Are game requirements and VRAM usage a joke today?

[evernessince]

That first chart is showing VRAM allocation. Games almost always allocate more than the bare minimum required to run smoothly. In the Far Cry 6 example, it's allocating over 10GB on the 6700XT, but still able to function within the 8GB memory pool of the 3070 once FSR is enabled. We have no way of knowing exactly how much crucial date is spilling over.

No, it's showing utilization as the chart says right at the top.

And yet, the 4060 Ti 16GB dusts its 8GB brother (and all other 8GB cards for that matter) in FC6 with RT. It's really one of the clearest examples there is.

Again though other reviews don't show these numbers and as that very chart shows, the 8GB A750 is performing fine. If 8GB is the problem then why is an 8GB card pulling so far ahead?

 

[Deleted member 234997]

No, it's showing utilization as the chart says right at the top.



Again though other reviews don't show these numbers and as that very chart shows, the 8GB A750 is performing fine. If 8GB is the problem then why is an 8GB card pulling so far ahead?

Intel has abilities that Nvidia don't.

From what I recall if your game is running ok for you then it don't matter.
My Geforce 3070 hated FC6 and it was unplayble for what I would like, it went from 35 down to 12, but if people are ok with that, that's all that matters.

 

[3DVCash]

No, it's showing utilization as the chart says right at the top.

They're synonymous... allocated VRAM is being utilized.

Again though other reviews don't show these numbers and as that very chart shows, the 8GB A750 is performing fine. If 8GB is the problem then why is an 8GB card pulling so far ahead?

I certainly wouldn't say the A750 is performing "fine." It doesn't choke quite as badly as other cards in that example, but drops behind the 3060 12GB and loses ground on the 16GB A770 compared to all the lower res benches. https://www.techpowerup.com/review/nvidia-geforce-rtx-4060-ti-16-gb/33.html

 

[evernessince]

They're synonymous... allocated VRAM is being utilized.

Guru3D reports the amount of VRAM actually used, stop with the nonsense semantics argument.

I certainly wouldn't say the A750 is performing "fine." It doesn't choke quite as badly as other cards in that example, but drops behind the 3060 12GB and loses ground on the 16GB A770 compared to all the lower res benches. https://www.techpowerup.com/review/nvidia-geforce-rtx-4060-ti-16-gb/33.html

It's loosing 23% of it's performance relative to the 73% the 3070 is loosing. This disproves your theory as clearly we can see an 8GB card here that doesn't take a 70% hit. Whether it's subjectively performing "fine" is irrelevant here, it's not having the dip your theory predicated.

Your theory of which had more sources disproving it to begin with, two other publican's numbers contradict this.

 

[Deleted member 234997]

semantics

My own experience, numbers in the teens at 1440P is not playable for me.
I think for the vast majority it's not either.

There are no hard rules on what someone deems as playable however.

 

[evernessince]

My own experience, numbers in the teens at 1440P is not playable for me.
I think for the vast majority it's not either.

There are no hard rules on what someone deems as playable however.

The provided chart is at 4K RT max, not 1440p. In addition, the A750 gets 33 FPS at those settings so not in the teens regardless. Mind you, the discussion was not around what's subjectively playable or not. It was around whether DLSS alleviates VRAM issues. 3DVCash submitted the idea that it does and provided a single example, Far Cry 6 4K RT Ultra. I then provided two other publications from HardwareUnboxed and Tom's Hardware demonstrating the same cards not having any issue under the same settings. I also pointed out in the one example that he did provide that the A750, with 8GB of VRAM, also does not exhibit the issue. The absolute performance is not what's relevant, what's relevant is the A750 is seeing performance scaling similar to high VRAM capacity cards and is not loosing 70% of it's performance like the 3070 and that other publicans contradict 3DVCash's provided graphic in regards to 8GB card performance.

