>
</RenderProfile>

Some of the settings above may read Low or High, but these are not selectable from the game menu, so I left them unchanged. Maybe they give additional eye candy but I've never bothered to check and compare. Really, not that picky. Just so you know.

I recommend playing this stupid game to everybody. It has good moments and lots of fun. Also as it appears, it does have an alternative ending, just be patient.

 

[AtenRa]

*some progress*

A very nice article on Far Cry 4 performance has been out for a few days, over at HardOCP. The game has been applied the latest patch (1.6), and gives some ideas on what to expect from various video cards.

Hell, at 1080p R9 280X is faster than GTX780

 

[Magic Carpet]

@AtenRa

And 280X is really 7970 Ghz Edition, which is a 2012 year design.

GCN just performs :thumbsup:

EDIT: No idea, why they wasted resources on that pos, a.k.a. 285. Might as well named it "The Empire Strikes Back" or something. Just ridiculous. They should stop mimic Nvidia in every way, for Christ's sake or indeed, they deserve to lose this game with a strong hand. Fools.

 

[DiogoDX]

What bothers me is that no site is investigating this like they love to do massive articles with any AMD problem.

 

[jpiniero]

Hell, at 1080p R9 280X is faster than GTX780

I'm not surprised at this point since it seems like nVidia has all but stopped with the cheats for Kepler.

 

[Magic Carpet]

I'm not surprised at this point since it seems like nVidia has all but stopped with the cheats for Kepler.

Maxwell is the current tech that needs as much attention, so why wasting resources elsewhere. But I am not competent enough to make comments about it, though. What we care about is the end result, and with the latest games, Kepler does not perform well enough. Sad, but true.

 

[soccerballtux]

Hell, at 1080p R9 280X is faster than GTX780

why is this?

 

[jpiniero]

Maxwell is the current tech that needs as much attention, so why wasting resources elsewhere. But I am not competent enough to make comments about it, though. What we care about is the end result, and with the latest games, Kepler does not perform well enough. Sad, but true.

It's just jarring how quickly they dropped Kepler though.

 

[caswow]

Maxwell is the current tech that needs as much attention, so why wasting resources elsewhere. But I am not competent enough to make comments about it, though. What we care about is the end result, and with the latest games, Kepler does not perform well enough. Sad, but true.

i would guess that there are way more people with kepler than with maxwell at this point.

 

[cmdrdredd]

I'm not surprised at this point since it seems like nVidia has all but stopped with the cheats for Kepler.

I think your choice of words is wrong. Cheats?

Increasing performance via driver updates is not cheating.

i would guess that there are way more people with kepler than with maxwell at this point.

Kepler is going on 3 years old now though isn't? I'm not saying that someone who bought a 780ti six months ago shouldn't expect some driver enhancements (I imagine there are always trickle down gains) but Nvidia is a business who wants money. Selling new GPUs makes them money. Releasing drivers doesn't. I expect when AMD releases their next lineup that there will stop being so many drivers focuses on increased performance on those 280 and 290 cards.

 

[el etro]

What bothers me is that no site is investigating this like they love to do massive articles with any AMD problem.

Seconded.

 

[RussianSensation]

The situation with Kepler is much different than normal obsolescene. First of all 780Ti beat 970 at launch and was only 5-8% slower than a 980. As soon as 970/980 came out and new games launched, Kepler performance completely tanked by massive amounts. If these newer games were too demanding for outdated architecture, then R9 280X/290/290X would also tank but this didn't happen at all. Not only that but NV charged $650 for 780 in May 2013, $1000 for Titan and $699 for November 2013 780Ti. These chips are not even 2 years old and their performance given their original standing and pricing is terribad! I could inderstand 670/680 due to low memory bandwith and 2GB of VRAM but there is no excuse to throw 780/780Ti under the bus considering Maxwellis a tiny fraction of the overall market. NV never treated GF8, GTX200 or Fermi this way.

A lot of people kept saying for years how AMD cuts frequent driver updates earlier than NV but what in NV's case their frequent driver updates seem to not even care for older generations. If this is a new trend, it's very alarming since not everyone wants to upgrade to new cards every 2 years. Ironically, Kepler performs poorly in FC4 and Watch Dogs, both GW titles. This is unacceptable when Maxwell does just fine in those games. Sounds like planned obsolescene by NV...

 

[Magic Carpet]

It's just jarring how quickly they dropped Kepler though.

Yeah, well... it's a fast-paced industry, **** happens. I've been stung before (hint: 3dfx V5).

i would guess that there are way more people with kepler than with maxwell at this point.

True. Hopefully, people will learn the lesson.

I expect when AMD releases their next lineup that there will stop being so many drivers focuses on increased performance on those 280 and 290 cards.

If it's still GCN, there is a bloody good chance all the existing AMD hardware might benefit from the future driver optimization as well. That's another reason, why buying 290's makes a lot of sense today.

Nvidia is a business who wants money. Selling new GPUs makes them money. Releasing drivers doesn't.

This, pretty much. They want all those 670/680's to upgrade to current tech. 780/780ti is just collateral damage, imo.

 

[gamervivek]

@AtenRa

And 280X is really 7970 Ghz Edition, which is a 2012 year design.

GCN just performs :thumbsup:

EDIT: No idea, why they wasted resources on that pos, a.k.a. 285. Might as well named it "The Empire Strikes Back" or something. Just ridiculous. They should stop mimic Nvidia in every way, for Christ's sake or indeed, they deserve to lose this game with a strong hand. Fools.

