[H] Battlefield 3 Open Beta Performance and Image Quality

[Keysplayr]

I don't suppose you've seen DICE's GDC DX11 presentation? It goes into some detail about how they're streaming resources, though it doesn't fully answer the question.

No, but thanks for the link. Gives me something to read on the way to work.
I'll check it out.

 

[Grooveriding]

Beware if you are on low-bandwidth connections. Here are some screenshots of different games with AA on and with AA off. I mention settings for each set. They are followed by a bunch of BF3 screenshots with different settings.

It's near impossible to get shots in the exact same spot for BF3 because when you change video settings they don't apply until you quit and restart the game. I had to use an empty server, and every time you rejoin the server it positions you in a different spot, and you can't move from that area.

Just look at memory usages though and you will get the idea. Really what I am seeing here is that AA uses more memory inf BF3 compared to other games. That overall BF3 just uses a lot of memory to begin with. Also, this puts to bed the idea that games use all the VRAM you have if they need it or not. Just not the case, every game makes different usage of VRAM, but none use as much as BF3 does. The only one that comes close is Crysis and my Crysis is heavily modded with upgraded textures.


Here is Crysis on settings above Very High using mods and texture packs.

This first shot is with 4xAA - 1.4GB usage

Without AA - 1.287GB Usage

Starcraft all on Ultra with 4xAA and 2xTRSAA - 1.185GB Usage

Witcher 2 All Maximum settings

With Ubersampling On - .934GB Usage

Ubersampling Off - .859GB Usage

BFBC2 Everything on Maximum setttings - These are interesting because this is the first iteration of FB and it also shows big memory usage from turning on AA, but not huge memory usage to begin with regardless of AA like I see in BF3.

With 8xMSAA - 1.084GB Usage

With no AA - .758GB Usage

Finally here is BF3 with some different settings


This is on all Ultra, HBAO, no post AA, 4x MSAA - 1.502GB Usage

This is on all Ultra, HBAO, no post AA, 2x MSAA - 1.414GB Usage

This is on all Ultra, HBAO, no post AA, No AA - 1.319GB Usage

Can only add 10 images per post, the links to the final two images.


This is on all Medium, HBAO, no post AA, 4x AA - 1.42GB Usage

http://i.imgur.com/9dipG.jpg

This is on all Medium, HBAO, no post AA, No AA - 1.245GB Usage

http://i.imgur.com/1COFE.jpg



BF3 is pretty monstrous on wanting VRAM. The only other game I've played that is near it is the first Crysis and that is with it heavily modded from the stock game.

No other game I play makes use of all the VRAM available just because it is there.

BF3 is making use of this VRAM, why so much ? When I run the ultra settings with 4xAA it hitches, backing off from the AA corrects it. First game I have played that causes hitching with 4xAA at my resolution.

I'm going with VRAM hungry.

 

[SirPauly]

I don't think it is ram issues because even the 6870 sends the GTX 560 TI packing and begging for frame-rate.

 

[happy medium]

To add some more evidence to our discussion, this guy has 2 3gb gtx580's with surround and is using, yes you guessed it.... about 2.7gb of video memory.


Quote:
"Been playing the beta with nvidia surround 6080*1200 and the game likes a lot the video ram.
Its taking 2.7 gbs vram at max settings with all on except hboa
It runs between 25-to 35 fps on metro map on the outside, inside the metro i get like 40 to 50 fps
i was able to get in to Caspian and the map was giving better fps than metro map, and felt smoother than metro. same settings. (on a good server, cause the majority of the caspian servers where bugged or lag)

I guess it will get better on the release, the thing that sucks of the game so far are the bugs .... just hopping to get most annoying ones fixed on the release. "


http://hardforum.com/showpost.php?p=1037830019&postcount=59

We just need a web site to do a 6950 1gb vs 1 6950 2gb performance /image comparison to see what the extra memory does if anything.

SO far NO one has said they are experiencing any performance problems or hitching except Groover ...........
The game will not use more than the memory you have so its impossible, it must be a beta//driver problem for him.

 

[Epsilon-Zero]

I dont think [H] has manipulated any of the benchmarks and those that shout shinanigans need to chill out.

Those that frequent [H] know that they usualy always gets the best AMD/ATI cards to test and there is certainly no mention of STOCK cards to get their result. Simmilarly if they had the best Nvidia cards Im sure we may have had the opposite result.

The model numbers may be apples to apples, but the hardware configurations are in doubt. They used what they had available just like every other hardware reviewer out there.