 

[Deleted member 234997]

The provided chart is at 4K RT max, not 1440p. In addition, the A750 gets 33 FPS at those settings so not in the teens regardless. Mind you, the discussion was not around what's subjectively playable or not. It was around whether DLSS alleviates VRAM issues. 3DVCash submitted the idea that it does and provided a single example, Far Cry 6 4K RT Ultra. I then provided two other publications from HardwareUnboxed and Tom's Hardware demonstrating the same cards not having any issue under the same settings. I also pointed out in the one example that he did provide that the A750, with 8GB of VRAM, also does not exhibit the issue. The absolute performance is not what's relevant, what's relevant is the A750 is seeing performance scaling similar to high VRAM capacity cards and is not loosing 70% of it's performance like the 3070 and that other publicans contradict 3DVCash's provided graphic in regards to 8GB card performance.

I run 1440P and FC6 was unplayable with maxed settings on the RTX 3070.

It's bad argumentation if you switch testing suites even for the same game as people generally don't always bench the same places in games.

Canned benchmarks can provide lighter loads or more extreme loads too compared to in-game.

My 3070 was unplayable in-game, the beginning to the part the tank goes by,, my FPS here dropped to the teens on 1440P, RT on , maxed textures. Around 100w power used, 7.7GB Vram used, 99% GPU usage.

VRAM necked.


No clue about Intel, never owned one.

The 7900 series are beastly in games like Starfield.

My card does 3.2ghz solidly without crashing in anything.

Luck sometimes...

 

[evernessince]

I run 1440P and FC6 was unplayable with maxed settings on the RTX 3070.

It's bad argumentation if you switch testing suites even for the same game as people generally don't always bench the same places in games.

Canned benchmarks can provide lighter loads or more extreme loads too compared to in-game.

My 3070 was unplayable in-game, the beginning to the part the tank goes by,, my FPS here dropped to the teens on 1440P, RT on , maxed textures. Around 100w power used, 7.7GB Vram used, 99% GPU usage.

VRAM necked.


No clue about Intel, never owned one.

The 7900 series are beastly in games like Starfield.

My card does 3.2ghz solidly without crashing in anything.

Luck sometimes...

You reported that the game uses 7.7GB at 1440p for you, which is the exact same that TechPowerUp reported with the 3090. If the game only uses 7.7GB with a 3090 with 24GB of VRAM then that's the max amount the game needs to run at those settings. Given that the 3070 8GB has enough VRAM for 7.7 GB of usage, VRAM should not be the cause of the problem you are describing unless there's some weird partitioning issue going on with like the 970 or some bug that's causing this. Really just having more VRAM removes the potential for issues like this alogether.

 

[Deleted member 234997]

You reported that the game uses 7.7GB at 1440p for you, which is the exact same that TechPowerUp reported with the 3090. If the game only uses 7.7GB with a 3090 with 24GB of VRAM then that's the max amount the game needs to run at those settings. Given that the 3070 8GB has enough VRAM for 7.7 GB of usage, VRAM should not be the cause of the problem you are describing unless there's some weird partitioning issue going on with like the 970 or some bug that's causing this. Really just having more VRAM removes the potential for issues like this alogether.

It's not enough my friend, no clue why you are defending it so much if you own a 4090 anyway.

 

[3DVCash]

Guru3D reports the amount of VRAM actually used, stop with the nonsense semantics argument.

Allocated VRAM is VRAM "actually used". If you're talking about *dedicated* VRAM, even that isn't an exact measurement of a game's requirements.

It's loosing 23% of it's performance relative to the 73% the 3070 is loosing. This disproves your theory as clearly we can see an 8GB card here that doesn't take a 70% hit. Whether it's subjectively performing "fine" is irrelevant here, it's not having the dip your theory predicated.

Your theory of which had more sources disproving it to begin with, two other publican's numbers contradict this.

...That doesn't disprove anything. It's a totally different architecture than Nvidia/AMD's cards *and* relies more heavily on ReBar than both. Perhaps it leverages it in a way that mitigates the performance hit better in VRAM constrained scenarios, but even it is still disproportionately impacted. Meanwhile you ignore the clear apples to apples comparison with the 4060Ti 16GB vs 8GB staring you right in the face.

