Question about Game Engines, Optimization and Framerates @ 4k

[DarkSideA8]

Looking at Guru3d's 6800xt review: AMD Radeon RX 6800 XT review - DX11: Unigine: Superposition (guru3d.com)

I see that several games at 4k, even with 3080 or 6800xt are pulling 60ish to sub 60 FPS numbers, while others are in the 100s. Note: these are without turning on RT. They're also all well above 60 at 1440p. At 4k:

• AC V, WD L, are both sub-60,
• RDR2,and ME are in the 60s
• SOTR gets up to the 80s
  • (note, not including MSFS on purpose).

I checked, and AC V and WD L use different engines, as do the others - so its not a problem of the baseline engine performance as far as I can tell. Or is it?

Beyond merely the increased load of the additional resolution, is there a facet to this of the game engine being optimized for 1440p as opposed to 4k?

If so, might we see future 'console/PC' titles being better optimized for 4k resolutions because the consoles now support 4k?

Or is it really as simple as 'can the card handle the load' which we are often led to believe?

 

[LukeTbk]

If so, might we see future 'console/PC' titles being better optimized for 4k resolutions because the consoles now support 4k?

Not necessarily, but one thing that will probably get more and better ported is dynamic resolution.

Console outputting 4K to tvs made most game have better and better dynamic resolution (those 4K games on console are often running around 1440p in action heavy or just harder to render moments), this on PC tend to not exist has much and has well, console can go under 1000p during "4K" gameplay if they need to.

for your example, on a 6800xt performance at 4K relative at 1440p

Assassin Creed: 55/90 = 61%
RDR2: 61/ 96 = 63%
SOTR: 78/ 146 = 53%

At first glance there seem to have nothing special about Tomb Raider and 4K (worst drop if anything), the game simply play faster at all resolution at high setting.

 

[Dan_D]

Looking at Guru3d's 6800xt review: AMD Radeon RX 6800 XT review - DX11: Unigine: Superposition (guru3d.com)

I see that several games at 4k, even with 3080 or 6800xt are pulling 60ish to sub 60 FPS numbers, while others are in the 100s. Note: these are without turning on RT. They're also all well above 60 at 1440p. At 4k:

• AC V, WD L, are both sub-60,
• RDR2,and ME are in the 60s
• SOTR gets up to the 80s
  • (note, not including MSFS on purpose).

I checked, and AC V and WD L use different engines, as do the others - so its not a problem of the baseline engine performance as far as I can tell. Or is it?

Beyond merely the increased load of the additional resolution, is there a facet to this of the game engine being optimized for 1440p as opposed to 4k?

If so, might we see future 'console/PC' titles being better optimized for 4k resolutions because the consoles now support 4k?

Or is it really as simple as 'can the card handle the load' which we are often led to believe?

It's a simple matter of how the game was designed. The engine isn't really the issue so much as the graphics. Polycount, texture resolution, map sizes, lighting and other factors determine how demanding a game is at a given resolution.

 

[LukeTbk]

Beyond merely the increased load of the additional resolution, is there a facet to this of the game engine being optimized for 1440p as opposed to 4k?

For that specific part, it could have changed since I worked on a small indie 3d engine (OpenGL 2.0 type of days, with only vertex and pixel shaders)

3D engine are quite resolution agnostic by default everything is placed in world not in pixel term but say in meters, an object is 2 meter long, the vertex is a 2 meter in X, 2.2 meter in Y, 2.12 meter in Z that why that for unlike 2D games or the UI everything 3d can usually play at all resolution without problem (back in the CRT days that was quite common to change to 532 to 640 to a large variety of resolution, LCD brought the concept of native one), the world does not know the resolution of the output until the pixel shaders

Rasterization is when the 3d theoretical world get flattened into the 2D screen in which pixels start to exist, now for every pixel an instruction from the pixel shaders will run and interpolation from the vertex one will be made.

You can make the 2D element (like texture, pre baked map) different for each resolution has a part of optimisation for them (say not use texture so big that they do not look better than smaller one at 1440p), level of details over distance algorithm as well, but 3D element tend to be quite agnostic of it.

