A look at VRAM use with AA enabled

[LTC8K6]

Agree, but show me another time in the last 10-15 years where the VRAM has differed so much between the top 3 options?

Titan X: 12GB
980Ti: 6GB
Fury: 4GB (rumored)

Add to the fact that we also have HBM mixed-in. This isn't 2GB vs 1GB, 2GB vs 1.5GB or even 3GB vs 2GB. We are talking 2-3x differences in some cases, with the LEAST difference being 50%. That's pretty unique.

But we've heard that Fury does 4K well with it's 4gb of HBM.

That was the tweet, anyway.

So I guess that is the hype target.

Fury should be a single GPU 4K gaming card.

 

[exar333]

But we've heard that Fury does 4K well with it's 4gb of HBM.

That was the tweet, anyway.

So I guess that is the hype target.

Fury should be a single GPU 4K gaming card.

For sure, I don't disagree at all. Just want to see a more in-depth review of that claim.

Kind of reminds of when Conroe launched. It seemed odd at first that a 2.4ghz CPU would be so much better than an existing 4.6ghz chip. The numbers don't always tell the whole story. I know that's a little different, but just trying to think of a relatively similar experience.

People dropping $600+ on a GPU should expect it's the best and have little in compromise. The 4GB is the only really uncertain aspect of Fury IMHO. If the shader count and so forth is what we expect, it will be killer.

 

[waldoh]

I am using 5100mb vram in gta 5.

DSR @ 4k
Ultra everything (except grass)
Extended view distance
fxaa/msaa x4/txaa/aa x16

 

[Silverforce11]

I am using 5100mb vram in gta 5.

DSR @ 4k
Ultra everything (except grass)
Extended view distance
fxaa/msaa x4/txaa/aa x16

Does that even work?

Super sampling at 4K with 4x MSAA active.

Also, what would be the point of running essentially SSAA and MSAA together?

 

[RibeyeAndScotch]

Interesting, I wonder how taxing MSAA is to vram

 

[dave1029]

I passed a staggering 6GB of VRAM in GTAV.

 

[happy medium]

The 4gb future of Fury. I hope people who spend 700$ on a card want it last a while. If you tri/crossfire 2/3 Fury's it still only has 4gb of memory right?

 

[IEC]

Yeah that doesn't tell us anything about whether you actually need more than 4GB to not stutter. Misrepresenting how much VRAM is used because it's available vs how much VRAM is required is disingenuous.

 

[krumme]

http://www.tweaktown.com/tweakipedia/90/much-vram-need-1080p-1440p-4k-aa-enabled/index.html

Looks like the 4gb line will be crossed if you like AA at 4K.

So you needed a new thread for this crap? You perfectly knew beforehand this is bs for arguments you already know and we have been over several times before in many threads. Flying under the mod rader in disguise is coward tactics. Troll.

 

[Gikaseixas]

The 4gb future of Fury. I hope people who spend 700$ on a card want it last a while. If you tri/crossfire 2/3 Fury's it still only has 4gb of memory right?

I'm sure reviews will address your concerns.

 

[Headfoot]

You guys just dont get it. You CANT quote your VRAM use without saying what card it is. VRAM use is dependent on your card. A Titan X will make use of more VRAM because it has more VRAM. It is not static, so don't treat it like it is. A 780 Ti will use less ram without stuttering than a Titan X will use similarly without stuttering.

 

[Gikaseixas]

You guys just dont get it. You CANT quote your VRAM use without saying what card it is. VRAM use is dependent on your card. A Titan X will make use of more VRAM because it has more VRAM. It is not static, so don't treat it like it is. A 780 Ti will use less ram without stuttering than a Titan X will use similarly without stuttering.

Too much logic in a single post

 

[happy medium]

VRAM use is dependent on your card

In all cases? So a 2gb gtx680 in tri sli will play Thief @ 4k?
I think it depends on the game and resolution and settings.

 

[Gikaseixas]

In all cases? So a 2gb gtx680 in tri sli will play Thief @ 4k?
I think it depends on the game and resolution and settings.

Why presenting a worse case scenario?

Try GTX 780ti SLI :biggrin:

 

[happy medium]

Why presenting a worse case scenario?

Try GTX 780ti SLI :biggrin:

I think that combo would fall on its face also.:thumbsup:

A high end setup ,especially when someone is buying 600$+ cards should feel comfortable that it will run all games maxed @ high resolutions.
Sucks 4gb I think is not enough for a high end card like the Fury.
I guess they have no choice until they can use HBM 2.

 

[Gikaseixas]

I think that combo would fall on its face also.:thumbsup:

poor Kepler owners

Sucks 4gb I think is not enough for a high end card like the Fury

How do you explain GTX 980ti SLI and 295x2 > Titan X @ 4k?

 

[Headfoot]

In all cases? So a 2gb gtx680 in tri sli will play Thief @ 4k?
I think it depends on the game and resolution and settings.

Yes, in all cases. VRAM use also depends on resolution, AA level and algorithm, and the game played. They are not mutually exclusive. 2 GB will not always be enough. Which is very obviously not what I said.