I have no idea how the two other "publicans" ran their benchmarks or what part of the game they tested. Maybe they were run in an area with a lighter VRAM load. Maybe they didn't properly apply updated settings between runs. Who knows??? But I've already shown you FOOTAGE of the built in benchmark turning into a slideshow under those conditions... in addition to, y'know, actually playing the game on an 8GB card myself... in addition to the benchmarks showing an obvious VRAM bottle neck on the site you're currently posting on. Clearly none of that matters though.

Pretty clear at this point I'm wasting my time with you, so you go ahead and believe whatever you want, alright. Just bizarre.

 

[Beginner Macro Device]

I'm just here to utter an obvious thing. More stuff beats less stuff.

You can argue all you please but 8 GB GPUs are no more sufficient for ultimate gaming even at 1080p. Yes, they give you very high FPS (given it's a 3060 Ti at least and not some little cheeky RX 570) if you turn some settings down from Ultra to more reasonable values... But 2010s are gone, so is the age of "8 GB is more than enough for anything."

Want a killer 1080p GPU? Get 12 GB. Or more. Won't hurt whatsoever.
Want a killer 1440p GPU? Get 16 GB. At least. The more the merrier.
Want ultimate 4K experience? Get a job 20+ GB.

 

[3DVCash]

It's not enough my friend, no clue why you are defending it so much if you own a 4090 anyway.

Right? Bro is really digging in his heels on something he has no experience with and that doesn't even affect him. Is this TPU or Reddit?

 

[80-watt Hamster]

Right? Bro is really digging in his heels on something he has no experience with and that doesn't even affect him. Is this TPU or Reddit?

It's a discussion forum. We discuss things. AFAICT, they're not denying your experience, but proposing a different cause of the issue based on other evidence available. Are they right? Short of somehow independently replicating the examples shown, that's difficult to prove.

 

[Deleted member 234997]

It's a discussion forum. We discuss things. AFAICT, they're not denying your experience, but proposing a different cause of the issue based on other evidence available. Are they right? Short of somehow independently replicating the examples shown, that's difficult to prove.

People who have owned the 3070 on 1440PO or higher know for a fact some games cap out the Vram and performance is lost.

But to say it is the end of the world is an absolute lie, I mean just turn down settings?

 

[3DVCash]

It's a discussion forum. We discuss things. AFAICT, they're not denying your experience, but proposing a different cause of the issue based on other evidence available. Are they right? Short of somehow independently replicating the examples shown, that's difficult to prove.

Well first he claimed upscaling doesn't lower VRAM requirements. That's already been proven false.

Then he claimed there was no situation where upscaling relieved a VRAM bottleneck and made otherwise unplayable settings playable. Also false. Far Cry 6 was just one fairly easy example to point to.

Now he's shifting the goalposts, again. It's a lil ridiculous. It's really not hard to find videos or benchmarks of 8GB cards choking at 4K with high res textures and RT enabled. Pretending like we have no idea what's going on is just silly.

 

[Fizban]

Amazing argumentation - one remake has poor textures, so no game has better graphics than 10 years ago... right...

And graphics is alot more than textures, genius.

The game also has nothing resembling PS3 textures. People are just really fond of gross hyperbole when attacking a game's graphics.

No, it's showing utilization as the chart says right at the top.



Again though other reviews don't show these numbers and as that very chart shows, the 8GB A750 is performing fine. If 8GB is the problem then why is an 8GB card pulling so far ahead?

Completely different architecture, it's very obvious the issue is VRAM, unless you believe the A750 is double the performance of a 3070 Ti. (Spoiler: It's more in line with a 3060.)

I'm just here to utter an obvious thing. More stuff beats less stuff.

