Initial experience going 295x2 quadfire to 980Ti SLI at 4K

[Jacky60]

Just installed a pair of reference model Zotac 980ti and the extra memory makes a big difference. Gone completely is the hitching in Arma 3 when flying/benchmarking and even GTA5's AA looks cleaner. I've gained on Valley over 1000 points probably because their are no momentary pauses while memory is shuffled about under the hood. Overall very pleased with an 'upgrade' that was cheaper than the original purchase and anyone who thinks 4GB ram is 'enough' for 4k is mistaken imho, I've been using 4.4gb VRAM without changing any settings already today.

 

[jackstar7]

Appreciate the review. I'm just moving from a 7990 to a single 980ti and hoping to see a solid increase, plus the fewer issues being on a single gpu. I'll be on 3440x1440 and play very few demanding games, so I expect a nice change.

Then in the future... maybe one of those 3440x1440 g-sync monitors. Or jump back to AMD for freesync. Just need to advancements to keep on rolling.

 

[Jacky60]

Although GTA5 while using more memory with the 980Ti's seems to lack the middle distance graphical fidelity on other cars and pedestrians that the 295x2s gave me. I don't 'swoon' when playing like I did with the 295x2s because everything not in the immediate foreground doesn't look as polished -much more like watching a PS4 version. Will have to fiddle with settings some more as GTA 5 definitely doesn't look as nice on exact same game settings as if middle distance textures have been downgraded.

 

[railven]

Although GTA5 while using more memory with the 980Ti's seems to lack the middle distance graphical fidelity on other cars and pedestrians that the 295x2s gave me. I don't 'swoon' when playing like I did with the 295x2s because everything not in the immediate foreground doesn't look as polished -much more like watching a PS4 version. Will have to fiddle with settings some more as GTA 5 definitely doesn't look as nice on exact same game settings as if middle distance textures have been downgraded.

Check your AA settings. I found that the blurring nature of MSAA on edges is inferior to straight up downsampling.

I've pretty much stopped using MSAA techniques if I got the brute force for downsampling. Image is clearler and there is no blur when turning which looks even more impressive on my IPS monitor.

 

[Headfoot]

and anyone who thinks 4GB ram is 'enough' for 4k is mistaken imho, I've been using 4.4gb VRAM without changing any settings already today.

ugh do we have to do this again?

4GB "being enough" depends on which card is using it. 295x2 uses MORE memory for CrossFire overhead than a single 290x. And the Fury X uses less memory due to overhead optimizations as has been shown in various reviews. Because your cards have 6GB VRAM the game will reserve more of that ram for itself via pre caching (and any game made in the last couple of years is going to work this way). 980 Ti 4.4GB VRAM use DOES NOT MEAN it requires 4GB VRAM absolutely for all cards. This incorrect hypothesis needs to die.

 

[railven]

ugh do we have to do this again?

4GB "being enough" depends on which card is using it. 295x2 uses MORE memory for CrossFire overhead than a single 290x. And the Fury X uses less memory due to overhead optimizations as has been shown in various reviews. Because your cards have 6GB VRAM the game will reserve more of that ram for itself via pre caching (and any game made in the last couple of years is going to work this way). 980 Ti 4.4GB VRAM use DOES NOT MEAN it requires 4GB VRAM absolutely for all cards. This incorrect hypothesis needs to die.

Well, I guess he's specifically talking about GTA5, which even Ryan found that 4GBs *could* be a problem in that game.

Adding a second card would alleviate the processing power that Ryan refers to, but it won't alleviate the memory pool. And if there is more overhead for CFX vs single, that makes it worse for Fury X in GTA5 scenario. I guess, you could just keep draw distance to low.

http://www.anandtech.com/show/9390/the-amd-radeon-r9-fury-x-review/7

Meanwhile with GTA5 we can break the R9 Fury X, but only at unplayable settings. The card already teeters on the brink with our standard 4K “Very High” settings, which includes 4x MSAA but no “advanced” draw distance enhancements, with minimum framerates well below the GTX 980 Ti. Turning up the draw distance in turn further halves those minimums, driving the minimum framerate to 6fps as the R9 Fury X is forced to swap between VRAM and system RAM over the very slow PCIe bus.

 

[ShintaiDK]

4GB not enough at 4K. Not surprising.

 

[amenx]

Although GTA5 while using more memory with the 980Ti's seems to lack the middle distance graphical fidelity on other cars and pedestrians that the 295x2s gave me. I don't 'swoon' when playing like I did with the 295x2s because everything not in the immediate foreground doesn't look as polished -much more like watching a PS4 version. Will have to fiddle with settings some more as GTA 5 definitely doesn't look as nice on exact same game settings as if middle distance textures have been downgraded.

