Computerbase - 3gb vs 4gb vs 6gb vs 8gb GDDR5 VRAM Frametime Testing

[maddie]

No it doesn't, HBM does not make the PCI-E bus magically faster. 4GB of VRAM is 4GB of VRAM, if it fits it fits, if it doesn't it goes to system memory via the PCI-E bus. That PCI-E bus runs at the same speed irrespective of your video card and what type of memory it has.

Yes, yes and yes.

Also pretty much irrelevant to what I was saying, but no matter. I guess this would take too many posts for the various viewpoints to be fully discussed.

 

[Bacon1]

Fury in Doom nightmare? If only the Fury manages to do well then we know it's HBM. If neither cuts it then we know it isn't. If both work fine then we just know it's an artificial limitation in the game.

Installed DOOM last night just for you.

Turning on Nightmare Shadows had little if no impact after restarting the game. It freaked out when first enabling it and dropped memory usage ~300mb and also halfed fps. Restarting the game it correctly showed ~200mb higher memory usage and ran at the same fps or within 1 (~75-78 fps @ Ultra/Nightmare 3440x1440p with 8x TSSAA in Vulkan).

Sadly Vulkan screenshot in steam is broken and fraps / dxtory didn't take screenshots so didn't have a way to do in game screenshots.

But I did waste a good chunk of time testing and proving that nightmare shadows work fine when forcing.

 

[Thinker_145]

Installed DOOM last night just for you.

Turning on Nightmare Shadows had little if no impact after restarting the game. It freaked out when first enabling it and dropped memory usage ~300mb and also halfed fps. Restarting the game it correctly showed ~200mb higher memory usage and ran at the same fps or within 1 (~75-78 fps @ Ultra/Nightmare 3440x1440p with 8x TSSAA in Vulkan).

Sadly Vulkan screenshot in steam is broken and fraps / dxtory didn't take screenshots so didn't have a way to do in game screenshots.

But I did waste a good chunk of time testing and proving that nightmare shadows work fine when forcing.

So we can say it is at least one big hassle which is in itself a negative point if we will have to do this in several games in the future.

Sent from my HTC One M9

 

[Bacon1]

So we can say it is at least one big hassle which is in itself a negative point if we will have to do this in several games in the future.

Big hassle?

I enabled the setting and restarted the game and it worked. It was just buggy when changing to it while in game.

The only thing that took a long time was verifying that was what was happening and testing many other settings and watching vram usage vs fps. Once I confirmed through multiple tests that it wasn't working right while changing in game it was simple to just restart the game after changing it and verified that it did work.

 

[UaVaj]

Hint, textures don't care about speed or bandwidth. They care about capacity. Doesn't matter how fast the VRAM is, if the textures routinely exceed the capacity, then the GPU will have to swap to system RAM, which incurs a massive latency penalty..

Larger VRAM pool means less texture swapping, which means smoother frame delivery. It ain't rocket science..

This was done at 4K, so AMD's poor CPU overhead in OpenGL shouldn't really matter. Just look at the RAM usage between the two, it's obvious the R9 Fury X is having to swap textures to system RAM way more often than the GTX 980 Ti. The Fury X is unplayable, but the GTX 980 Ti is still playable.

Looks like the Fury X wasn't even using it's full 4GB framebuffer though.

this Fury X memory management is as piss poor as 970's memory management. swapping texture to slower memory is never a good thing. thanks for clearing it up.

given that game setting. looks like it wants ~4.9GB. sadly frame are hovering at a ~40fps.

if you got time. run that demo again. only this time. reduce game setting to maintain frame at ~60fps.

 

[Bacon1]

this Fury X memory management is as piss poor as 970's memory management. swapping texture to slower memory is never a good thing. thanks for clearing it up.

given that game setting. looks like it wants ~4.9GB. sadly frame are hovering at a ~40fps.

if you got time. run that demo again. only this time. reduce game setting to maintain frame at ~60fps.

That demo was bad. He was using the machine to record the benchmark and was using opengl not vulkan.

I just used Ultra preset / Nightmare Shadows / 8x TSSAA / 16x AF and had maybe 1fps hit vs Ultra shadows @ 3440x1440 in vulkan. Was @ 75-78 fps on both last night.

 

[Carfax83]

this Fury X memory management is as piss poor as 970's memory management. swapping texture to slower memory is never a good thing. thanks for clearing it up.

given that game setting. looks like it wants ~4.9GB. sadly frame are hovering at a ~40fps.

if you got time. run that demo again. only this time. reduce game setting to maintain frame at ~60fps.

It wasn't me that was running the demo. It was some guy on YouTube.

 

[Carfax83]

That demo was bad. He was using the machine to record the benchmark and was using opengl not vulkan.

