Known/suspected games to eat more than 1GB video memory at 1920x1200

[drifting]

I wish reviewers would do a review on 6950 1GB vs 6950 2GB on all the so called "used more than 1GB of VRAM" games on 1920 res 4xAA to show the actual performance difference between the two.

Bit-tech has done a few for 6950 1GB vs 6950 2GB at 1920x1200 4xAA 16xAF:
http://www.bit-tech.net/hardware/2011/02/11/amd-radeon-hd-6950-1gb-review/1
Of all the games they testing, the 6950 2GB deliver identical frame rate as the 6950 1GB at all games, but with only one exception where in Black OPs the 6950 1GB deliver 2fps less minimum frame rate than the 6950 2GB, and even that is 53fps vs 55fps so it's not like it's that huge impact with "running out of memory".

I'm also curious to see, however as I mentioned before, the ad-hoc results from the reviews do not necessarily cover the worst case for each game.

A simple example: what's the min fps you can find, for World of Warcraft Cataclysm DX11 mode? How many reviews would state a min fps of merely 30? The fact is that even my 580 SLI could get less than 30 fps in Orgrimmar when there are over 40 player characters in sight.

 

[whipgeez]

wow, metro looks great, is it any good tho as a game?

 

[arc@css]

wow, metro looks great, is it any good tho as a game?

Yeah, I'd say so. The first level where things feel a little on rails seemed to drag on a little, but other than that it is a decent shooter. Eye candy of the highest quality makes it about the best looking game I have seen yet.

 

[Paracelsus]

2007-11: Crysis / Warhead: No more lag while quickly rotating my camera. Confirmed to exceed 1GB video memory easily.

Doesn't even hit 700mb in dense jungle with me, all very high 1920x1200 no AA. Maybe AA will change that, but without numbers you simply can't throw terms like 'confirmed' around.

 

[drifting]

Doesn't even hit 700mb in dense jungle with me, all very high 1920x1200 no AA. Maybe AA will change that, but without numbers you simply can't throw terms like 'confirmed' around.

I don't know about the "dense" jungle you are talking about. I can no longer find a working link to the 1200p screenshots. I can only show you some 1080p screenshots during some easy scenes. Note these were captured on a card with only 1GB vram. If it had more than 1GB vram then we could probably even see more than 1GB vram usage at 1080p in these screenshots.

s/s from some review site for 1200p:

s/s for 1600p:

I don't always have the time to show you guys where the numbers are from. It's totally up to you whether to trust me or not.

 

[bhavv]

I dont doubt that there are games that use more than 1 Gb Vram at 1920x1200 resolution.

I do however strongly doubt that this causes a performance hit on decent 1 Gb cards. I've ran metro 2033 on max settings with 4x AA, and it was completely smooth and playable on my SLI setup.

I dont care about screenshots showing how much Vram a game uses up, I want solid evidence that any current mainstream game is going to take a significant performance hit of far more than a few FPS by having its Vram limit exceeded.

In the games I've played reported to use up more than 1 Gb Vram (Metro and TDU2), I have not seen any slowdown or lag that other people complain about. Any slowdown that does occur in a game like Metro is fully due to the card being too slow for the game, not to do with the Vram. If you go SLI or crossfire, the game runs smooth as silk when fully maxed out with AA on a pair of decent 1 Gb cards.

Even in some situations where you can show a game like Crysis eating up around 1200-1500 Mb of ram, a 768 Mb Geforce GTX 460 is still able to run the game completely fine in those cases.

As far as I can tell, offloading the extra Vram information to the allocated amount of virtual memory you have set on your PC (4 Gb total Vram on mine for example) does not cause anymore than a few FPS reduction, nor any kind of lag that some people keep on imagining to be the case.

A simple example: what's the min fps you can find, for World of Warcraft Cataclysm DX11 mode? How many reviews would state a min fps of merely 30? The fact is that even my 580 SLI could get less than 30 fps in Orgrimmar when there are over 40 player characters in sight.

That doesnt have anything to do with Vram, thats simply because WoW is a horribly coded game. The engine in Guild Wars is capable of displaying HUNDREDS of player characters in the same outpost without any performance hit whatsoever.

Also I saw Dragon Age 2 mentioned in the OP. DA2 is a blatant example of a shoddy console port that simply eats away at ram ad is so CPU limited entirely due to how poorly it is coded to run on PCs. Crysis 2 also displays exactly the same problems from being an unoptimized console port where the game is simply incapable of making use of a decent hardware setup because it is so badly coded.

 

[Marine-RX179]

I'm also curious to see, however as I mentioned before, the ad-hoc results from the reviews do not necessarily cover the worst case for each game.

A simple example: what's the min fps you can find, for World of Warcraft Cataclysm DX11 mode? How many reviews would state a min fps of merely 30? The fact is that even my 580 SLI could get less than 30 fps in Orgrimmar when there are over 40 player characters in sight.