Nvidia default settings are not very good imo. A couple I always change once a driver is installed:

Anistropic filtering = x16 (not application controlled)
Texture filtering = High quality over quality.

These alone make a big difference to me.

 

[RussianSensation]

4GB not enough at 4K. Not surprising.

Yet another totally unsubstantiated comment by you being passed off as a fact. To conclude the above for Arma 3, we would need a much deeper analysis such as 970 SLI vs. 980 SLI vs. 980Ti OC. If 970 SLI or 980 SLI is actually better or even at least as good as 980 Ti OC at 4K, 4GB of VRAM may not even be a big factor. Alternatively, Tri-fire 290X 8GB vs. Tri-Fire 290X 4GB. Similarly, 390X Quad-fire vs. 295X2 CF.

Right now the stutters are more than likely related to Quad-fire scaling/driver issues in that lower frames are perceived as stutters since the 3rd and 4th GPUs aren't scaling linearly -- the nature of multi-GPUs.

Normally VRAM bottlenecks aren't just minor stutters but massive drops in FPS, and can often even be a slide-show in fact. More likely than not, 980Ti SLI could easily be producing superior frame times than 295X2 CF in this title at 4K. We cannot just conclude that VRAM is the fundamental factor that's allowing for smoother gameplay at 4K in this game.

Furthermore, VRAM required doesn't mean dynamic VRAM used as per MSI AB/Precision X.

A quick Google search shows that in Arma 3 @ 4K even a single 970 with its 3.5GB of VRAM has superior frame times (I.e, it maintains a flatter frame time variance up to and including 99th percentile) than a 290X, despite lower FPS:

http://www.tomshardware.com/reviews/nvidia-geforce-gtx-980-970-maxwell,3941-9.html

That alone casts some doubt into the assumptions made by the OP. It's looking like Maxwell + NV drivers just do a better job in this game at 4K than Hawaii cards, irrespective of VRAM. This game could still very well be impacted by 4GB limit but comparison of 295X2 CF vs. 980Ti SLI is not enough to make a definitive conclusion on this hypothesis.

The end result though is the OP could add a 3rd and a 4th 980Ti to his system and end up with more horsepower to push battlefield draw distance further at which point the benefits of > 4GB could manifest themselves. Also, he will most definitely have superior performance since even a single 980Ti OC is more or less as fast as a 295X2. Sounds like a good upgrade since 295X2s don't always scale well.

 

[Hitman928]

Although GTA5 while using more memory with the 980Ti's seems to lack the middle distance graphical fidelity on other cars and pedestrians that the 295x2s gave me. I don't 'swoon' when playing like I did with the 295x2s because everything not in the immediate foreground doesn't look as polished -much more like watching a PS4 version. Will have to fiddle with settings some more as GTA 5 definitely doesn't look as nice on exact same game settings as if middle distance textures have been downgraded.

Interesting you noticed that. There's someone at the overclockers uk forum that posted comparison videos of a TitanX and FuryX and people noticed the same in BF4

I hadn't said anything about it because I was seeing how it played out before bringing it up, but it seems like you are experiencing the same thing.

 

[railven]

Yet another totally unsubstantiated comment by you being passed off as a fact. To conclude the above for Arma 3, we would need a much deeper analysis such as 970 SLI vs. 980 SLI vs. 980Ti OC. If 970 SLI or 980 SLI is actually better or even at least as good as 980 Ti OC at 4K, 4GB of VRAM may not even be a big factor. Alternatively, Tri-fire 290X 8GB vs. Tri-Fire 290X 4GB. Similarly, 390X Quad-fire vs. 295X2 CF.

Right now the stutters are more than likely related to Quad-fire scaling/driver issues in that lower frames are perceived as stutters since the 3rd and 4th GPUs aren't scaling linearly -- the nature of multi-GPUs.

Normally VRAM bottlenecks aren't just minor stutters but massive drops in FPS, and can often even be a slide-show in fact. More likely than not, 980Ti SLI could easily be producing superior frame times than 295X2 CF in this title at 4K. We cannot just conclude that VRAM is the fundamental factor that's allowing for smoother gameplay at 4K in this game.

Furthermore, VRAM required doesn't mean dynamic VRAM used as per MSI AB/Precision X.

A quick Google search shows that in Arma 3 @ 4K even a single 970 with its 3.5GB of VRAM has superior frame times (I.e, it maintains a flatter frame time variance up to and including 99th percentile) than a 290X, despite lower FPS:

http://www.tomshardware.com/reviews/nvidia-geforce-gtx-980-970-maxwell,3941-9.html

That alone casts some doubt into the assumptions made by the OP. It's looking like Maxwell + NV drivers just do a better job in this game at 4K than Hawaii cards, irrespective of VRAM. This game could still very well be impacted by 4GB limit but comparison of 295X2 CF vs. 980Ti SLI is not enough to make a definitive conclusion on this hypothesis.