I just used Ultra preset / Nightmare Shadows / 8x TSSAA / 16x AF and had maybe 1fps hit vs Ultra shadows @ 3440x1440 in vulkan. Was @ 75-78 fps on both last night.

Great, now why don't you go ahead and enable the other nightmare setting for textures.. You do know there are two nightmare settings right? One for textures, and the other for shadows..

Enabling nightmare shadows whilst leaving textures on ultra doesn't help your argument at all.. Also the guy that ran that demo ran it at 4K. I doubt Vulkan would have made a difference at that resolution, as the game would be heavily GPU bound..

 

[Thinker_145]

Great, now why don't you go ahead and enable the other nightmare setting for textures.. You do know there are two nightmare settings right? One for textures, and the other for shadows..

Enabling nightmare shadows whilst leaving textures on ultra doesn't help your argument at all.. Also the guy that ran that demo ran it at 4K. I doubt Vulkan would have made a difference at that resolution, as the game would be heavily GPU bound..

Thanks for pointing that out. I didn't buy Doom yet so wasn't aware.

 

[Final8ty]

Big hassle?

I enabled the setting and restarted the game and it worked. It was just buggy when changing to it while in game.

The only thing that took a long time was verifying that was what was happening and testing many other settings and watching vram usage vs fps. Once I confirmed through multiple tests that it wasn't working right while changing in game it was simple to just restart the game after changing it and verified that it did work.

Indeed there are a lot of games that require a restart before the settings take.

 

[Bacon1]

Great, now why don't you go ahead and enable the other nightmare setting for textures.. You do know there are two nightmare settings right? One for textures, and the other for shadows..

Enabling nightmare shadows whilst leaving textures on ultra doesn't help your argument at all.. Also the guy that ran that demo ran it at 4K. I doubt Vulkan would have made a difference at that resolution, as the game would be heavily GPU bound..

Thanks for pointing that out. I didn't buy Doom yet so wasn't aware.

The nightmare texture setting doesn't actually increase texture quality, at least not from what I can tell when testing it. High, Ultra and Nightmare all look 100% identical. Nightmare dropped fps a small amount. I took screenshots with low, high, ultra and nightmare and the later 3 were all the same.

What it does it reserve VRAM for texture streaming (Hence the NAME of the option, Virtual Texture Page Size), so when you change areas there is less texture streaming and thus "pop" in.

You can see the issue here, where it shows up on the 970 often compared to the 480.

https://www.youtube.com/watch?v=FVb25eomcrI

So there is no point in trying to use Nightmare which just forces it to use over 4GB of VRAM to store textures that aren't in use as the IQ doesn't change.

http://imgur.com/a/A0zWd

And the popin happens even on a 1070: https://youtu.be/w5vLJHDuWjI

Fell free to do your own testing - Game is $30 right now: https://www.humblebundle.com/store/doom

 

[UaVaj]

nightmare texture does increase texture quality. of course a quality monitor capable of delivering that difference does help differentiate.
point is. most simply cannot tell the difference beyond high texture. rather that be their eyes or their hardware or both. hence why console is so popular with its "high" settings. and forget 60fps minimum too. lol.

also why block out the benchmark info on the screenshots?

anyway back to the pop in texture (aka texture lag). the guy in the video notices the pop in but has zero clue as to why. that pop in is not exclusive to doom. it happens to other games too. point is. if you dial in massive texture. you need a gpu with enough power to handle such texture in real time in preventing texture lag. that is simply the difference of 480 having more memory bandwidth than 970.



lastly. in order to have a healthy discussion about these particular topics. one need access to exotic resolutions and access to quad hi-end gpu to ensure that most bottlenecks are minimize. that is exclusive to the 2% of pc gamers.

 

[Bacon1]

also why block out the benchmark info on the screenshots?

So you can point out which is which since apparently I can't see the differences. Running in openGL since vulkan screenshots aren't supported and FPS isn't the focus of the test, but seeing if IQ did change or not was.

Obviously Nightmare should look much better than Ultra right? So I posted high, ultra and nightmare screenshots. Let me know which is which

 

[UaVaj]

you missing the point again.

nightmare texture are not across 100% of the game.
those particular screenshots (you took) is the same across all settings. that particular spot simply does not have the nightmare texture.
that is why frames did not tank.

nightmare texture are only available where the developer feel it will make a visual difference to the game play.
so go find a another spot that does have nightmare texture and you will see the visual difference.
rest assure frames will dip.

 

[Bacon1]

so go find a another spot that does have nightmare texture and you will see the visual difference.
rest assure frames will dip.

How about you send me a link to a save game next to that spot. I'm not going to waste my time looking for one. If it was an obvious difference then it would show up all over.

Also there was a slight FPS difference in that spot

 

[96Firebird]

We've gone from "4GB is enough!" to "those settings don't matter!"