I don't mean to be rude, but I'm starting you think you don't really know what are the causes of low frame rate. WOW has to be the worst example for representing graphic performance. Yes since the dx11 update the graphic has became more demanding, but not THAT demanding. The fact that you are getting low frame rate with many characters on screen at the same time points to the CPU being the bottleneck, rather than graphic card running out of VRAM or the cards themselves being not fast enough. Check your GPU usage for the GTX580 SLI while in WOW...I wouldn't be surprised at all if they are at under 50% usage.

In fact, majority of games that got lots of characters/units appear on screen at the same time (mmorpgs, Total Wars series etc) are heavily CPU limited...because of the games are not written to use all four cores of the CPU (in fact, many of them barely make full use of 2 cores, and WOW being one of them). Another example is that for me in Global Agenda, my frame rate would drop to 25~30fps when there are lots of characters on screen at the same time, while my 5850's GPU usage from the usual 60-70% down to around 50-60% usage. People with CPU like overclocked like i5/i7 using the same card barely have their frame rate drop to below 40fps.

 

[bhavv]

And its also funny how most of the other games I didnt mention listed in the OP are poorly optimized console ports as well, maybe you could say that about Crysis 2?

Such games were not designed for PC hardware, they were made for consoles and then ported over to the PC. They tend to be heavily CPU dependent, while gobbling up an excessive amount of Vram that wouldnt even be needed if the game was properly made and optimized for the PC.

 

[Kevin1234]

When i ran crysis on my 5850s x3 on 10.11 drivers vram around 999MB hitting 59fps on testbench now with 11.3 drivers vram has shotup to 1400MB stuttering same settings as before 1920x1080 v.high 64bit and only 43fps in benchtest?

 

[Marine-RX179]

When i ran crysis on my 5850s x3 on 10.11 drivers vram around 999MB hitting 59fps on testbench now with 11.3 drivers vram has shotup to 1400MB stuttering same settings as before 1920x1080 v.high 64bit and only 43fps in benchtest?

AMD nerfing older cards performance to courage people to spend money on buying new not much faster 2GB cards may be?

 

[Kevin1234]

AMD nerfing older cards performance to courage people to spend money on buying new not much faster 2GB cards may be?

My thoughts exactly........

 

[Buckster]

^^ er don't agree ... DCS A10 certainly designed for PC - Afterburner showed 1480MB of VRAM used up

Arma 2 used all of my VRAM too

both PC games only

 

[drifting]

That doesnt have anything to do with Vram, thats simply because WoW is a horribly coded game. The engine in Guild Wars is capable of displaying HUNDREDS of player characters in the same outpost without any performance hit whatsoever.

I don't mean to be rude, but I'm starting you think you don't really know what are the causes of low frame rate. WOW has to be the worst example for representing graphic performance. Yes since the dx11 update the graphic has became more demanding, but not THAT demanding. The fact that you are getting low frame rate with many characters on screen at the same time points to the CPU being the bottleneck, rather than graphic card running out of VRAM or the cards themselves being not fast enough. Check your GPU usage for the GTX580 SLI while in WOW...I wouldn't be surprised at all if they are at under 50% usage.

Don't get me wrong, but I have NEVER said in the OP that starting one instance of DX11 WOW could lead to vram shortage and hence cause low fps. I gave you the screenshots WITHOUT vram usage but merely with fps, only to prove that the numbers from the reviews does not necessarily reflect worst cases. You probably should NOT take screenshots without vram usage as evidence to support my list. The point here is, you should *play* the game yourself to find out, instead of just quoting numbers from reviews.

I have ONLY said in the OP that starting two instances of DX11 WOW on a 1GB card could give you single-digit fps and this is because of vram shortage.

 

[drifting]

And its also funny how most of the other games I didnt mention listed in the OP are poorly optimized console ports as well, maybe you could say that about Crysis 2?

Such games were not designed for PC hardware, they were made for consoles and then ported over to the PC. They tend to be heavily CPU dependent, while gobbling up an excessive amount of Vram that wouldnt even be needed if the game was properly made and optimized for the PC.

I wouldn't care how you call these console ports but anyway my conclusion would be 1GB cards are getting out-of-date for these "console ports" thing or whatever you call them. A simple example would be GTA4: if you only have 1GB card you wouldn't even be able to max out the graphics card settings.

I do however strongly doubt that this causes a performance hit on decent 1 Gb cards. I've ran metro 2033 on max settings with 4x AA, and it was completely smooth and playable on my SLI setup.

I dont care about screenshots showing how much Vram a game uses up, I want solid evidence that any current mainstream game is going to take a significant performance hit of far more than a few FPS by having its Vram limit exceeded.

