Digital Foundry Article Technical Discussion [2023]

[Silent_Buddha]

you’re evidence is a single point without the ability to actually test what is happening, but OK. That’s far from thorough testing as to why the shadows are like that.

C'mon, we all know that looking at a single static screenshot is the best way to determine if the lighting is dynamic GI influenced by atmospheric conditions or baked.

I mean the fact that he continually relies on finding the absolutely worst screenshots he can find and then tries to apply that as a blanket example for the rest of the visuals, already shows either an agenda, ignorance or just plain inability to admit that they are wrong.

I mean sure, everything in there in isolation is possible on last gen consoles. Just like everything in the new Ratchet and Clank game is possible on last gen consoles in isolation. It's when you take the whole that the argument falls apart.

The scale of Starfield which is so far above anything else released in gaming probably makes it difficult for people that don't have a good grasp of game technology to understand.

There are games that do some of what Starfield does better as you've pointed out. NMS has a larger galaxy and more planets. But it doesn't have even 1/100th of the hand crafted content. RDR2 likely has a similar amount of hand crafted content but the world is less than 1/100th the size of Starfield. Etc. Etc.

People that can't grasp that single concept are going to have a hard time conceptualizing just what Bethesda has accomplished with Starfield.

Graphically, yes, Starfield is never going to challenge for the top spot. But then there is no other game currently in existence that's going to challenge what Bethesda are doing with Starfield ... until Starfield 2.

And this considering the last Bethesda game I was a huge fan of was ... TES: Morrowind.

Regards,
SB

 

[DavidGraham]

It's almost like the texture streaming system is completely broken

It's probably the usual culprit, suboptimal VRAM management caused by inexperience with DX12 from the developer (it's their first DX12 title afterall), I am sure they fix it soon via patches. The game is set to have Ray Tracing post launch anyway.

 

[snc]

...
Graphically, yes, Starfield is never going to challenge for the top spot. But then there is no other game currently in existence that's going to challenge what Bethesda are doing with Starfield ... until Starfield 2.

Generaly agree with your post but would wait for a release of game before such a statement It can be truth as showcase showed potential but in the end it can be yet again too ambitious game with procedural worlds not that amazing and deep. Btw about 30fps, if 60 is not possible on consoles, why not at leas give 40fps mode for 120hz tvs?

 

[Subtlesnake]

I'll leave it to you to explain how this is "next-gen" graphics

If you take out the people, the materials and detailing look ok to me. The lighting looks a bit flat, but Bethesda probably don't have an effective indoor radiosity system. The 3rd shot looks pretty blurred, so I am not sure if it is YouTube compression or perhaps some kind of motion artifact.

The main focus of Starfield is in creating beautiful outdoor vistas with high fidelity (generated) terrain detail in the foreground that doesn't break down in the mid-range. As a consequence of being unable to pre-compute any of the lighting and using a procedural character generation system that doesn't seem to be quite fit for purpose, the worst looking parts of the game will be indoor rooms filled with people taken at oblique angles where you don't get to see all the material detailing. And of course it's no surprise that your first two shots are exactly that.

If Starfield was just a corridor shooter featuring rooms full of people and was still running at 30 FPS and 1296p, you would have a point. But when we go to the places in your screenshots on the PC, we are probably not going to see huge drops

The really challenging parts will likely be the city areas full of people, merging with the generated landscape. And of course you didn't post any screenshots of those areas, nor of any landscapes. It's familiar to these kinds of "graphics are PS2 quality" debates to post the worst possible screenshots and then compare them to a completely different kind of game on the prior generation. Which why I don't consider them good faith discussions. If you want to properly debate the merits of the graphics of a particular title, you need to consider the good and bad. Because games are full of compromises that often break down.

 

[hughJ]

Mass Effect II vs Frontier: Elite 2. Fight!

(Everyone wins, both are compelling games)

Yeah, I think Star Citizen, Starfield, and NMS, (and Elite Dangerous) are similar only if you boil them down to a one or two sentence elevator pitch. They're all going to differ in the degree of how authored, realized/simulated, contiguous, persistent, and shared vs. instanced they are. Different goals, different audiences, different platforms. If I want to explore planet surfaces and build bases then I play NMS, if I want to hook up my hotas and pilot spaceships then I play Elite Dangerous or Star Citizen. NMS has a lot of systems of planets, but the systems don't have stars or planetary orbits, so the illusion to me of a galaxy built of star systems devolves into being a bunch of instanced volumes and some spheres that you point yourself at and zip to/from.

