Will GPUs with 4GB VRAM age poorly?

[DavidGraham]

Ghost Recon Wildlands consumes about 5GB of VRAM @1080p Ultra, FuryX drops to the level of 980/1060/480 at these settings:

http://gamegpu.com/action-/-fps-/-tps/ghost-recon-wildlands-test-gpu

https://www.techpowerup.com/reviews/Performance_Analysis/Ghost_Recon_Wildlands/5.htmlhttp://gamegpu.com/action-/-fps-/-tps/ghost-recon-wildlands-test-gpu

 

[Malo]

Well it is an extremely demanding game. You wouldn't run it at Ultra anyway unless you had a 1080 or better. Maybe a 1070.

 

[Silent_Buddha]

http://gamegpu.com/action-/-fps-/-tps/ghost-recon-wildlands-test-gpu

https://www.techpowerup.com/reviews/Performance_Analysis/Ghost_Recon_Wildlands/5.html

It's not the memory amount that's dragging it down to those levels. It also happens at Very High 1080p which only reserves 3083 MB of memory.

What's interesting here is that as resolution increased, and hence memory usage increases, Fury X pulls away from both the 390, 8 GB and 480 8 GB. As well, despite it having less memory than the 1060, it pulls ahead of that as well as resolution and memory use increase.

To make things more interesting. With Very high, it doesn't use more memory until it hits 4k. Yet when the Fury X is supposedly limited by it's amount of memory it finally passes the 980 TI which isn't limited by memory.

IE - this isn't exactly showing how 4 GB is a limitation.

Ultra 1920x1080 (5067 MB memory reservation) shows the same relative performance difference as Very High 1920x1080 (3083 MB) between the 390, 480, and Fury X. Although the Fury X does drop every so slightly more. 38.3%, 37.9%, 39.4% respectively.

Basically the bottleneck is shifting all over the place depending on quality setting and resolution. What it looks like to me is that it can deal relatively well with memory usage over 4 GB, but as the amount increases, it's going to have some effect regardless of how well it can manage memory usage over 4 GB.

Regards,
SB

 

[DavidGraham]

Could this be another geometry bottleneck? ala Forza Horizon 3?

What's interesting here is that as resolution increased, and hence memory usage increases, Fury X pulls away from both the 390, 8 GB and 480 8 GB. As well

Well, this can happen due to the pressure on computing units, the 480 might have 8GB of RAM that is capable of holding 4K data, but it can never have the processing power to render 4K at high enough fps. The fps drop due to lack of processing power on the 480 is bigger than the fps drop on the VRAM starved FuryX. It's a frequent phenomenon. I think I've seen it multiple times already.

 

[DavidGraham]

Looking at OverClock3D review, VRAM usage is barely touching 3GB at Very High preset, yet Fury X is barely ahead of the 480, This is definitely a geometry bottleneck, just like Forza Horizon 3, Anno 2205, AC Syndicate and Fallout 4.

Spoiler: PGH Links

Here are alternative links to Forza Horizon 3, Anno 2205, AC Syndicate and Fallout 4 from PCGamesHardware

 

[DavidGraham]

Is this another geometry bottleneck? According to PCGH, 980Ti is 40% faster than FuryX in Mass effect Andromeda @1080p, FuryX is barely faster than a 480 or 1060. This is despite AMD releasing a game ready driver for Andromeda and despite VRAM usage not exceeding 3GB @1080p.
http://www.pcgameshardware.de/Mass-...55712/Specials/Mass-Effect-Andromeda-1223325/
According to GameGPU 980Ti is 23% faster @1080p. The only cards that do well against NV are Polaris cards.
http://gamegpu.com/action-/-fps-/-tps/mass-effect-andromeda-test-gpu

 

[Malo]