Round and round we go.

 

[Bacon1]

We've gone from "4GB is enough!" to "those settings don't matter!"

Round and round we go.

PROVE to me they do. I've tested and shown no differences and its a "Well those settings don't matter there you have to go find some other place".

So provide me with a place where the texture IQ is different between Nightmare and Ultra and I'll gladly test it on my machine and report back the results.

Like I said before, this setting doesn't appear to change the actual texture quality, but how much textures are "cached" since it streams in textures. The higher you set it, the more textures it will hold in memory (page size).

http://www.tweaktown.com/guides/7757/doom-graphics-performance-tweak-guide/index5.html

 

[UaVaj]

Round and round we go.

round and round does NOT apply to the few of us that DO have access to exotic resolutions and access to quad hi-end gpu.

 

[Bacon1]

round and round does NOT apply to the few of us that DO have access to exotic resolutions and access to quad hi-end gpu.

Sorry forgot that 3440x1440 isn't exotic enough here. Amazing you respond to that trolling post and not mine asking you to prove that the setting does actually change IQ and where I can go to test it.

 

[Carfax83]

PROVE to me they do. I've tested and shown no differences and its a "Well those settings don't matter there you have to go find some other place".

You are completely missing the point. The reason why you were asked to run nightmare quality textures, is because we want to know whether the Fury's HBM has the "special sauce magic" that AMD fans says it does, which allows it to somehow bypass the normal limitations that other non HBM cards have when it comes to VRAM management..

But I'm sure you're going to come up with some excuse again..

 

[Carfax83]

BTW, Doom is just one game of many, and isn't even close to being one of the most egregious consumers of VRAM.. Deus Ex Mankind Divided is perhaps the greatest of them all. In this video, the guy benchmarks a Radeon Fury X vs a 980 Ti and a GTX 1080.....with ULTRA quality textures at 1080p.

Most of the comparisons are stationary, and during these, the Radeon actually looks quite capable considering it only has 4GB of VRAM. But at 3:57 is when it starts to get interesting, as that's when the uploader has Jensen moving through Prague

During this segment, you can visually see that even though the frame rate of the Fury X might match the 980 Ti at times, the frame time variance is far less stable. You can actually SEE the microstutters. Also, the Radeon is using about 3.5GB more system RAM than the GTX 1080, which is obviously due to having to swap out textures often.

And then at 4:31, the Radeon starts to have some very noticeable stutters, whilst the 980 Ti and GTX 1080 coast by..

 

[Bacon1]

You are completely missing the point. The reason why you were asked to run nightmare quality textures, is because we want to know whether the Fury's HBM has the "special sauce magic" that AMD fans says it does, which allows it to somehow bypass the normal limitations that other non HBM cards have when it comes to VRAM management..

But I'm sure you're going to come up with some excuse again..

I'm not making up excuses.

I spent time testing, found no difference and then was told "Oh that isn't a valid test place, the nightmare textures only show up in specific spots".

So I asked "Where are those spots?"

And everyone is ignoring my request to find those spots and saying I'm trying to change the subject.

I'm asking you to provide me with a place to test or admit that my testing is valid and there is no difference.

 

[IEC]

I tried Doom with Ultra settings and also with both Nightmare settings turned on. I don't get any perceivable stutters, but the Nightmare settings seem to decrease my FPS without increasing the visual fidelity. Probably because you'd need a 5K+ monitor to take advantage of those.

 

[Carfax83]

I'm asking you to provide me with a place to test or admit that my testing is valid and there is no difference.

After doing some research, I doubt you'll find any differences because Doom uses virtual textures. The PC and consoles are using the same virtual textures broken up into tiles or chunks, but the PC is able to have better looking textures due to having more memory in the sense that the highest quality textures are loaded further away rather than up close.

Nightmare quality just pushes the distance further away from the camera for when the highest quality textures will be loaded relative to the ultra setting. This is why I never see texture pop in when playing Doom, unless it's after an initial load..

But, this is besides the point. This discussion isn't about whether Doom's nightmare settings increases texture quality. It's about whether the Fury X and the Fury can run nightmare quality shadows and textures without dropping frames or creating lots of stutters due to texture swapping.

 

[Carfax83]

I tried Doom with Ultra settings and also with both Nightmare settings turned on. I don't get any perceivable stutters, but the Nightmare settings seem to decrease my FPS without increasing the visual fidelity. Probably because you'd need a 5K+ monitor to take advantage of those.

See my explanation above. The texture resolution isn't any greater in nightmare quality, but the distance that the highest quality textures are loaded is significantly increased relative to ultra and high quality.

So this means that the game will theoretically look better over distance, as well as eradicating pop in with nightmare quality.