Read the detailed review though, the 285 does the highest level of godrays better than 280x.

The AMD Radeon R9 285, being based on newer technology did better than we thought, but ultimately is held back in performance we think due to its memory bandwidth. What is interesting is that we were able to run with the highest Godray setting of Enhanced on the R9 285. We found that selecting between "Volumetric Fog" and "Enhanced" did not change performance much at all. There was only a few FPS difference. It is as if Enhanced Godrays just don't burden this GPU much.


This actually makes a lot of sense. You see, the R9 285 (Tonga) received some tessellation performance improvements in that new iteration of GCN. You can refer back to this table where we outlined all the improvements in each iteration of GCN. You will notice that improved tessellation is one of the upgrades with Tonga. Since Godrays are tessellation based, this newer architecture does very well with it compared to any other AMD GPU. Even R9 290X cannot touch how well R9 285 is doing tessellation now. The R9 285 also uses a new color compression scheme which also improves performance.

And selling the better dies to apple.

 

[Magic Carpet]

Read the detailed review though, the 285 does the highest level of godrays better than 280x.

Some things 285 does better, tessellation has been vastly improved. Better use of memory bandwidth (compression). Possibly, a few more things. But it doesn't perform. It should have beaten any 280x class card. At least, it would have lived up to the name. Instead all it does, is customer confusion. And to top it off, 2 gigs of vram. Come on, AMD, what you been smoking? :whiste:

A card to forget, really. Maybe AMD realized, 2 gigs is "fast enough" for 285. In which case, it should have been 275 really. Would have collected less hate, most definitely.

This is with enhanced "God Rays".

 

[ShintaiDK]

Hell, at 1080p R9 280X is faster than GTX780

Quite a massive difference between 280X and 285 as well. Looks odd. Unless the 280X numbers are off, AMD should pay more attention to the few people that bought the 285.

But some sites have radically different numbers. I wonder wht there is so large difference between them.

This also looks wierd.

 

[skipsneeky2]

Is there even a recent hardware review with recent FC4 numbers?The whole kepler conspiracy as true as it appeasr due to the Op and its original review numbers that are 3 months old does sound incredibly crazy too.Crazy like Gary Busey yelling at his t.v in a commercial.

I can't take the 3 month old numbers off the Op as even serious.

 

[AtenRa]

Is there even a recent hardware review with recent FC4 numbers?The whole kepler conspiracy as true as it appeasr due to the Op and its original review numbers that are 3 months old does sound incredibly crazy too.Crazy like Gary Busey yelling at his t.v in a commercial.

I can't take the 3 month old numbers off the Op as even serious.

http://www.hardocp.com/article/2015/01/07/far_cry_4_video_card_performance_review/1#.VLF66nvcBJM

Wednesday, January 07, 2015

 

[ShintaiDK]

There isnt any ingame benchmark is there? I just took a look and couldnt find any. That might explain why some sites got very different numbers and performance is basicly all over the place.

 

[skipsneeky2]

http://www.hardocp.com/article/2015/01/07/far_cry_4_video_card_performance_review/1#.VLF66nvcBJM

Wednesday, January 07, 2015

Very interesting,surprised anyone would take the time to even do a review so long after a game comes out.Pretty admirable if you ask me.

 

[Hitman928]

There isnt any ingame benchmark is there? I just took a look and couldnt find any. That might explain why some sites got very different numbers and performance is basicly all over the place.

That bench was done back in November when the game was still having issues on AMD cards. From the website they advise people with AMD cards to wait for updates before purchasing the game. Patches and a new driver fixed it shortly thereafter. The hardOCP benches were done with latest drivers and the latest patch.

 

[cmdrdredd]

Can anyone see the difference between TXAA and SMAA? I looked at Nvidia's comparisons and couldn't tell if one blurred more than another.

 

[Magic Carpet]

There isnt any ingame benchmark is there? I just took a look and couldnt find any. That might explain why some sites got very different numbers and performance is basicly all over the place.

This is a major "problem" with modern games. Few come with any sort of benchmarking capabilities. It's not like in the Quake times, where people used to record demos and what not. So, for a quick bench, you would just type in:

timedemo 1
map demo1.dm2

To get accurate numbers and compare them with your friends rig. Easy peasy.

Now really, was it that difficuld to record a cooperative demo of Far Cry 4, 5 mins long or so, with massive gunfire and what not? Sigh. Instead, we have to rely on, god knows what techniques review sites are using to get those numbers from. It's merely just a guess game, since you can't replicate those conditions. They can't just be ran multiple times over and average the result. So yeah, you have to take every benchmark with a huge grain of salt. Allow -/+1-10% to human/stats error.

Can anyone see the difference between TXAA and SMAA? I looked at Nvidia's comparisons and couldn't tell if one blurred more than another.

This?

You can drag the slider. TXAA gives you a bit more detail. Note the blue skies. I am fine with SMAA, though. Good trade-off, quality versus perf.

That bench was done back in November when the game was still having issues on AMD cards. From the website they advise people with AMD cards to wait for updates before purchasing the game. Patches and a new driver fixed it shortly thereafter. The hardOCP benches were done with latest drivers and the latest patch.

Yeah, but 750 TI giving better minimums than 780? I call that bs.

If you are any serious review web site, you must update your results, so they don't just hang out there with dinosaur old drivers, confusing people.