The end result though is the OP could add a 3rd and a 4th 980Ti to his system and end up with more horsepower to push battlefield draw distance further at which point the benefits of > 4GB could manifest themselves. Also, he will most definitely have superior performance since even a single 980Ti OC is more or less as fast as a 295X2. Sounds like a good upgrade since 295X2s don't always scale well.

So, basically what you're saying is "4GB is not enough for 4K on AMD hardware."

 

[Headfoot]

So, basically what you're saying is "4GB is not enough for 4K on AMD hardware."

No dude come on. Stop trying to twist words. It seems to be not enough to max out GTAV, maybe.

For the most part the reviews have shown it IS enough, for now.

The real question is: Will Fury X have enough power that the 4GB becomes relevant? My guess is that in future games where it runs out of VRAM it will already have unplayably low frame rates, so the point is moot. This is how it works for existing games where it runs out of RAM (GTAV). It probably does factor into a SLI vs CF argument where you have more processing power but the same amount of VRAM.

 

[Headfoot]

Interesting you noticed that. There's someone at the overclockers uk forum that posted comparison videos of a TitanX and FuryX and people noticed the same in BF4

I hadn't said anything about it because I was seeing how it played out before bringing it up, but it seems like you are experiencing the same thing.

Huh, that's very strange behavior... It's very interesting OP also independently noticed this.

 

[railven]

No dude come on. Stop trying to twist words. It seems to be not enough to max out GTAV, maybe.

My comment was more a tongue in cheek joke because it was the second time in this thread that someone said something along the lines of "AMD drivers have more over head" so it seems where a GTX 970 can get away with 3.5GBs a R9 290X couldn't with 4GBs.

For the most part the reviews have shown it IS enough, for now.

The real question is: Will Fury X have enough power that the 4GB becomes relevant? My guess is that in future games where it runs out of VRAM it will already have unplayably low frame rates, so the point is moot. This is how it works for existing games where it runs out of RAM (GTAV). It probably does factor into a SLI vs CF argument where you have more processing power but the same amount of VRAM.

We're not talking about a single card in this thread. That was never the issue. This is specifically about CFX/SLI setups. You negate your own post.

 

[Gikaseixas]

Interesting you noticed that. There's someone at the overclockers uk forum that posted comparison videos of a TitanX and FuryX and people noticed the same in BF4

I hadn't said anything about it because I was seeing how it played out before bringing it up, but it seems like you are experiencing the same thing.

Would this cause a FPS penalty on the Fury X platform?

 

[Hitman928]

Would this cause a FPS penalty on the Fury X platform?

Assuming it's an actual render quality issue, then the TitanX could be getting a fps boost. I'm not sold on it yet, but seeing Jacky's observations in this thread made me give it more consideration. I remember when Mantle first launched on BF4 there was some render issues as well, but in the end, it was a bug and the card performed the same after it was fixed.

 

[railven]

Nvidia default settings are not very good imo. A couple I always change once a driver is installed:

Anistropic filtering = x16 (not application controlled)
Texture filtering = High quality over quality.

These alone make a big difference to me.

Assuming it's an actual render quality issue, then the TitanX could be getting a fps boost. I'm not sold on it yet, but seeing Jacky's observations in this thread made me give it more consideration. I remember when Mantle first launched on BF4 there was some render issues as well, but in the end, it was a bug and the card performed the same after it was fixed.

Post above might explain it, especially since he came from a clean install.

I don't use NV Panel as much as AMD CCC (frankly I hate NV Panel) but I do remember at least on AMD CCC sometimes installing a new driver would reset my CCC options (especially for profiles I created).

Could be NV is pulling another "ole optimize" trick to get the FPS crown? I'm sure reviewers would have caught this.

 

[ShintaiDK]

Reading Anandtechs Fury X review. its quite clear VRAM is an issue. AMD have dedicated 2 people to sit and try hand managing memory for demanding games.

Which is why for Fiji, AMD tells us they have dedicated two engineers to the task of VRAM optimizations. To be clear here, there’s little AMD can to do reduce VRAM consumption, but what they can do is better manage what resources are placed in VRAM and what resources are paged out to system RAM. Even this optimization can’t completely resolve the 4GB issue, but it can help up to a point. So long as game isn’t actively trying to use all 4GB of resources at once, then intelligent paging can help ensure that only the resources that are actively in use reside in VRAM and therefore are immediately available to the GPU when requested.