 

[Dan_D]

Another thing that needs to be understood is that rendering a game at 1920x1080, 2560x1440, and 3840x2160 isn't a linear increase in pixel count. 1920x1080 is roughly 2MP while 3840x2160 is 8.2MP. It isn't as though going to 4K is only twice as demanding as 1080P is.

 

[DarkSideA8]

LukeTbk and Dan_D ( + others): I realize the thing I'm trying to get at:

In virtually every review over the years when two competing cards come out, there is inevitably one game or another where the 'generally lower spec' card outperforms the other. In the case of the article I posted in the OP, the 3080 generally outperforms the 6800xt (although not by much). But in DS, AC V, Strange, and FC - the 6800xt is ahead (again, not by much). This drives me to question what it is about the architecture of a game that makes it work better with one card or another. And if there is some factor (whether the engine or not) where software can be optimized in a way to be more efficient with a card at the demands presented at 4k? Is there any reason to think that with more 4k capable machines coming online that 4k will get 'better' / more efficient due to software improvements as studios start to write with 4k as the target?

(I get that 4k requires far greater rendering power than lower resolutions, and its interesting to read what you've written... I'm not ignoring it - but perhaps I'm on to something, or rather asking a question for which I don't know the proper terms.)

When I look at say Witcher 3 and see the frame drops between 1440 and 4k - that's clearly a matter of computational demand - 4k performance is roughly 60% of 1440 regardless of card. But when I look at MSFS (bad example, maybe, but its there) - the frame drop from 1440 to 4k is negligible. Borderlands 4k frames are roughly 60% of 1440, but then Far Cry is relatively close at 80%.

Is this purely a factor of how much the card is being asked to draw?

 

[Dan_D]

LukeTbk and Dan_D ( + others): I realize the thing I'm trying to get at:

In virtually every review over the years when two competing cards come out, there is inevitably one game or another where the 'generally lower spec' card outperforms the other. In the case of the article I posted in the OP, the 3080 generally outperforms the 6800xt (although not by much). But in DS, AC V, Strange, and FC - the 6800xt is ahead (again, not by much). This drives me to question what it is about the architecture of a game that makes it work better with one card or another. And if there is some factor (whether the engine or not) where software can be optimized in a way to be more efficient with a card at the demands presented at 4k? Is there any reason to think that with more 4k capable machines coming online that 4k will get 'better' / more efficient due to software improvements as studios start to write with 4k as the target?

(I get that 4k requires far greater rendering power than lower resolutions, and its interesting to read what you've written... I'm not ignoring it - but perhaps I'm on to something, or rather asking a question for which I don't know the proper terms.)

When I look at say Witcher 3 and see the frame drops between 1440 and 4k - that's clearly a matter of computational demand - 4k performance is roughly 60% of 1440 regardless of card. But when I look at MSFS (bad example, maybe, but its there) - the frame drop from 1440 to 4k is negligible. Borderlands 4k frames are roughly 60% of 1440, but then Far Cry is relatively close at 80%.

Is this purely a factor of how much the card is being asked to draw?

I don't think you are getting it. Simply put: Games, game engines, game visuals and graphics cards are rearely created equal. It comes down to differences in the physical architectures of each GPU as well as the performance of the selected game engine as well as the design choices of the studio which impact the game's performance. When game developers decide to make a game, they have an endless sea of choices when it comes to implementing the visuals of that game. They can choose what effects to use. Different game engines also have specific strengths and weaknesses when it comes to performance and visuals. Some games are well suited towards the multiplayer FPS genre, but ill suited towards single player RPG's. Frostbite is a legendary example of this. Also, specific physics technology can be licensed as well leading to some differences in how a game behaves. Everything a game developer chooses to do impacts performance one way or the other.