Is this another geometry bottleneck? According to PCGH, 980Ti is 40% faster than FuryX in Mass effect Andromeda @1080p, FuryX is barely faster than a 480 or 1060. This is despite AMD releasing a game ready driver for Andromeda and despite VRAM usage not exceeding 3GB @1080p.
http://www.pcgameshardware.de/Mass-...55712/Specials/Mass-Effect-Andromeda-1223325/
According to GameGPU 980Ti is 23% faster @1080p. The only cards that do well against NV are Polaris cards.
http://gamegpu.com/action-/-fps-/-tps/mass-effect-andromeda-test-gpu

The game heavily uses tessellation which is only able to be turned off by setting terrain to medium or lower quality. The tessellation levels are quite high which as we know favors Nvidia architecture and Polaris can better handle the levels of geometry.

 

[Ike Turner]

The game heavily uses tessellation which is only able to be turned off by setting terrain to medium or lower quality. The tessellation levels are quite high which as we know favors Nvidia architecture and Polaris can better handle the levels of geometry.

I can confirm that the Fury X (and all other GCN GPU apparently given the results) is definitely not running to its max potential here and it's not a 4GB Vram issue given what the ram usage is. The latest drivers are aimed at improving performance on Polaris only but also include a tessellation profile for the game which should theoretically help the Fury X too I guess. But frankly I can't see much difference in polycount even compared to the Xbox version.

 

[Malo]

But frankly I can't see much difference in polycount even compared to the Xbox version.

You couldn't really see much difference in the jersey barrier either

 

[Ike Turner]

You couldn't really see much difference in the jersey barrier either

Nvidia's proudest TWIMTBP moment!

 

[CarstenS]

Is this another geometry bottleneck? According to PCGH, 980Ti is 40% faster than FuryX in Mass effect Andromeda @1080p, FuryX is barely faster than a 480 or 1060. This is despite AMD releasing a game ready driver for Andromeda and despite VRAM usage not exceeding 3GB @1080p.
http://www.pcgameshardware.de/Mass-...55712/Specials/Mass-Effect-Andromeda-1223325/
According to GameGPU 980Ti is 23% faster @1080p. The only cards that do well against NV are Polaris cards.
http://gamegpu.com/action-/-fps-/-tps/mass-effect-andromeda-test-gpu

Important note: The PCGH benchmarks were done without the lower- tessellation profile. In the text, there's mentioned how much fps this costs.

At 1080p on a Fury X:
40/47,2 (min/avg) -> 44/53,2 with "optimized" profile. With tessellation manually disabled via RSCE: 51/59,5 fps.

And on an R9 390 (also 1080p)
30/35,2 -> 34/41,6 fps

So IMHO the tessellation adjustment in AMDs driver is a nice knob to have in order to selectively improve game performance. It should, however, remain solely in the users discretion to enable it.

 

[DavidGraham]

The game heavily uses tessellation which is only able to be turned off by setting terrain to medium or lower quality. The tessellation levels are quite high which as we know favors Nvidia architecture and Polaris can better handle the levels of geometry.

Important note: The PCGH benchmarks were done without the lower- tessellation profile. In the text, there's mentioned how much fps this costs.

Thanks, was wondering why AMD bothered with the Optimized Tessellation profile this time, now it makes sense.

Presumably, this makes it the 5th game in 3 months to take a hit on Fury cards due to a geometry bottleneck.

Conan Exile
980Ti is 45% than FuryX @1080p, 28% faster @1440p! 980 is as fast as FuryX at both resolutions!
http://gamegpu.com/mmorpg-/-онлайн-игры/conan-exiles-test-gpu

ARK Survival
98Ti is 25% faster than FuryX @1080p, the only resolution that matters.
http://gamegpu.com/mmorpg-/-онлайн-игры/ark-survival-evolved-test-gpu

Styx Shadow Of Darkness
980Ti is 36% faster than FuryX @1080p, 34% faster @1440p, and 22% faster @4K! Even a regular 980 is almost as fast as FuryX @4K!
http://gamegpu.com/rpg/роллевые/styx-shards-of-darkness-test-gpu

Ghost Recon Wildlands
980Ti is 28% faster than FuryX @1080p and @1440p!
http://gamegpu.com/action-/-fps-/-tps/ghost-recon-wildlands-test-gpu

But frankly I can't see much difference in polycount even compared to the Xbox version.