I fully expect Starfield to comparatively come up short in some respects and succeed in others just as they all do. Given that it's not out yet I also expect there to be an ample amount of overshooting and undershooting in terms of promises and expectations. For whatever it's worth, the last Bethesda game I was happy with graphically was Oblivion, but even that was only after some aggressive config fiddling to stretch the LOD/zone radius well beyond default. As long as Starfield doesn't look utterly third-rate I'll probably be satisfied with it graphically.

 

[Shifty Geezer]

If you would take the time to look at that picture - which you won't - you would see that there are (at least) two light sources casting shadows. And as I'm sure you're aware, scattering effects can vary based on distance, layers of atmosphere passed through etc etc

I've had a good nose. I'm not seeing evidence of two light sources. All terrain shadowing is in the direction of the visible star. Only the arch seems to cast contrarily. There aren't secondary shadows or other-side illumination on the hills to indicate a second lightsource higher up and to the left either. Any second-star illumination would have be as bright as the first star given the cast shadow is of equal darkness and very obvious.

As a last curio...



The arch shadow in part doesn't match the hill shadow.

I guess there could be a second light source causing the subtle darker shadow bumps throughout the image, but these dark blobs are too subtle to match the arch's full shadow. There should be at least three distinct areas of brightness - both sources occluded, maximum darkness; one occluded; and neither occluded, full brightness. We're only seeing two.

 

[VFX_Veteran]

Wrong. Starfields planets are no where the size of Star citizens planet and we saw that in the showcase. I don’t know that one can argue that Starfields planets are even bigger than no man’s sky’s planets talk-less of Star citizen. It also doesn’t matter if they have more planets since most of them are pea sized in comparison.

How can you compare a planet in Starfield when you haven't explored it yourself?

All of this is irrelevant to the technical discussion at hand. All I read here is you trying to defend starfield by putting star citizen down while discussing no actual technical features of the game. Again, I’m not saying starfield is a bad game, I’m just saying that it’s not technically impressive.

But there isn't any technically impressive games out this generation other than CP2077 OD mode. All of these games are still using rasterization with RT features sprinkled here and there on top. Having said that, I think Starfield is doing a hell of a lot more than current gen games are doing technically due to it's scope alone - and that's assuming all games using tech from last generation.

I can pull out no less than 20 horrible scenes from that showcase and I wouldn’t even be trying. Just the facial animations alone are a huge eyesore.

Yes, the facial animations have to be procedural as there is way too much dialogue in the game. Most animations in games are procedural as well during gameplay.

Firstly, No Man’s sky is bigger in scale and that’s a fact. Secondly, if the game started development 8 years ago, then it should be far more performant than it is.

Are you insinuating that the longer a game is in development the most performant it will be?

 

[function]

I've had a good nose. I'm not seeing evidence of two light sources. All terrain shadowing is in the direction of the visible star. Only the arch seems to cast contrarily. There aren't secondary shadows or other-side illumination on the hills to indicate a second lightsource higher up and to the left either. Any second-star illumination would have be as bright as the first star given the cast shadow is of equal darkness and very obvious.

I think these areas are showing light coming from a light source to the left and above the area in the screenshot:



I'm pretty confident there are two light sources in this image. Even with a pretty scattered light source these areas shouldn't be lit the way they are by the partially risen sun / whatever in the middle of the screen.

This looks like a thin atmosphere with lots of dust in it - kind of like Mars. Blinding sun head on, but to the sides of this image you can see the stars and the darkness of space.

As a last curio...

View attachment 9068

The arch shadow in part doesn't match the hill shadow.

I guess there could be a second light source causing the subtle darker shadow bumps throughout the image, but these dark blobs are too subtle to match the arch's full shadow. There should be at least three distinct areas of brightness - both sources occluded, maximum darkness; one occluded; and neither occluded, full brightness. We're only seeing two.

Good spot. Definitely something going on with the shadow system. Perhaps some areas marked as shadowed are left alone for other / subsequently calculated shadow casting light sources under some circumstances? There's also something going on with the sharpness of shadows cast by the arch as you move further from the camera.

 

[snc]

Are you insinuating that the longer a game is in development the most performant it will be?

Thats interesting, do you insinuating that time of production does no corelating with quality of code/performance choice ?