Game developers have to make choices about everything. Polygon counts, texture sizes, map sizes, etc. You can't remotely compare two different games to one another. Even when they use the same engines, they are not the same visually. Think of it this way. You have X number of polygons, X number of textures, and X amount of processing power to devote to physics etc. before no system can run the game with decent performance. Think of it as a performance budget. When you get close to using all of that budget, the game becomes more demanding. Use less of it, and the game will run on a potato. It will probably look bad as well, but that's the way it goes. MMO's never look as good as the latest FPS games do. It isn't that you couldn't theoretically make an MMO look as good, it's that the massive worlds take longer to craft so they end up dated by the time they release. But, stripping the time to make such a game out of the equation and you could reach the same polycount as an FPS game surprisingly fast. It's just used far differently. You have less detailed, albeit much larger game worlds.

This is why Destiny 2's instances are so freaking small. You never see more than a dozen players in most of the zones save for the main quest and social hub. The game also has smaller maps than something like Far Cry 5 would have. Yet, the game runs smoother and has a better feel to its shooting mechanics than just about anything else out there. Why? Well, it's because the developer made specific design decisions to allow the game to both look and run really good. However, the trade off is smaller maps and areas. A great deal more instancing takes place than you would see in a traditional MMO. Decisions that impact performance can also come down to lighting and a variety of other factors. You can't compare the Witcher 3 to Borderlands or Far Cry 5. The games use different engines and virtually everything about them is different. Even if the result achieves parity visually, the choices made to get their yield different results in terms of performance.

Take Ghost Recon Breakpoint for example. It's still a great looking game, but it doesn't exactly run all that well. Especially not in DX12. The game's world is massive and very detailed. That comes at a cost. Getting 120FPS at 4K just isn't possible with it. Simply switching from DX12 to Vulkan for rendering increases performance by 20%. Differences in API's and other factors all impact performance.

Different GPU architectures are better at specific things. For example: The 3080 and 3090 are better at raytracing than AMD's 6800XT and 6900XT cards are. Naturally, anything with ray tracing will favor NVIDIA cards. The cards are not all equal in other respects either, but one game may work better due to being optimized for a given card's feature set. Optimized isn't even necessarily the right word as much as leveraged might be. You can make choices in the design of a game that favors performance aspects of a given card which might not work as well on another. One thing you have to realize though, is that the games are ultimately built for an API, not the cards directly. However, game developers can make choices with the game's design and visuals which they know will leverage a specific card better than another. For the most part though, they develop for the lowest common denominator. IE: They build for somewhat older cards, but do tend to scale some of the game's visuals for specific newer generation hardware. Mostly at the behest of AMD or NVIDIA, who give money to do so and slap their logo on the game.

Moreover, NVIDIA and AMD have vastly different design approaches. Their architectures may be used for similar things, but they are about as different as they can be. As a result, one option may excel at one game while faltering in another. It's been this way since the dawn of 3D gaming. The cards aren't equal and that's why you will see different performance in the same game. Why XYZ game looks as good to you as another but performs worse often comes down to small design decisions you don't see. Crysis 1 for example has several design choices which negatively impacted performance at the time of its release, but contributed heavily to the game still looking better than anything else at the time and aging surprisingly well. Yet some aspects of the game don't look that good because those were design decisions made to help improve performance somewhat. There is a good video on Youtube about that, but I'm too lazy to go find it for you.

More specifically:

LukeTbk and Dan_D ( + others): I realize the thing I'm trying to get at:

In virtually every review over the years when two competing cards come out, there is inevitably one game or another where the 'generally lower spec' card outperforms the other. In the case of the article I posted in the OP, the 3080 generally outperforms the 6800xt (although not by much).

GPU designs aren't equal. Therefore, their results will not be 100% equal when comparing them in the same game. Also, I wouldn't necessarily call the 3080 "lower spec" to the 6800XT aside from their RAM configurations. But then again, you need to understand that RAM does not equate to speed. A larger frame buffer means jack shit if its not being utilized. Beyond that the architectures are vastly different and not necessarily directly comparable. Sure, you can count ROPS, etc. but that doesn't tell the whole story.