IMHO, there is quite a difference in the terrain of several planets available in the game. Rock formations, mountains and tress are affected. I wish I could go back and take screenshots but my 10-hour trial is now over. Anyway this option is reminiscent of the Tessellation technique used in the like of Battlefield 3 and Battlefield 1. Though terrain is slightly more complex in Mass Effect.

 

[Alessio1989]

Lego Worlds is the new non-plus-ultra for the 4GB VRAM destroy race: when playing in huge worlds, 4GB VRAM is not enough :\

The game become suddenly unplayable, but at least it advises you gently with an in-game pop-up telling you to restart the game OR to lower the graphics and visual LOD settings.

 

[DavidGraham]

Lego Worlds is the new non-plus-ultra for the 4GB VRAM destroy race

Can I ask what GPU you are using? and at what resolution?

 

[Alessio1989]

Can I ask what GPU you are using? and at what resolution?

Sapphire R9 380X 4GB Nitro (with backplate)
Resolution: 2560x1080
Settings: al maxed, both quality and landscape LOD, DX11 mode.

Please note that Lego worlds is a procedural games where everything is made by bricks. On huge worlds, it is easy to saturate the VRAM pool, especially if you modify the landscape.

The game has been released last week, so there could be still some engine issues, epsecially if we consider that a lot of changes has been made since the early access program (some of them to make the game run good on consoles I guess)..

 

[DavidGraham]

Resolution: 2560x1080
Settings: al maxed, both quality and landscape LOD, DX11 mode.

I can't find much information about the game. Is it resolution sensitive or Geometry sensitive? Meaning does the VRAM usage go up quickly with resolution more than geometry or vice versa? Did you try reducing the resolution to see if it clears up enough space?

 

[Alessio1989]

I can't find much information about the game. Is it resolution sensitive or Geometry sensitive? Meaning does the VRAM usage go up quickly with resolution more than geometry or vice versa? Did you try reducing the resolution to see if it clears up enough space?

Most of the VRAM usage is related to the dimension of the world, the distance of the LOD and how much you alter the landscape. The graphics quality settings does not impact so much on performance.
I do not have any detailed informations about how the engine works, but I guess it does not use a simple vertex-geometry approach to store the modifications you do to the world. I notice this issues only today, so this issue could be just a bug introduced with the last patch.
The game also have a simple profiler showing GPU and CPU timings, but nothing more.

EDIT: as for performance, only shadow quality and LOD distance impact a lot on performance, everything else (ambient occlusion, blur, FSAA, characters shadows quality, geometry quality, texture quality, depth of field etc...) does not affect so much the framerate.

EDIT2: to be clear, you need a decent CPU, most of times I am CPU bound (running on an old Nehalem i5 750 machine, +50% overclock on fsb and final frequency). I also observed over 8GB of commited memory on RAM, and over 4GB of working set. So, at least 16GB of RAM is recommended for huge worlds.

 

[-Sweeper_]

Apparently Fury can't do NieR at 4K

Spoiler

''In these settings appropriate measure FPS showed a GeForce GTX 1080. Video cards with a volume of less than 6 gigabytes of video in this mode, go into a stupor ...''

Granted, it's sub 30 FPS on most cards anyway.

 

[Malo]

I'm not sure that post even belongs in this thread.

 

[digitalwanderer]

Just my 2 cents on it real quick, I think 4GB is going to be fine for a while still if you're gaming at 1080. I think when you go larger res than that you should really be looking at more for any kind of "future proofing". (I don't believe in the term, but I get its incorrect usage.)