***THE BF4 BENCHMARKS THREAD***

[Rusty0611]

The game just isn't optimised yet - that's why the frame rate is lowish. It's as simple as that: running out of VRAM results in a complete loss of fluidity and it breaks the immersion completely. Simply put: the game becomes unplayable.

I've ran at 2030MB in BF3 on 680s with no discernible loss in performance compared to running at 1900MB. I think 2GB will be pushing it a bit if you game at 1440/1660 and have multi GPU. But if you don't meet both those criteria the normal rules apply. What is shown above by Matt doesn't constitute evidence either because the game ain't all that smooth even on my heavily OC'd 780.

You only need to see the 770 on its own having a higher minimum than the 690 to show that the game isn't working properly on nVidia cards at the mo neither is SLI producing scaling.

Also you're not taking into account the memory bandwidth angle - this will impact performance at super high resolution with AA applied. But VRAM amount? No way not from those numbers.

 

[Kaapstad]

No home... well maybe just a little.

I have just got my GTX 690s working @1600p maxed out.

Min 65

Ave 100

Max 145

 

[=XDC=FluphyBunny]

So do we have enough evidence yet to see if LtMatt's predictions of 2GB VRAM not being enough to run more than 2xMSAA with playable framerates?

Don't know as no one has played the actual game yet.

No reason to make decisions on a beta.

 

[Gregster]

I have just got my GTX 690s working @1600p maxed out.

Min 65

Ave 100

Max 145

@ 1600P or was it lower?

 

[LtMatt]

I'm siding with the tech site who have done proper benchmarks on all the cards at all the settings. The stark difference between the 690/770SLI results here tells you the 2gb buffer is struggling. You can see the low minimums on Ultra with high vram usage, vs medium settings with low vram usage while the minimum fps are much better as the buffer is not being filled.

Swe clockers come to the same conclusion i do. And so do all of their user comments.

When the resolution is screwed up also increases demands for more memory. Ultra lands according to GPU-Z of 2.7 GB, well above what eg GTX 690 or GTX 770 can offer. It may also be the explanation for why these cards suffer from noticeable frame rate drop in comparison with video cards that can provide more memory.

 

[Kaapstad]

@ 1600P or was it lower?

GTX 690s quad sli Maxed @1600p and they ran very smooth just like my Titans.

The mistake I was making earlier was running in a window. I reinstalled and ran fullscreen, sorted.

 

[LtMatt]

GTX 690s quad sli Maxed @1600p and they ran very smooth just like my Titans.

The mistake I was making earlier was running in a window. I reinstalled and ran fullscreen, sorted.

I'd love to see a gameplay video of you with fraps running on a 64 man conquest large server. Make it happen Kaap.

 

[Rusty0611]

I'm siding with the tech site who have done proper benchmarks on all the cards at all the settings. The stark difference between the 690/770SLI results here tells you the 2gb buffer is struggling. You can see the low minimums on Ultra with high vram usage, vs medium settings with low vram usage while the minimum fps are much better as the buffer is not being filled.

Swe clockers come to the same conclusion i do.

No - they say may.

True VRAM related slow downs are single digits or complete freezes. As I've said: if it really was VRAM then why has the 770 got a higher minimum than the 690?

The reason is the lack of general nVidia optimisation. You're interpreting the results based on what you want them to say as opposed to what they are actually saying. The memory bandwidth will struggle a lot more at ultra settings due to the application of MSAA as well.

 

[SS-89]

You're interpreting the results based on what you want them to say as opposed to what they are actually saying.

You expected any less?

 

[Kaapstad]

I'd love to see a gameplay video of you with fraps running on a 64 man conquest large server. Make it happen Kaap.

Unfortunately I am rubbish at the game and all you would see is me getting killed.

 

[LtMatt]

No - they say may.

True VRAM related slow downs are single digits or complete freezes. As I've said: if it really was VRAM then why has the 770 got a higher minimum than the 690?

The reason is the lack of nVidia optimisation. You're interpreting the results based on what you want them to say as opposed to what they are actually saying.

Because the 770 SLI has faster memory chips, samsung at 1750mhz as far as i know the 690 does not have these. The 770 SLI also has higher core clock speeds does it not?

In your opinion. Other tech sites seem to think that it can also present itself in other forums. Lower fps than normal, stuttering or hitching. I'm saying both are possible.

I was told before it was because SLI was not working, when it clearly is. Now its down to optimisation but we have people, setter for example, reporting 99% constant usage on both his Nvidia cards. Both sides no doubt need a bit more tweaking, but only 2gb cards are suffering massive fps drops at high res. Isn't that strange.

Unfortunately I am rubbish at the game and all you would see is me getting killed.

Do it anyway. Will be good to see your results.

 

[Silent_Scone]

You're going back in time a little bit there, Rusty? Not had VRAM cause that sort of performance in the last 8 years lol. Don't forget the way GPU's manage VRAM now is a bit less Model-T than that lol. It'll use what resources it has available to it, unless the game is broken you won't just hit a wall. You will see worse performance though

 

[LtMatt]

You're going back in time a little bit there, Rusty? Not had VRAM cause that sort of performance before. Don't forget the way GPU's manage VRAM now is a bit less Model-T than that lol. It'll use what resources it has available to it, unless the game is broken you won't just hit a wall. You will see worse performance though

Thank god. Frosty i could give you a kiss. Someone has finally seen the light.