You can argue all you please but 8 GB GPUs are no more sufficient for ultimate gaming even at 1080p. Yes, they give you very high FPS (given it's a 3060 Ti at least and not some little cheeky RX 570) if you turn some settings down from Ultra to more reasonable values... But 2010s are gone, so is the age of "8 GB is more than enough for anything."

Want a killer 1080p GPU? Get 12 GB. Or more. Won't hurt whatsoever.
Want a killer 1440p GPU? Get 16 GB. At least. The more the merrier.
Want ultimate 4K experience? Get a job 20+ GB.

Eh, 16 GB has honestly served me just fine for 4K with literally '1' exception. Might not be the case for a lot longer though. Current setup is a 4090 Mobile with a 4k/144 Hz display, 4090 Mobile is roughly the performance of a 4070 Ti, but with the VRAM quantity of a 4080. It's 4080 silicon, but with slower VRAM, and a wattage cap, but the overall performance is closer to a 4070 Ti, except in very specific scenarios where it shreds the 4070 Ti. ie. Anytime 12 GB VRAM isn't enough I crush the 4070 Ti, and in things like path-tracing where the number of tensor cores seems to matter far more than the clock speed of the GPU's cores I get much closer to 4080 performance than 4070 Ti performance.

Might not call it the ultimate 4k experience, but it's certainly very 4k-capable in the vast majority of games.

 

[Apocalypsee]

Talking about game requirements and VRAM usage is quite interesting these days, couple of days ago my 3080 10GB died so I back with GTX 1070 8GB, the slowdowns is quite noticeable because it only have 38% of the power but unlike back then on early 2000 years where any sort of 'optimization' of the game will be mocked (ATi trilinear cheating, nvidia 3DMark03 scandal, HL2 DX9 mixed mode etc.) today it has become a feature with all this super sample thing so enabling AMD FSR on this game made quite a noticeable improvement, sometime up to 50% framerate improvement.

I have a budget gaming laptop with modest 4GB RTX 3050 which is the lowest RTX 3000 lineup with only 2048 shaders, 64 TMU and 32 ROP, plus 64 tensor cores and 16 ray tracing units (CPU is i5 11400H). I quite surprised how it performed in CP2077, without DLSS it gets around 30-40fps but set DLSS to balanced makes it jump to 60fps in certain situation and even more, that is at 1080p with everything on high settings (no ray tracing of course). RAM usage is high for course, laptop only have 16GB and game uses up to 11-12GB at this settings but hey there isn't any jarring stutter. So despite the low amount of VRAM, with resizable BAR, and using SSD for storage the performance impact can be reduced as data transfer between these three is better than back then with HDD.

My point here is, yes despite the high requirement of newer games, if you adequately informed, you can tweak these games to run well on any modern-ish hardware. With various article showing what setting impact the performance and which isn't, it won't take people a few minutes to get their favorite game to run well with their systems. Of course the quality is lower than using native, but at least you have the options to control how it looks like and performance you can expect from it.

But this tech is double edged sword IMO, it is good for lesser hardware to run newer games but it made hardware devs skimp on VRAM, come on 1070 was released back in 2016. 3070 Ti not supposed to have the same VRAM as 1070 two gens back.

 

[Fizban]

Talking about game requirements and VRAM usage is quite interesting these days, couple of days ago my 3080 10GB died so I back with GTX 1070 8GB, the slowdowns is quite noticeable because it only have 38% of the power but unlike back then on early 2000 years where any sort of 'optimization' of the game will be mocked (ATi trilinear cheating, nvidia 3DMark03 scandal, HL2 DX9 mixed mode etc.) today it has become a feature with all this super sample thing so enabling AMD FSR on this game made quite a noticeable improvement, sometime up to 50% framerate improvement.

