Alan Wake II on gtx 1080Ti

[SunRise-]

This mighty card will remain forever in our hearts !! gtx 1080ti R.I.P.!
Pascal R.I.P !

 

[Toothless]

Turning off mesh shaders allows the card to get a better powerpoint presentation.

 

[Shihab]

This is the first footage I've seen for this game, and honestly, I've got less issue with my 1080 not being able to run it properly even at low, than fact you start the game greeted with an overweight fellas bare arse...

 

[Vayra86]

Strange performance happens in that video.

All the way from >200 FPS to sub 30.
Its clear Nvidia isn't doing driver work for Pascal here, first of all. Its also clear there is a certain graphical feature / LoD change that just kills any and all performance. Might be mesh shader related?

Another thing that's clear is that the IQ you get at low on this card makes no sense given its FPS. Its one big blurry mess and various games look a lot better at low while giving better perf on top.

Let's keep in mind Nvidia REALLY wants to tell you all that Pascal is EOL. This game spells it out, and is Nvidia sponsored - and not just a little bit either.
But in the meantime, they release cards in Ada that can barely rival 1080ti raw perf and even carry substantially lower bandwidth. Go figure. If you read between the lines here, you could say there is a lot more in the tank for this 1080ti, its just not extracted.

 

[izy]

The difference mesh shaders make in terms of performance (i think this is on RTX 3060 Ti) :

Performance improvement of using mesh over vertex: 250% (GTX16), 400% (RTX20), 500% (RDNA2), 700% (RTX30)

 

[SunRise-]

When i started the game,and when i saw i have 57-60 fps on 1280x720 resolution i was so excited to play at decent resolution HD,but the happines end ,when i saw constantly .......18 -24 fps..This game is GPU killer in my case.A long time ago, i thaught The last of us is a gpu killer,but i was wrong, Alan wake II is a new Crysis era gpu killer from 2007....or is bad optimized ,but i don;t think so
For this game only rtx 30 and 40 series will work great.
Enjoy play this game.!

 

[izy]

When i started the game,and when i saw i have 57-60 fps on 1280x720 resolution i was so excited to play at decent resolution HD,but the happines end ,when i saw constantly .......18 -24 fps..This game is GPU killer in my case.A long time ago, i thaught The last of us is a gpu killer,but i was wrong, Alan wake II is a new Crysis era gpu killer from 2007....or is bad optimized ,but i don;t think so
For this game only rtx 30 and 40 series will work great.

My 2060 super does decent at 1080p DLSS Quality, medium-high quality, 1080ti just lacks mesh shaders.

 

[Shihab]

All the way from >200 FPS to sub 30.
Its clear Nvidia isn't doing driver work for Pascal here, first of all. Its also clear there is a certain graphical feature / LoD change that just kills any and all performance. Might be mesh shader related?

I only jumps high in pre-rendered sequences; the cutscenes (beginning, in the car), in the ones played as flashback thingy during gameplay (the jumpscare face, the smoke in dark scene effect, etc).
Actual, real-time rendering performance is around 30. Goes to near 60s at the begining.

 

[Beginner Macro Device]

Let's keep in mind Nvidia REALLY wants to tell you all that Pascal is EOL.

To be fair, aftermarket tells the same story. In some countries, Russia and Ukraine included, 1080 Ti is available for as little as $110—150. Some guys here earn more in a day. And if we exclude a couple shady (pun intended) games this is an absolute banger for its money. Especially if you pay stupid little money for electricity.

It's also valid to ditch Pascal in this particular game because no GPU of 1080 Ti's calibre performs well here. Even the best case scenario, an RTX 3060, is derping in 40s at native 1080p. Why optimising a game for GPUs which even theoretically can't deliver good experience? This makes development easier and more bug free.

This is still greedy but this time, greed is valid. Unlike that greed in VRAM challenged mid- and high-range GPUs of two latest generations. Unlike that greed in Ada GPUs pricing. Unlike that greed in disabling Frame Generator for everything that's not Ada...

 

[evernessince]

Strange performance happens in that video.

All the way from >200 FPS to sub 30.
Its clear Nvidia isn't doing driver work for Pascal here, first of all. Its also clear there is a certain graphical feature / LoD change that just kills any and all performance. Might be mesh shader related?

Another thing that's clear is that the IQ you get at low on this card makes no sense given its FPS. Its one big blurry mess and various games look a lot better at low while giving better perf on top.

Let's keep in mind Nvidia REALLY wants to tell you all that Pascal is EOL. This game spells it out, and is Nvidia sponsored - and not just a little bit either.
But in the meantime, they release cards in Ada that can barely rival 1080ti raw perf and even carry substantially lower bandwidth. Go figure. If you read between the lines here, you could say there is a lot more in the tank for this 1080ti, its just not extracted.