 

[Rusty0611]

Because the 770 SLI has faster memory chips, samsung at 1750mhz as far as i know the 690 does not have these. The 770 SLI also has higher core clock speeds does it not?

No I said 770 not 770 SLI.

In your opinion. Other tech sites seem to think that it can also present itself in other forums. Lower fps than normal, stuttering or hitching. I'm saying both are possible.

I was told before it was because SLI was not working, when it clearly is. Now its down to optimisation but we have people, setter for example, reporting 99% constant usage on both his Nvidia cards. Both sides no doubt need a bit more tweaking, but only 2gb cards are suffering massive fps drops at high res. Isn't that strange

Have you got a link to the theory that running close to the wire on VRAM results in a performance decrease and the technical reasons for it? Because everything I've seen personally from testing 2GB cards at triple screen resolution would suggest that this is nonsense. The performance impact comes when you physically completely fill the VRAM. Until that point there is no performance impact.

The frame rate is dropping more because:

a) the game isn't that optimised for nVidia cards
b) the memory bandwidth is strained at ultra settings at 2560x1440

You're going back in time a little bit there, Rusty? Not had VRAM cause that sort of performance in the last 8 years lol. Don't forget the way GPU's manage VRAM now is a bit less Model-T than that lol. It'll use what resources it has available to it, unless the game is broken you won't just hit a wall. You will see worse performance though

No - less than a year ago in BF3 and MoHWF. Also tanked in Crysis 2 while modded, Hitman Absolution with 8x MSAA. Same effect each time. You aren't getting an average of 66 FPS if you don't have enough VRAM on your card (which is what the 690 got in that benchmark).

 

[sand_dune]

Swe OC clearly shows that 2GB limits frame rates even at 1080p but far not as significantly as at higher resolutions. So expect running out of Vram if ever you are on a huge map with lots of action or if BF4 graphics are slightly enhanced.

 

[=XDC=FluphyBunny]

Arguing over benchmarks for a game that's not finished

 

[Silent_Scone]

I'm all for being disproved but I think you'll find running close to the wire and your drivers working their arse off so that you don't are two separate things

 

[Gregster]

4-Way GTX 680 SLI Sony FW900 Surround Reaching VRAM (copy and paste and search youtube)

That has swearing in it (hence the lack of a link) but it is our Vega benching on youtube with 4 way 680's. You can skip to 7mins to see what happens when VRAM ran out for Vega. It is a slideshow.

I have nothing else to add to this VRAM debate and am out of any further VRAM discussions.

 

[LtMatt]

No I said 770 not 770 SLI.



Have you got a link to the theory that running close to the wire on VRAM results in a performance decrease and the technical reasons for it? Because everything I've seen personally from testing 2GB cards at triple screen resolution would suggest that this is nonsense. The performance impact comes when you physically completely fill the VRAM. Until that point there is no performance impact.

The frame rate is dropping more because:

a) the game isn't that optimised for nVidia cards
b) the memory bandwidth is strained at ultra settings at 2560x1440. Unless you're going to show me a result where the game



No - less than a year ago in BF3 and MoHWF. Also tanked in Crysis 2 while modded, Hitman Absolution with 8x MSAA. Same effect each time.

Aside from the links already provided and a ton of feedback from people using 2gb cards on various forums around the web and on battlelog. (yes i know )

I have here some benchmarks where the 2gb limit was not exceeded but resulted in the problems i mentioned above. Given the demands of bf4 its easily possible to see this similar behaviour happening again with much higher vram usage. Be it fps drops, stuttering, hitching or what not. It might not always mean bang 1fps.

Frame times in general are pretty good at 4K resolutions with the 2GB GTX 680 the lone outlier with some noticeable and significant spikes and jumps in performance. The fact that the 4GB variant does not exhibit that problem tells us that the 2GB frame buffer is just not enough to keep up 3840x2160 at these settings.

Both the Titan and the HD 7970 have the best / lowest frame time variance with the GTX 680 2GB coming in last. It is interesting to see the added frame buffer of the 4GB GTX 680 making a noticeable difference in potential stutter.

The GTX Titans in SLI have a very narrow and smooth band of frame times. The GTX 680 4GB frame times are also noticeably tighter and doesn't exhibit nearly as many spikes in frame times as the 2GB models do.

I find it very interesting that the GT 680 4GB cards from EVGA in SLI mode exhibit much less potential stutter than the 2GB cards, in line with the GTX Titans.

HardOCP also posted something exactly like this regarding Hitman Absolution. Can't remember the link or settings but it showed exactly the same as Swe clockers and pcper's conclusion.

 

[Rusty0611]

I'm all for being disproved but I think you'll find running close to the wire and your drivers working their arse off so that you don't are two separate things

I don't comprehend... can you elaborate?