These things are factors contributing to how much VRAM is consumed, in all cases: 1) Total VRAM of the card used, 2) Game being played, 3) AA levels and algorithm used, 4) resolution, 5) in game settings. If you leave off information as to ANY of these variables, your data is junk.

 

[happy medium]

Yes, in all cases

wrong, not every game streams memory.

 

[Piroko]

In all cases? So a 2gb gtx680 in tri sli will play Thief @ 4k?

Would 4 GB gtx680 tri sli be able to run it at >20 fps? Not even GT980s in tri sli can sustain 60fps at 4k afaik, so that's a bit of a moot point really.

My take on this discussion is that 290x and GTX980 will most likely run out of shader performance before running out of vram in future games - even in cf/sli.
Fiji on the other hand might be one of those cards that run out of vram in some games at least.

On the other hand, we've seen high res textures now and there's only so far that game devs can go with texture resolution without being stupid. There are easier ways to improve image quality than upping every texture to 4k *4k and improving the image quality by adding more lights and more sophisticated shaders doesn't really affect vram usage that much.

And then there is DX12, which could end up like dx9 where the new shader model 2.0 absolutely murdered the first 2 generations' shader performance. Increasing draw calls by a lot can allow some very neat effects that may or may not increase vram usage, but most definitely increase GPU load.

 

[Headfoot]

wrong, not every game streams memory.

Maybe some old games dont. All modern games do. I'll modify my previous statement here to be more accurate, "Yes in all cases in current reviewer benchmark lineups on current drivers." If we want to talk about Quake 3 or something then sure total VRAM probably doesnt influence it. This is no longer the case in modern games for the vast majority if not all cases (depending on your definition of "modern game"). Any game where we actually have to worry about running out of VRAM on modern cards is going to be influenced by Total VRAM.

 

[boozzer]

Even the last COD used more than 4GB vram. The question isn't so much usage as to whether its required.

ie. Performance suffers due to lack of vram.

Then one has to factor in whether there's actually enough GPU grunt to run games at 4K with 4x MSAA or lol 8x MSAA (yes, TT tested 4K with 8x MSAA or SSAA!). Considering many modern games tank hard in performance as soon as MSAA is used, its really only for those who QUAD SLI Titan X.

The most realistic usage scenario for single or dual card of Titan X performance caliber at 4K is with MSAA disabled.

no single gpu can run 4k AAA games with 4x msaa or higher I 100% agree.

wonder why some keep on bringing up useless metrics?

yes you can run with enough vram for 4x msaa and you get around 20 fps max for it. enjoy your 20 fps. :twisted:

 

[IEC]

As those of us already running 4K/UHD resolutions know, you're going to hit a wall in GPU performance before you hit a VRAM limitation. And forget about enabling MSAA with a single GPU in newer titles, even with the Titan-X. Terrible performance hit for not much gain.

I don't anticipate needing more than 4GB VRAM. Will there be corner cases where this is a limitation? Possibly, but the difference between 15fps and 7fps isn't going to make me happy either way...

 

[amenx]

This is a terrible review they did. Showing consumption measured in Afterburner tells you nothing about how much VRAM you actually need. You can use two different video cards with different amounts of VRAM in the same system, same game and same settings - but see two different values of VRAM consumed. You have to show framerate over time to give a real indication of how much VRAM you actually need. When you actually run out of VRAM your game drops down to sub 5fps and stutters while the memory is refilled. Just because software says the game is using X amount of VRAM, it does not mean that is how much VRAM you need.

For example my VRAM used in Battlefield 4 was different on my 780ti cards vs my Titan X cards; the Titan Xs used more VRAM. This has no impact on visuals or gameplay, both were identical, just different amounts of VRAM consumed.

The only time I've actually ran out of VRAM in recent memory was trying to run Shadow of Mordor at 2560x1600 on my 3GB 780ti cards. The game ran out of VRAM constantly and stuttered horribly. I had to reduce the texture settings to make it playable. On my Titan X cards I saw VRAM usage over 5GB, but if you look at reviews of GTX 980 cards with 4GB, they don't stutter at 2560x1600 with maxed settings. The 980 with 4GB can even run Mordor at Ultra settings 4K without running out of VRAM:

TLDR: This only shows how much VRAM is being filled in monitoring software, not how much VRAM you actually need. To show actual VRAM requirements you need to show minimum framerate over time in a graph that also displays VRAM usage over the same synchronized timeline.

Yep, always been my argument. All these vram usage reviews are useless without having identical hardware with different vram capacities to test. Like maybe a 4gb 290x vs 8gb 390x running at same clocks and benching the performance impact in FPS and FCAT charts. Vram usage is not very meaningful all by itself . It needs context and the context is performance impact.

 

[MrTeal]

It would be interesting to see if a BIOS flash would disable VRAM on a 8GB 390X or 290X.

If someone would like to donate the cards, I would be more than happy to benchmark quad-290X at 5k Ultra in 4GB and 8GB modes.

 

[cen1]

I will wait for the actual Fury reviews but the 4GB FUD mills are certainly starting to spin around the web.