 

[yamaci17]

It's probably the usual culprit, suboptimal VRAM management caused by inexperience with DX12 from the developer (it's their first DX12 title afterall), I am sure they fix it soon via patches. The game is set to have Ray Tracing post launch anyway.

Why is texture scaling is so horrible across so many games? It is incurable? :/ I'm sure most 8 GB users would be fine with degraded textures but not to the point they turn to PS3.

 

[iroboto]

I've had a good nose. I'm not seeing evidence of two light sources. All terrain shadowing is in the direction of the visible star. Only the arch seems to cast contrarily. There aren't secondary shadows or other-side illumination on the hills to indicate a second lightsource higher up and to the left either. Any second-star illumination would have be as bright as the first star given the cast shadow is of equal darkness and very obvious.

As a last curio...

View attachment 9068

The arch shadow in part doesn't match the hill shadow.

I guess there could be a second light source causing the subtle darker shadow bumps throughout the image, but these dark blobs are too subtle to match the arch's full shadow. There should be at least three distinct areas of brightness - both sources occluded, maximum darkness; one occluded; and neither occluded, full brightness. We're only seeing two.

The challenge is using this cut up video as proof that their RTGI and lighting system. If it were, it would be consistently busted. What you’re probably seeing is a mix of old and new builds in a single video.

As Alex said, you can’t test squat without the game in your hands. Edited video is edited video and they still have 3 more months to refine it further. We have no clue how old
Some footage is, but if it looked nice the trailer and video directors may have used it but that doesn’t mean it’s an accurate depiction of the release candidate.

 

[function]

Thats interesting, do you insinuating that time of production does no corelating with quality of code/performance choice ?

It depends on what the man hours are spent on. Adapting an engine? Adding new features? Allowing for new interactions?

Performance and scope and adaptation all tend to challenge each other.

And how do you judge "quality" of code? Performance? Readability? Maintainability? Adherence to studio practices? Stability?

 

[iroboto]

Thats interesting, do you insinuating that time of production does no corelating with quality of code/performance choice ?

Depends on what you’re trying to accomplish right?

There is a strong correlation if the product is underbaked. You cannot optimize forever and continually generate better results. Star Citizen is a fairly good example that increasing scope doesn’t work, because it has drawback effects of constantly requiring more hardware. Even if they went back to refactor their code for the next 3 years there’s an upper limit to how performant it will be.

Some loads are just too large to handle for certain hardware.

But let’s be real 3 years is not enough to finish star citizen

 

[snc]

Depends on what you’re trying to accomplish right?

There is a correlation only if the product is underbaked. You cannot optimize forever and continually generate better results. Star Citizen is a fairly good example that increasing scope doesn’t work, because it has drawback effects of constantly requiring more hardware. Even if they went back to refactor their code for the next 3 years there’s an upper limit to how performant it will be.

Some loads are just too large to handle for certain hardware.

But let’s be real 3 years is not enough to finish star citizen

Sorry star citizen is scam not proper game made by aaa studio :d

 

[cwjs]

Some loads are just too large to handle for certain hardware.

This is true, but it's hard to imagine anything about the experience starfield or star citizen provide that shouldn't be able to run at 240fps on an xbox in principle -- the idea that a big empty game world or some procedural planets to stream in magically makes your game fundamentally impossible to run fast is silly. Whether the studio culture is amenable to it or the core architecture of the tehcnology is designed to make that realistic without spending tens of millions of dollars in order to please like ~5% of their userbase is another question, of course, it's very reasonable for the game to ship at 30fps!

 

[see colon]

Correct me if wrong, but aren't real life planets connected by space?

@davis.anthony is right.

I'll leave it to you to explain how this is "next-gen" graphics

It doesn't look next gen. It looks this gen. In fact, it runs on this gen hardware.

 

[BitByte]

If you take out the people, the materials and detailing look ok to me. The lighting looks a bit flat, but Bethesda probably don't have an effective indoor radiosity system. The 3rd shot looks pretty blurred, so I am not sure if it is YouTube compression or perhaps some kind of motion artifact.

Firstly, I must say, I appreciate the cordial response. The materials in the 3rd shot look ok but, it’s not revolutionary or something we haven’t seen before…

The main focus of Starfield is in creating beautiful outdoor vistas with high fidelity (generated) terrain detail in the foreground that doesn't break down in the mid-range. As a consequence of being unable to pre-compute any of the lighting and using a procedural character generation system that doesn't seem to be quite fit for purpose, the worst looking parts of the game will be indoor rooms filled with people taken at oblique angles where you don't get to see all the material detailing. And of course it's no surprise that your first two shots are exactly that.