But in DS, AC V, Strange, and FC - the 6800xt is ahead (again, not by much). This drives me to question what it is about the architecture of a game that makes it work better with one card or another. And if there is some factor (whether the engine or not) where software can be optimized in a way to be more efficient with a card at the demands presented at 4k? Is there any reason to think that with more 4k capable machines coming online that 4k will get 'better' / more efficient due to software improvements as studios start to write with 4k as the target?

This was primarily what I addressed above, so I'll leave it at that except to say, it isn't a matter of software efficiency improving the 4K experience. On consoles, they'll employ specific trickery as well as specific targeted optimizations which will allow for better visuals at a lower performance cost over time. This is a natural progression as developers learn the hardware and begin to utilize it better over time. On the PC side, games are largely developed for the lowest common denominator. However, games as I said are usually scalable to utilize newer features on newer, higher end cards to an extent. That being said, on the PC side games are developed or ported to an API. They do not get the same level of targeted optimization that consoles do. This is why more can be done with less on the console side. Even when consoles launch with relative parity to the PC, the PC hardware life cycle is much quicker than the consoles, leading to the PC surpassing the consoles in six to twelve months.

(I get that 4k requires far greater rendering power than lower resolutions, and its interesting to read what you've written... I'm not ignoring it - but perhaps I'm on to something, or rather asking a question for which I don't know the proper terms.)

When I look at say Witcher 3 and see the frame drops between 1440 and 4k - that's clearly a matter of computational demand - 4k performance is roughly 60% of 1440 regardless of card. But when I look at MSFS (bad example, maybe, but its there) - the frame drop from 1440 to 4k is negligible. Borderlands 4k frames are roughly 60% of 1440, but then Far Cry is relatively close at 80%.

Is this purely a factor of how much the card is being asked to draw?

Again, I cover this in detail above. The short version is that 4K is 8.3 million pixels versus 1080P's 2.1 million. It's not a linear progression but rather a logarithmic one. 4K is vastly more demanding than 1080P is. But differences in a game visuals, engines, physics, world sizes, etc. all factor in. You can't compare Assassin's Creed Odyssey to Far Cry 5. You can't even compare games with the same exact engines and expect the same results. Each game is unique.

 

[LukeTbk]

This drives me to question what it is about the architecture of a game that makes it work better with one card or another.

Really do not know enough about massive game with latest engine

But video card has difference in discrete way not just overall strength even just looking at specs, infinity cache vs faster memory bandwidth with gddr6x,

Ram bandwidth
6800xt: 512 GB/s
3800: 760.3 GB/s

Is one potential reason that we can see a shift in relative performance when memory use become massive like in 4K gaming.

Other difference (very theoric, GPU are good at wasting potential maximum usage):
3800 vs 6800xt
Pixel rate: 164.2 GPixel/s vs 288.0 GPixel/s
Texture rate: 465.1 GTexel/s vs 648.0 GTexel/s
FP16 (half): 29.77 TFlops vs 41.47 TFLOPS
FP32 (float): 29.77 TFlops vs 20.74 TFlops
FB64(double) : 465.1 GFLOPS vs 1,296 GFLOPS

L1 Cache: 128 KB vs 128 KB
L2 Cache: 5 MB vs 4mb
L3 cache: None vs 128 MB

How well a game take advantage of the L3 Cache and so on could affect how well it goes on different architecture.

It is easy to loose calculation on a GPU, they love to do the exact same operation on a giant set of data with a lot of parallelisation, it is easy to have many core doing almost nothing if it is not well feed, that in part why theoretical TFlops do not track performance that well specially among different architecture.

 

[DarkSideA8]

I can't 'like' the above posts enough. Thank you very much for taking the time to write all of that!

Dan_D So the part about the budget was when the bell went off. I also did not know/think about that they're writing to and API, and forgot that they have to appeal to the generic audience (ala WOT, which has been written with RU base average computers rather than NA/EU). Combined with what LukeTbk writes, I can see there's budgeting and decisions all around that impact performance.

I'll have to re-read and reflect to figure out why I got so far afield on this. It just stuck in my head for some reason - and I appreciate your time in helping me get perspective.