Combined with Pro Nvidia benchmarks from others, I think the benchmark from [H] proves that it doesnt matter which GPU camp you choose, what matters is the hardware configuration.

 

[Grooveriding]

SO far NO one has said they are experiencing any performance problems or hitching except Groover ...........
The game will not use more than the memory you have so its impossible, it must be a beta//driver problem for him.

Yeah, the game cannot use more memory than you have. It's why I get hitching when I try to use 4xAA at 2560x1600 with ultra.

I haven't seen anyone running this game on a 1GB card with all in-game settings cranked, because they can't, or rather they get an unpleasant experience. People are running it with some settings on ultra and turning others off with 1GB cards. Anyone who has been running the maximum settings is using a GTX 580/480, 2GB 6950 or a 6970.

No one has said you cannot run this game on a 1GB card. You cannot run BF3 maxed out on a 1GB card, but you can certainly run it. I've posted ample evidence of memory usage scenarios in demanding games contrasted to BF3.

I just would like to know more about why BF3 wants so much VRAM, I'm past determining if it does or not, that is proven.

 

[happy medium]

You cannot run BF3 maxed out on a 1GB card

Groover read the thread, especially the last 10 posts on page one.

Read this post by Keys and toyota experienced the same thing I am.
YOU WILL NOT RUN OUT OF MEMORY with 1gb cards.
http://forums.anandtech.com/showpost.php?p=32363272&postcount=91

I have no idea what the game does with the extra memory but your game will not lose performance if you run at max settings @ 1080p with a 1gb card. No hitching,it runs as smooth as a beta will run.

Like Virge said the game must streaming sources including memory.
http://forums.anandtech.com/showpost.php?p=32363360&postcount=100

Now we just have to see see the difference in image quality from a 69501gb vs a 6950 2gb @ 1080p max settings because performance seems to be the same.

 

[TheAdvocate]

I don't think it is ram issues because even the 6870 sends the GTX 560 TI packing and begging for frame-rate.

This is absurd. Fanboi much?

 

[SirPauly]

Yeah, the game cannot use more memory than you have. It's why I get hitching when I try to use 4xAA at 2560x1600 with ultra.

I haven't seen anyone running this game on a 1GB card with all in-game settings cranked, because they can't, or rather they get an unpleasant experience. People are running it with some settings on ultra and turning others off with 1GB cards. Anyone who has been running the maximum settings is using a GTX 580/480, 2GB 6950 or a 6970.

No one has said you cannot run this game on a 1GB card. You cannot run BF3 maxed out on a 1GB card, but you can certainly run it. I've posted ample evidence of memory usage scenarios in demanding games contrasted to BF3.

I just would like to know more about why BF3 wants so much VRAM, I'm past determining if it does or not, that is proven.

The key is the testing evaluation that Hardocp did offer and the settings they used -- not necessarily your subjective findings. With the 6870 and GTX 560 Ti, and yet the GTX 560 TI had to reduce settings and still the 6870 won handily and both were not memory limited.

In Hardocp's testing evaluation, don't believe the difference is based on memory limitations.

 

[happy medium]

I just ran the beta @ 1080p max settings 4xaa and used 981 mb of memory max.
I used cpuz.

Then I hooked up the old monitor @ 1600x1200, same settings , guess what? it used 980mb of memory .
Also using gpuz.

For me this issue about running out of memory is proven false.

 

[SirPauly]

This is absurd. Fanboi much?

Can you expand? The 6870 won by about 15 percent in over-all average, and the GTX 560 TI had to reduce settings as well in comparison. Would say that would be a huge win for a product, sent the GTX-560 TI packing and begging for frame-rate. The key is why? Considering the GTX 560 TI offers more over-all performance than a 6870 -- Why is the 6870 winning so dramatically?

 

[Grooveriding]

Groover read the thread, especially the last 10 posts on page one.

Read this post by Keys and toyota experienced the same thing I am.
YOU WILL NOT RUN OUT OF MEMORY with 1gb cards.
http://forums.anandtech.com/showpost.php?p=32363272&postcount=91

I have no idea what the game does with the extra memory but your game will not lose performance if you run at max settings @ 1080p with a 1gb card. No hitching,it runs as smooth as a beta will run.

GTX 560 - 1920x1200 Ultra No AA Min=25 Avg=38

Smooth ? And without AA.

I'm not going to continue. I've proven the game uses lots of VRAM. You've just said it isn't so, if you prefer to believe that, feel free.