I have a budget gaming laptop with modest 4GB RTX 3050 which is the lowest RTX 3000 lineup with only 2048 shaders, 64 TMU and 32 ROP, plus 64 tensor cores and 16 ray tracing units (CPU is i5 11400H). I quite surprised how it performed in CP2077, without DLSS it gets around 30-40fps but set DLSS to balanced makes it jump to 60fps in certain situation and even more, that is at 1080p with everything on high settings (no ray tracing of course). RAM usage is high for course, laptop only have 16GB and game uses up to 11-12GB at this settings but hey there isn't any jarring stutter. So despite the low amount of VRAM, with resizable BAR, and using SSD for storage the performance impact can be reduced as data transfer between these three is better than back then with HDD.

My point here is, yes despite the high requirement of newer games, if you adequately informed, you can tweak these games to run well on any modern-ish hardware. With various article showing what setting impact the performance and which isn't, it won't take people a few minutes to get their favorite game to run well with their systems. Of course the quality is lower than using native, but at least you have the options to control how it looks like and performance you can expect from it.

But this tech is double edged sword IMO, it is good for lesser hardware to run newer games but it made hardware devs skimp on VRAM, come on 1070 was released back in 2016. 3070 Ti not supposed to have the same VRAM as 1070 two gens back.

Yeah, I have a bit older laptop than yours, it's from 2019, and has a 1660 Ti, it gets solid performance in Cyberpunk as well.

Amazing performance? Not really, no. But very playable on a laptop that was under $1000, 4-5 years ago.

It might be able to reach closer to 60, with lower quality FSR presets, but it might ultimately just be held back by the CPU, as it's 2 gens older than the CPU in your 3050 laptop.

 

[Deleted member 234997]

Cyberpunk barely eats Vram due to most textures being lower quality and also bad poly count which is more akin to TF2 in some areas.
Ray tracing fixes a lot of the slight issues but at the same time barely shows much difference, and no real life still does not look like what Cyberpunk does. I don;t know it may be the art style but RT barely adds anything if I were to play that game myself from videos and the images.

It's a Nvidia owners hype because it also cost them the priveledge also so it's a game most Nvidia owners will use. Emotional, not at all objective. Though this is also ok as I am not objective too, visuals are subject to a users preference.


Now Vram?

Test a game that actually uses more than 8GB.

The guy did mention that I stated the 3070 at 1440P with maxed settings in FC6 capped at 7.7GB then told me the 3090 did too so it means my point is moot.
Well dips to 12 FPS seem to differ, I bet the 3090 never dipped that low.


I kinda wish I still had my 3070 on hand but it is merely for argumentation sake, it would end some of the debate.

Stutters were observed in Horzon Zero Dawn on maxed settings at 1440P too but back then I had a 5800X non 3D and just 16GB RAM (Still 3733mhz) so it could be part CPU too.

 

[Fizban]

Cyberpunk barely eats Vram due to most textures being lower quality and also bad poly count which is more akin to TF2 in some areas.

This is getting really silly now. It is in no way akin to TF2.

 

[Deleted member 234997]

This is getting really silly now. It is in no way akin to TF2.

Go up to things close up without cherry picking. Leave out the wide panning views like most cyberpunk players because they already know the game is ugly up close... and voila!

The game also has very low draw distance.

 

[Fizban]

Go up to things close up without cherry picking. Leave out the wide panning views like most cyberpunk players because they already know the game is ugly up close... and voila!

Wow, you're right.

I can't believe it took me so long to realize these two things look the same:

I'm so glad you showed me the light and made me aware of the error of my ways.

Or maybe you meant the ground?

Or am I supposed to run TF2 at 4K with 8x MSAA so it's using 11k textures? Or close my eyes?

 

[Deleted member 234997]

That's a texture, not poly count.

If you want to meme all day then we cannot have a good faith discussion.

 

[Shihab]

Go up to things close up without cherry picking. Leave out the wide panning views like most cyberpunk players because they already know the game is ugly up close... and voila!

You've just described every 3D video game ever...

 

[Fizban]

That's a texture, not poly count.

If you want to meme all day then we cannot have a good faith discussion.

You're not having a good faith discussion when you try to claim Cyberpunk looks like Team Fortress 2.