In the games I've played reported to use up more than 1 Gb Vram (Metro and TDU2), I have not seen any slowdown or lag that other people complain about. Any slowdown that does occur in a game like Metro is fully due to the card being too slow for the game, not to do with the Vram. If you go SLI or crossfire, the game runs smooth as silk when fully maxed out with AA on a pair of decent 1 Gb cards.

Even in some situations where you can show a game like Crysis eating up around 1200-1500 Mb of ram, a 768 Mb Geforce GTX 460 is still able to run the game completely fine in those cases.

As far as I can tell, offloading the extra Vram information to the allocated amount of virtual memory you have set on your PC (4 Gb total Vram on mine for example) does not cause anymore than a few FPS reduction, nor any kind of lag that some people keep on imagining to be the case.

May I ask how much vram do you have on your card? If you only have 1GB, a simple example would be to run the ending of Crysis Warhead. During the realtime animated cinema scene in which Psycho fights the Korean boss, every time when the camera switches, the game would lag for like 0.2 seconds or so, which is pretty obvious on 1GB cards. However 2GB cards doesn't have that problem.

 

[drunkenmaster]

You're back to claiming 1gb isn't enough in GTA 4, despite saying you've used settings on a 1gb card with the game claiming it needs 1.2gb, with no drop in performance.

Yet you can't put 1 + 1 together. If a 1gb card isn't running into a massive performance drop when some random application claims its using 1.2gb, its very simple theres two options, its not REALLY using 1.2gb, or the card magically generated an extra .2gb of memory chips on the card that you can't see, because they are cloaked.........

Posting a screen shot + vmem supposedly being listed + performance + showing an uncharacteristic drop in performance would prove vmem is being exceeded, infact, you can do this without the vmem being shown in a screen shot.

I've shown you how games look when you go from non memory limited, to memory limited, you have not. 20fps on a 1gb card to 30fps on a 2gb faster card, just suggests that card is faster, not much else, it proves NOTHING in relation to memory.

4xaa being slower, but not horrifically bad in a game vs faster and playable at 0xaa does NOT prove memory is being exceeded, 4xaa in a game will run slower than 0xaa.

You need to add these things together and show a vmem drop, not screen shots with vmem listed, again you have NO proof these numbers are accurate. Its like the threads where a guys unhappy his temps have gone up to 60c load from 50c load, theres no problem, its not crashing, the source is higher ambients, without seeing the number he'd be perfectly happy, seeing a number he automatically assumes something is wrong.

I'm afraid thats all I'm seeing here, can you be sure the numbers listed are being read correctly? Does it matter that MS suggests the driver should not be able to, that it doesn't work on AMD cards, that its a call designed for Win xp, that it also wouldn't work on Vista, that on Vista it would show both vmem + system mem reserved together as a much larger number?

The chiphell screen shot showing 1.7gb for instance, in what relation is that, was performance shockingly terrible, or was performance fine and the 1.7gb was clearly wrong, you're giving us half the story, and the half you want.

Are you sure those SLI screen shots at 1.7gb weren't reading the total the driver see's both cards using and the "real" number is only half that, no, well firstly, be sure, secondly compare these readings to performance, one without the other is JUST a number. If a game is playing fine at 30fps at 0xaa at a high res, and goes to 25fps at 4xaa but the memory monitor is saying its gone from 950mb to 1.4gb, does that suggest the memory monitor is incorrect?

With a real memory limit you'd be dropping from 30fps to 5fps, maybe 0.5fps.

You've shown and "confirmed" it yourself that GTA 4 is not giving accurate memory numbers, so why are you blindly assuming everything else is.

E-mail a reviewer, ask them to look into it, e-mail Nvidia ask them if the gpu-z numbers are accurate.

Ask them if its raw textures being read before they get compressed and sent to the memory.

The simple fact is, real memory limits = horrific performance, I've yet to see reviews show 1920x1080 is limited in ANY of the games you've suggested.

 

[drifting]

You're back to claiming 1gb isn't enough in GTA 4, despite saying you've used settings on a 1gb card with the game claiming it needs 1.2gb, with no drop in performance.

I haven't said I've used settings on a 1gb card with the game claiming it needs 1.2gb with no drop in performance. When I was using 1gb cards, the game didn't even allow me to max out the settings. I have only said other people were using 1.5gb card to capture a screenshot of vram usage, when the game asks for 1.2gb to max the settings, msi afterburner measures only about 1gb.

Yet you can't put 1 + 1 together. If a 1gb card isn't running into a massive performance drop when some random application claims its using 1.2gb, its very simple theres two options, its not REALLY using 1.2gb, or the card magically generated an extra .2gb of memory chips on the card that you can't see, because they are cloaked.........