 

[Atomic Playboy]

Although GTA5 while using more memory with the 980Ti's seems to lack the middle distance graphical fidelity on other cars and pedestrians that the 295x2s gave me. I don't 'swoon' when playing like I did with the 295x2s because everything not in the immediate foreground doesn't look as polished -much more like watching a PS4 version. Will have to fiddle with settings some more as GTA 5 definitely doesn't look as nice on exact same game settings as if middle distance textures have been downgraded.

GTA V has an issue with the ingame Anisotropic Filtering on Nvidia cards; basically, it doesn't work right. Disable AF in the ingame menu and force 16x AF through the Nvidia control panel (make sure to choose "Override application setting"). Also set the filtering quality to "High Quality." That should fix the texture filtering issues (as shown in this comparison shot).

 

[sm625]

What effect does that "issue" have on benchmarks? This seems as shady as ever for nvidia...

 

[Hitman928]

Post above might explain it, especially since he came from a clean install.

I don't use NV Panel as much as AMD CCC (frankly I hate NV Panel) but I do remember at least on AMD CCC sometimes installing a new driver would reset my CCC options (especially for profiles I created).

Could be NV is pulling another "ole optimize" trick to get the FPS crown? I'm sure reviewers would have caught this.

This very well may be, but then the question becomes (as sm625 said), how does it effect performance? It could be that the card is still doing the "work" but it's not getting to the output correctly like with the Mantle issue early on, or it could be giving inflated fps for lowered detail. Haven't seen anything convincing yet, but is something I think should be investigated. Also don't think reviewers would have really caught it by now, in today's market there's lots of pressure to push out the reviews so I don't think they really pay as much attention as they use to unless they get "tips" of certain things to test.

 

[Atomic Playboy]

Considering the performance cost of Anisotropic Filtering in modern GPUs is basically 0, I can't imagine Nvidia would intentionally degrade their filtering, as it's not going to result in any measurable gain in FPS. It's more likely a driver issue or a problem specific to filtering within a single game. Regardless, a jpg from a compressed YouTube video is probably the least effective way to measure such discrepancies as there are so many other quality issues that can be introduced along that path. Let's see some uncompressed screenshots from the source.

 

[manimal]

Although GTA5 while using more memory with the 980Ti's seems to lack the middle distance graphical fidelity on other cars and pedestrians that the 295x2s gave me. I don't 'swoon' when playing like I did with the 295x2s because everything not in the immediate foreground doesn't look as polished -much more like watching a PS4 version. Will have to fiddle with settings some more as GTA 5 definitely doesn't look as nice on exact same game settings as if middle distance textures have been downgraded.

I noticed an IQ reduction going from my 6950 to gtx980 too. Chalked it up to my looking for problems.

Hmm.

 

[ShintaiDK]

Interesting you noticed that. There's someone at the overclockers uk forum that posted comparison videos of a TitanX and FuryX and people noticed the same in BF4

I hadn't said anything about it because I was seeing how it played out before bringing it up, but it seems like you are experiencing the same thing.

Should post the source.
https://www.youtube.com/watch?v=FBijzEQ6x6g

Looking on the FPS numbers. Its quite clear they dont run the same. Because DX11 benchmarks shows them much closer than this.

However I suspect the Fury X is running it with the Mantle API. This would explain the performance delta.

And be careful trying to compare frame by frame since they are not synced. This gets obviously clear in the end of the video. And its not taken in the same places either since its a manual run.

 

[KaRLiToS]

Quad Hawaii to dual GM200 is a total downgrade. Sorry.

It's been like 5 years since people noticed an obvious lower Image Quality on Nvidia's side.

I noticed it when I went from Quad 7970 to tri GTX 780 to Quad R9 290x. Same thing when I went from HD 6970 trifire to Sli GTX 680.

Considering the performance cost of Anisotropic Filtering in modern GPUs is basically 0, I can't imagine Nvidia would intentionally degrade their filtering, as it's not going to result in any measurable gain in FPS. It's more likely a driver issue or a problem specific to filtering within a single game. Regardless, a jpg from a compressed YouTube video is probably the least effective way to measure such discrepancies as there are so many other quality issues that can be introduced along that path. Let's see some uncompressed screenshots from the source.

I can't imagine Nvidia releasing a 3.5GB whinning card as a 4GB card...but they did.


We need a third competitor because at this point, I questionned Nvidia honesty or AMD competence. Both are getting quite ridiculous. Fury x is a total fail to my eyes and everwhere on forums we see threads of Nvidia releasing a GTX 980ti for ONLY 650$ or 700$ for the classifed. Are they freaking serious? Only 650$ is way too expensive...but people still buy the 3% faster Titan X for 1000$.

I have no interest in both companies as of right now.