That’s a fair criticism to make but the beauty I see in the outdoor vistas is primarily art driven. The textures used are high quality but, the visuals in the outdoor shots are not better than existing games out right now in my opinion. The best I can do is describe the game as cross gen not next gen.

If Starfield was just a corridor shooter featuring rooms full of people and was still running at 30 FPS and 1296p, you would have a point. But when we go to the places in your screenshots on the PC, we are probably not going to see huge drops

The funny thing is that there are games out right now that are not corridor shooters that look better than Starfield. Like a lot better, not marginally better. I chose infinite warfare because it’s a space game from last gen that has similar visuals to starfield. It’s funny that you describe it as a “corridor” shooter when imo, the best visuals in the direct took place while shooting in a corridor or spending time in the ships interior.

The really challenging parts will likely be the city areas full of people, merging with the generated landscape. And of course you didn't post any screenshots of those areas, nor of any landscapes. It's familiar to these kinds of "graphics are PS2 quality" debates to post the worst possible screenshots and then compare them to a completely different kind of game on the prior generation. Which why I don't consider them good faith discussions. If you want to properly debate the merits of the graphics of a particular title, you need to consider the good and bad. Because games are full of compromises that often break down.

We saw new Atlantis city I think, not sure what it’s called and it was perhaps the least impressive part of the demo as far as I’m concerned. I’ve watched the starfield direct in full 4 times now and each time I watch it, I become less and less impressed. Foliage rendering, cloud rendering, explosions, terrain lighting, materials, the way animals/life forms are rendered, skin rendering and hair rendering are things that have serious issues. The more I watch, the more issues I see and I’m not even actively looking for it. Take the materials for example. Metallic, plastic and some cloth look ok while others display an unnatural plastic sheen to them that’s clearly not right. Or look at how some of the terrain floor has plastic sheen to it. Do not mistake my criticism for bad faith arguments. I truly am not impressed by these visuals. Others might be able to hand wave it away because Starfield is doing a lot. It’s as if there’s this belief that when you’re fighting some random npc on a random planet, starfield is wasting resource simulating what’s going on with npcs on new Atlantis.

 

[BitByte]

@davis.anthony is right.

It doesn't look next gen. It looks this gen. In fact, it runs on this gen hardware.

You know what i mean and I don’t think it belongs to this gen. It’s clearly cross gen.

 

[see colon]

You know what i mean and I don’t think it belongs to this gen. It’s clearly cross gen.

But it wouldn't be the worst looking game this generation. So I don't see what the problem is. Starfield doesn't have to be the best looking game ever. No Bethesda game has ever approached that title, so I wouldn't have expected Starfield to either.

 

[BitByte]

How can you compare a planet in Starfield when you haven't explored it yourself?

I’ve seen the direct and I know none of the planets have a diameter of 15000km that is fully explorable. Nothing they’ve show even remotely suggest that at all. If it turns out that I’m wrong, I’ll gladly put my hand up and acknowledge it.

But there isn't any technically impressive games out this generation other than CP2077 OD mode. All of these games are still using rasterization with RT features sprinkled here and there on top. Having said that, I think Starfield is doing a hell of a lot more than current gen games are doing technically due to it's scope alone - and that's assuming all games using tech from last generation.

I keep hearing scope used but all I see is a few handcrafted environments and then a bunch of procedurally generated maps with pointless missions and filler content. It’s a cool check box to have but, it’s not impressive to me. I won’t visit most of the planet as the “content” is an insulting use of my time. All of this is to say that I don’t consider procedurally generated content as impressive scope. Procedural generation has been used in the industry for a while and I don’t consider it to be scope. Does no man’s sky have the biggest scope ever because they created a deterministic procedurally generated system that could have millions of planets? Common, let’s be serious here.

Yes, the facial animations have to be procedural as there is way too much dialogue in the game. Most animations in games are procedural as well during gameplay.

That’s fine, but the system they’re using is objectively not good. We’ve seen better.

Are you insinuating that the longer a game is in development the most performant it will be?

No, what I was saying to the other poster is that say for example you were targeting a specific spec during development. During the development, you spend time optimizing your game to run on that machine. Then when the game is 75% done, you find out that the actual spec you’ll be releasing on is more powerful. All that optimization you had done previously would pay dividends. On a higher spec machine allowing you’d be able to run the game faster and better than you initially thought.