 

[notty22]

Nobody is going to be running the game 'maxed out' (even @1080p) with a single card regardless.
So welcome to dual gpu, and micro-stuttering !
Techspot has their review up, of course its more like everyone else, opposite Hocp.
http://www.techspot.com/review/448-battlefield-3-beta-performance/page2.html

 

[Arkaign]

Can you expand? The 6870 won by about 15 percent in over-all average, and the GTX 560 TI had to reduce settings as well in comparison. Would say that would be a huge win for a product, sent the GTX-560 TI packing and begging for frame-rate. The key is why? Considering the GTX 560 TI offers more over-all performance than a 6870 -- Why is the 6870 winning so dramatically?

I wouldn't be so quick. Although at the moment I'm happy I have a 2GB 6950 lol.

What we see VERY often in situations like this (big new release, new engine) is one company lag in getting decent drivers out, often because the game company is still tweaking the game and engine as well.

I'd fully expect improvements in performance for both the AMD 5xxx and 6xxx series along with the Nvidia GTX4xx/5xx series over the next few weeks.

Beta drivers on a beta release doesn't mean a whole lot to me, and shouldn't be taken without a significant grain of salt.

 

[happy medium]

I've proven the game uses lots of VRAM

No one is debating this. We are trying to find out if 1gb is enough at 1080p/ultra /4xaa and as 3 members here as well as a few others on hardforum have shown it is BUT does it suffer from a image quality loss with 1gb is still up in the air.

I get better fps than Hardocp's gtx 560ti with my gtx460 as are many other web sites. I dont know how to explain that. I will say ,I would not play @ 1080p max setting with my card, its a little slow.

 

[Grooveriding]

This is a good bench to show the [H] review is showing fair results. The 560 is slower than the 6950 2GB in this bench on high, the same way it is slower than the 6950 2GB in [H]'s bench on ultra.

[H]ard actually shows the 560 doing better at 38fps on ultra while techspot shows it at 34fps on high. Both without AA. The 560 simply cannot run AA and ultra settings and give playable frames.

Most likely why neither site benched the card @ those settings.


Also every card they benched @ ultra with 4xAA MSAA has over 1GB of VRAM

Good stuff. Notice here the 560 is not included anymore, but the 6950 2GB is still being used. Why not ? 560 could not handle the AA settings, while the 6950 2GB could.

Interesting how close the 6970 is to the 580 and the 6970CF is to 580SLI, I would of thought the 580 would of been faster over the 6970 by a greater margin than that. There is hardly any difference, about 9%, but the 580SLI is 16% faster than the 6970CF, and the 6990 should be up there with the 590, if not faster. Looks like CF scaling needs to be improved in BF3, SLI is showing better scaling.

 

[zebrax2]

I just ran the beta @ 1080p max settings 4xaa and used 981 mb of memory max.
I used cpuz.

Then I hooked up the old monitor @ 1600x1200, same settings , guess what? it used 980mb of memory .
Also using gpuz.

For me this issue about running out of memory is proven false.

Question about the programs used to measure the vram.
Does it measure the vram used on the GPU or does it measure the vram that the program requires because if its the former then you will never exceed 1gb since your card only has 1gb of vram

 

[SirPauly]

I wouldn't be so quick. Although at the moment I'm happy I have a 2GB 6950 lol.

What we see VERY often in situations like this (big new release, new engine) is one company lag in getting decent drivers out, often because the game company is still tweaking the game and engine as well.

I'd fully expect improvements in performance for both the AMD 5xxx and 6xxx series along with the Nvidia GTX4xx/5xx series over the next few weeks.

Beta drivers on a beta release doesn't mean a whole lot to me, and shouldn't be taken without a significant grain of salt.

Agreed, It is beta, based on testing that is tough to gauge apples-to-apples evaluations.

The key is AMD came up on top by a noticeable degree, even with sku's with the same amount of memory as nVidia, based on Hardocp's evaluation.

 

[WMD]

you cant be using ultra settings and 4x AA because your ram usage is nothing like anybody elses. it would be using around 1011 mb for your 1gb card with those settings.

Here it is: Ultra preset 1680x1050 4XMSAA and HBAO. Max VRAM usage 845MB

Edit: Even at those settings, the framerate is low but smooth at 35-45fps. I am not sure how you are hitting over 1.1GB at 1280x960 but I think the game really does try to use up available video memory. I think 1GB cards will be fine up to 1920x1080 4XAA or 2560x1600 with post AA.