They pulled the same thing when they transitioned to CUDA based GPUs as well. They were pushing devs to implement GameWorks features that utilized CUDA like PhysX which just so happened to trash performance on their products that didn't support CUDA.

It's also valid to ditch Pascal in this particular game because no GPU of 1080 Ti's calibre performs well here.

The game performs very well on the consoles, it's really the PC port that is god awful.

 

[Beginner Macro Device]

The game performs very well on the consoles

Last time I checked, consoles didn't even remotely represent PC image fidelity. And I'm yet to hear of a good PC port of a console game.

 

[izy]

They pulled the same thing when they transitioned to CUDA based GPUs as well. They were pushing devs to implement GameWorks features that utilized CUDA like PhysX which just so happened to trash performance on their products that didn't support CUDA.



The game performs very well on the consoles, it's really the PC port that is god awful.

It drops under 60 fps with mostly LOW settings on consoles too

 

[SunRise-]

even on 640x360 resolution i have same fps

 

[izy]

even on 640x360 resolution i have same fps

It doesnt scale with resolution because of the mesh shaders.

 

[evernessince]

It drops under 60 fps with mostly LOW settings on consoles too

Mostly but not all low, the presentation is better than low on consoles. If you compared the OP's video to the console footage the console footage is vastly smoother and looks better.

Last time I checked, consoles didn't even remotely represent PC image fidelity. And I'm yet to hear of a good PC port of a console game.

In a like for like comparison the consoles seem to get better visuals and performance as compared to the 1080 Ti.

It goes without saying that PC will have a higher overall visual quality at higher settings but that's irrelevant to the topic. This is a like for like comparison. Not a $1,600 4090 vs $500 console one.

 

[izy]

Mostly but not all low, the presentation is better than low on consoles. If you compared the OP's video to the console footage the console footage is vastly smoother and looks better.

I said mostly , yeah, and in my opinion, if i understood this right, if you dont have mesh shaders on your GPU (the consoles have mesh shaders i guess) it will use the vertex shaders instead and those are very slow on processing meshes and that is the limiting factor. (edit: from my understanding you have to calculate some geometry that is mostly the same at every resolution and thats where the mesh shaders are good)

For example:
NVIDIA GeForce RTX 3070 - Mesh Shaders off: 52.86 FPS - Mesh Shaders on 357.07 FPS - Difference 575.5 % - https://www.3dmark.com/ms/3658

 

[evernessince]

I said mostly , yeah, and in my opinion, if i understood this right, if you dont have mesh shaders on your GPU it will use the vertex shaders instead and those are very slow on processing meshes and that is the limiting factor.

For example:
NVIDIA GeForce RTX 3070 - Mesh Shaders off: 52.86 FPS - Mesh Shaders on 357.07 FPS - Difference 575.5 % - https://www.3dmark.com/ms/3658

That's what I read as well in regards to mesh shaders, at least in this game. I have to wonder if that's down to yet again poor optimization of the alternative shader path when the card doesn't support mesh shaders.

 

[izy]

That's what I read as well in regards to mesh shaders, at least in this game. I have to wonder if that's down to yet again poor optimization of the alternative shader path when the card doesn't support mesh shaders.

I have no idea how they work but im guessing , as i said before, that the geometry that has to be processed by the mesh shaders will be processed by vertex shaders instead and that is the alternative path they were talking about, it makes the game work on older GPUs but at the cost of a big performance hit (guessing again - limiting factor being on how fast your vertex shaders can process that geometry that it should had been processed by mesh shaders and i also think that the geometry processed by mesh shaders has mostly the same processing power requirements at all resolutions ) , before release i heard that the game wasnt even supposed to start on GPUs without mesh shaders.

How im seeing it lets say 3070 was equal to 1080ti in performance on other games

NVIDIA GeForce RTX 3070 - Mesh Shaders off: 30 FPS - Mesh Shaders on 100 FPS
NVIDIA GeForce GTX 1080Ti- 30 fps

In this case if you could turn mesh shaders off on the RTX 3070 it should perform same as an GTX 1080TI , 30fps lets say and turning on mesh shaders on would make the 3070 go up to 100fps while 1080TI will have the same 30 fps as it doesnt have the extra hardware to increase the performance and the geometry still has to be calculated because thats how the game was made, also when then 3070 will have lower frames even with mesh shaders on the 1080TI will feel the hit too going under 30 fps .. and so on, its like turning ON NVIDIA Hair Works on AMD cards. (feel free to correct me if im wrong)

 

[Shihab]

I have no idea how they work but im guessing , as i said before, that the geometry that has to be processed by the mesh shaders will be processed by vertex shaders instead and that is the alternative path they were talking about, it makes the game work on older GPUs but at the cost of a big performance hit (guessing again - limiting factor being on how fast your vertex shaders can process that geometry that it should had been processed by mesh shaders ) , before release i heard that the game wasnt even supposed to start on GPUs without mesh shaders.