Posting a screen shot + vmem supposedly being listed + performance + showing an uncharacteristic drop in performance would prove vmem is being exceeded, infact, you can do this without the vmem being shown in a screen shot.

I've shown you how games look when you go from non memory limited, to memory limited, you have not. 20fps on a 1gb card to 30fps on a 2gb faster card, just suggests that card is faster, not much else, it proves NOTHING in relation to memory.

4xaa being slower, but not horrifically bad in a game vs faster and playable at 0xaa does NOT prove memory is being exceeded, 4xaa in a game will run slower than 0xaa.

You need to add these things together and show a vmem drop, not screen shots with vmem listed, again you have NO proof these numbers are accurate. Its like the threads where a guys unhappy his temps have gone up to 60c load from 50c load, theres no problem, its not crashing, the source is higher ambients, without seeing the number he'd be perfectly happy, seeing a number he automatically assumes something is wrong.

Let me put it this way. If you only have 2GB system memory, and you run Windows 7 x64. There are 3 categories I want to list here:

1) When your memory usage is below 2GB, you can alt+tab switch between programs smoothly.

2) When your memory usage is sort of between 2GB and 3GB, you don't notice any obvious lag unless you alt+tab switch between programs fast.

3) When your memory usage is far above 4GB, you probably notice that even your current program is laggy, waiting for heavy swap of pages between memory and hdd.

Most of the games I listed here fall into scenarios similar as category 2. You wouldn't notice obvious drop of average fps, but you'd suffer from worse min fps, which is usually caused by rotating your camera fast or scene switch or something suddenly appears in your sight. You are claiming that you only think scenarios similar as category 3 can be regarded as vram shortage, but everyone can make his own choice. Personally I would not even say 6GB system memory is enough, simply because I don't want any page swap between my system memory and my SSD when I switch between programs while I run several virtual machines concurrently. I hope you get my point.

I'm afraid thats all I'm seeing here, can you be sure the numbers listed are being read correctly? Does it matter that MS suggests the driver should not be able to, that it doesn't work on AMD cards, that its a call designed for Win xp, that it also wouldn't work on Vista, that on Vista it would show both vmem + system mem reserved together as a much larger number?

The chiphell screen shot showing 1.7gb for instance, in what relation is that, was performance shockingly terrible, or was performance fine and the 1.7gb was clearly wrong, you're giving us half the story, and the half you want.

Are you sure those SLI screen shots at 1.7gb weren't reading the total the driver see's both cards using and the "real" number is only half that, no, well firstly, be sure, secondly compare these readings to performance, one without the other is JUST a number. If a game is playing fine at 30fps at 0xaa at a high res, and goes to 25fps at 4xaa but the memory monitor is saying its gone from 950mb to 1.4gb, does that suggest the memory monitor is incorrect?

With a real memory limit you'd be dropping from 30fps to 5fps, maybe 0.5fps.

You've shown and "confirmed" it yourself that GTA 4 is not giving accurate memory numbers, so why are you blindly assuming everything else is.

E-mail a reviewer, ask them to look into it, e-mail Nvidia ask them if the gpu-z numbers are accurate.

Ask them if its raw textures being read before they get compressed and sent to the memory.

The simple fact is, real memory limits = horrific performance, I've yet to see reviews show 1920x1080 is limited in ANY of the games you've suggested.

If nVidia does not return the correct vram usage, how are CUDA C developers supposed to code and debug those GPGPU programs under Windows 7? I can't guarantee these readings are always accurate but it is an important reference.

 

[TheRealDeal]

There was a few reviews back when the 4870 1gb was released and compared to the 4870 512mb. The fps was similar between both cards but the reviews said the 1gb version was a much smoother gaming experience. I would put that down to vram usage on the 512mb being to high. If you just looked at the fps given from the games then both looked the same on paper but the gaming experience was not the same.

I think to say using more vram than your card has will make your performance drop to the levels of 0-1fps is wrong if you are using slightly more than you have. I think using slightly more will show up as choppy performance and using way more will result in the massive drops to 0-1fps.

 

[RedvGreen]

As I am looking for a card to last me 18months, what is your conclusion? if we are looking to play the games mentioned, we all need to jump to AMD/ATI?

1.5GB GTX 580 are useless?

 

[TheRealDeal]

As I am looking for a card to last me 18months, what is your conclusion? if we are looking to play the games mentioned, we all need to jump to AMD/ATI?

1.5GB GTX 580 are useless?

No 1.5gb gtx580 is fine. I do think the amd 69xx with 2gb of ram is a good selling point though. I am not worried about my card having 1gb as if i need to i will turn down some settings.

 

[bigjimmyauk]

i can confirm that in the metro 2033 benchmark on max setting at 1920x1080 , 35 seconds in a get a massive dip in fps which i think is due to memory limitations fps drop from 20's to single digits even tho there is not a lot more stuff happening on screen