 

[happy medium]

I thought BF3 @ 1080p ultra settings with 4xaa uses 1.4gb of memory?
Why does a gtx 570 seem to only need 1.25gb's? I'll tell you why because it uses memory avalable, if not the gtx 570 would lose to a 6950 and a gtx480 with 4xaa applied.

 

[SirPauly]

This is a good bench to show the [H] review is showing the same. The 560 is slower than the 6950 2GB in this bench, the same way it is slower than the 6950 2GB in [H]'s bench.

[H]ard actually shows the 560 doing better at 38fps, while techspot shows it at 34fps. Both without AA. The 560 simply cannot run AA and ultra settings and give playable frames.

Most likely why neither site benched the card @ those settings.


Also every card they benched @ ultra with 4xAA MSAA has over 1GB of VRAM

Good stuff. Notice here the 560 is not included anymore, but the 6950 2GB is still being used. Why not ? 560 could not handle the AA settings, while the 6950 2GB could.

Interesting how close the 6970 is to the 580 and the 6970CF is to 580SLI, I would of thought the 580 would of been faster over the 6970 by a greater margin than that. There is hardly any difference, about 9%, but the 580SLI is 16% faster than the 6970CF, and the 6990 should be up there with the 590, if not faster. Looks like CF scaling needs to be improved in BF3, SLI is showing better scaling.

If your memory limitation theory made sense -- then why is the GTX 570 offering more frame-rate than a 6950 with 2 gigs of ram based on Techspots findings? It is hardocp's findings that are out of whack compared to everyone else's -- the 6950 hands the GTX 570 a new one based on HardOCP evaluation as well and in that 15 percent range, when the GTX 570 is faster over-all than a Radeon 6950.

 

[happy medium]

Here it is: Ultra preset 1680x1050 4XMSAA and HBAO. Max VRAM usage 845MB

That seems right I get 980mb at 1600x1200 but I also use the same amount 981mb @ 1900x1080.

 

[happy medium]

If your memory limitation theory made sense -- they why is the GTX 570 offering more frame-rate than a 6950 with 2 gigs of ram based on Techspots findings? It is hardocp's findings that are out of whack compared to everyone else's -- the 6950 hands the GTX 570 a new one based on HardOCP evaluation as well and in that 15 percent range, when the GTX 570 is faster over-all than a Radeon 6950.

Hurray! I'm glad I'm not the only one that sees it.

2 conclusions

Hardocp preview sucks and you dont need more than 1gb for 1080p ultra 4xaa. The game somehow uses what it has avalable.

 

[TheAdvocate]

So wait - the 2GB Radeon 6950 is faster than the 1GB 560Ti at high resolutions where VRAM becomes a bottleneck.

Holy crap, next youa re going to tell me the Pope is Catholic and water is wet. Stunning revelation there.

If you want to do a side by side comparison, either 1) Compare a 1GB Radeon to the Ti or 2) Compare the 2GB Ti to the 2GB Radeon, so you can neutralize the VRAM discrepency. Otherwise, you're doing a VRAM bench for a game that is obviously demanding it whenever you enable every visual option at high resolutions.

Next, cut the fanboi crap like "sends XXX packing" which #1 is simply not true, and #2 is ridiculously juvenile. Did you actually fist pump when you said that?

Also, that 2GB Radeon costs a good deal more on average than a 1GB Ti. Not surprisingly, the 2GB versions are within about $10 of each other.

Finally, please find me a stock 560Ti 822MHz that is still regularly sold by a prominent vendor such as Newegg. You can't. So this test of [H]'s, besides being setup to be ridiculously biased towards AMD (I have just about written that site off at this point - they've lost all integrity), is usueless to someone like me, because it doesn't test my gaming conditions.

Here's what I do know - I have 10 hours in the beta so far and I can, and have, played entire matches @ ultra everything on my rig and never dipped below 47 FPS per Afterburner. However, I prefer to shut HBAO off because the performance hit is ridiculous compared to the minimal visual quality gains. There's the real story - what settings are worth it and what are not? Last night I played with 16xAF and no AA or HBAO and was getting about 115 fps in the subway. That's cute, but since I never dropped anywhere close to 60 fps, even in a fire fight, I probably should turn 2xAA back on and settle in. My goal is to find the best mix of settings that still delivers a 60 fps average framerate without stuttering/huge dips/etc.

 

[TheAdvocate]

double