I doubt the underlined part. Vertex shaders are indeed -theoretically- slower, but, from what the video shows, the geometry isn't that complex for it to be a limiting factor.

My guess is that Remedy hackjob'ed a fallback code to handle converting MS's meshlets to standard vertex/index buffer combo that VS uses, the way Microsoft recommends, and they did it poorly. A last minute addition, you could say.
If my guess is correct, the bottleneck would be before VS part of the pipeline.

Not saying that they are at fault here. They did state the only support MS-enabled hardware. Writing the fallback code isn't part of the what what was promised, and it somewhat makes me optimistic that they would improve it and release a better performing version down the line.

 

[izy]

I doubt the underlined part. Vertex shaders are indeed -theoretically- slower, but, from what the video shows, the geometry isn't that complex for it to be a limiting factor.

My guess is that Remedy hackjob'ed a fallback code to handle converting MS's meshlets to standard vertex/index buffer combo that VS uses, the way Microsoft recommends, and they did it poorly. A last minute addition, you could say.
If my guess is correct, the bottleneck would be before VS part of the pipeline.

Not saying that they are at fault here. They did state the only support MS-enabled hardware. Writing the fallback code isn't part of the what what was promised, and it somewhat makes me optimistic that they would improve it and release a better performing version down the line.

You might be right but looking at 3dmark mesh shader benchmarks makes me think that its not much you can do if the game was built with meshes unless you rework the game and make a version of it that doesnt use meshes, the difference in performance is so big , 575% off vs on and its such a small increase on different tier of cards with it off.

NVIDIA GeForce RTX 3070 - Mesh Shaders off: 52.86 FPS - Mesh Shaders on 357.07 FPS - Difference 575.5 % - https://www.3dmark.com/ms/3658
NVIDIA GeForce RTX 3090 TI - Mesh Shaders off 64.71 FPS - Mesh Shaders on 605.38 FPS - Difference 835.6 % - https://www.3dmark.com/ms/9000

Edit: AMD seems to be even worse at processing meshes without mesh shaders (as you can see that 5700xt is doing very bad in this game, worse than GTX 10 series) :
AMD Radeon RX 6900 XT - Mesh Shaders off 35.42 FPS - Mesh Shaders on 592.53 FPS -Difference 1572.8 %

 

[N/A]

How im seeing it lets say 3070 was equal to 1080ti in performance on other games

3070 was equal to 2080 ti or 2070 / 1080 Ti +50%.

 

[izy]

3070 was equal to 2080 ti or 2070 / 1080 Ti +50%.

I was just giving an example to show my point, the numbers were not meant to be accurate.

 

[Shihab]

You might be right but looking at 3dmark mesh shader benchmarks makes me think that its not much you can do if the game was built with meshes unless you rework the game and make a version of it that doesnt use meshes, the difference in performance is so big , 575%.
NVIDIA GeForce RTX 3070 - Mesh Shaders off: 52.86 FPS - Mesh Shaders on 357.07 FPS - Difference 575.5 % - https://www.3dmark.com/ms/3658

Synthetic benchmarks are usually specifically crafted to single out the aspects they test. I wouldn't be surprised if the meshes in this test were subdivided four times to overwhelm geometry pipeline.
In a high-poly scene, yes, a mesh shader would be dramatically superior. But games don't typically come with insane amount of triangles per scene. And this game doesn't look different from its contemporaries.

 

[izy]

Synthetic benchmarks are usually specifically crafted to single out the aspects they test. I wouldn't be surprised if the meshes in this test were subdivided four times to overwhelm geometry pipeline.
In a high-poly scene, yes, a mesh shader would be dramatically superior. But games don't typically come with insane amount of triangles per scene. And this game doesn't look different from its contemporaries.

You may be right, maybe something like this could be implemented (see the link) but who knows if its possible and easy to do or not (for both AMD and NVIDIA) / or if its worth the extra work and how much will gain from it , if its just 20% -30% i dont think it makes sense to do it anyway. (going from 20 fps to 24 fps or something like that)

Mesh Shader Emulation - Tellusim Technologies Inc.

Tellusim Technologies Inc. Mesh Shader Emulation

tellusim.com

 

[Shihab]

You may be right, maybe something like this could be implemented (see the link) but who knows if its possible and easy to do or not (for both AMD and NVIDIA) / or if its worth the extra work and how much will gain from it , if its just 20% -30% i dont think it makes sense to do it anyway. (going from 20 fps to 24 fps or something like that)

Mesh Shader Emulation - Tellusim Technologies Inc.

Tellusim Technologies Inc. Mesh Shader Emulation

tellusim.com

Yeah. This uses the same approach I mentioned earlier. Merge meshlets into single vert and index buffers, pass to vertex shader as usual.
Weird results tho...