FAST Fusion - Switch 2 - Digital Foundry Analysis

I'm excited to play this after the patch. These devs have historically been great at listening to feedback and making updates, so I'm sure we'll get this to a great place eventually.

Honestly, the biggest takeaway from the video for me was that they were able to use DLSS to upscale to 4K. DF previously believed that would be impossible for a 60fps game, since their testing on similar hardware had a frame time cost of ~18ms to upscale to 4K.

That means that the Switch 2 has some kind of bespoke implementation of DLSS. It'll be really interesting to hear more about that eventually.
 
The game is great, but I'll wait until the patch. It looks horrendous docked. Those who arent seeing how bad it is need a trip to specsavers
 
The answer is the same as the one your girlfriend gives you during sex.

e69b41ff33ffaadd4661e41bbeaa6fe2.gif
Jokes on you, i dont have a girlfriend or do sex.
 
Last edited:
yeah, the image quality was conspicuously terrible here, especially in handheld mode. Now we know why, upscaled from 360p!
 
Docked
1080P | 60fps - 540p
1440P | 60fps - 504p
4K | 60fps - 648p
4K | 30fps - 720p

Handheld
1080p | 60fps - 360p

360p is diabolical

Those resolutions .... :messenger_grimmacing_ Switch 2 is not competing with Xbox Series S ....
 
I mean. i did notice the image is not super clean. But to be honest, with the fast action going on. It's kinda hard to get bothered by when you are trying to keep your eyes almost bleeding on the track ( especially when it's on high speed level ).

the game in general looks pretty, and it's really very addictive.

with that being said. i will continue playing it next week when its gets updated too
 
Regardless of it's look (which I still think looks great) game is fucking amazing! Plays so well. Is extremely fun and a really overlooked game in the Switch 2's launch. Some would say a racing game that is very mature even though supposedly Nintendo release "kiddy" racers...
 
You didn't play the game though did you?
I'm serious, you won't have time to look at any image artifacts, you risk crashing just quickly glancing down at your boost meter. With that insane speed there is so much you have to be laser focused at; curves, jumps, opponents, color boosts, pickups, different routes, obstacles, etc. And you quickly advance and get faster hover crafts and more complicated tracks. The game is awesome 👌
I've played the game and I easily notice those artifacts, it's like a very visible shimmer around all edges, game is beautiful but they stopped the ball with the IQ
 
Honest question man, is it a ..."game" game or is it more of a tech demo to show off the S2's capabilities ?

This looks like Rogue flight - a game which looked amazing (bought it, finished it) but it was very light gameplay wise and never played it again...

Cheers
At just 15$, it is worth it. Will you notice the artifacts? Most probably yes, it doesn't matter when you are going full speed. The courses are decent, and I haven't tried it yet but split-screen should be a blast (just like fastRMX). It is a great game that deserves to be appreciated by a wider audience. I hope Shinen has a PC port lined up.
If they're releasing a pure 1440p patch, why not do 900p or 1080p and then use DLSS for image quality from there?
Because DLSS is not free on a mobile SoC with a pathetically low tensor core count. The render time budget might be too much and that GPU time might be better spent on generating native pixels than having the tensors compute fake pixels. It's just speculation though, I hope there are some GDC presentations that can shed some light on the render cost of DLSS upscaling on the S2.
 
Last edited:
At just 15$, it is worth it. Will you notice the artifacts? Most probably yes, it doesn't matter when you are going full speed. The courses are decent, and I haven't tried it yet but split-screen should be a blast (just like fastRMX). It is a great game that deserves to be appreciated by a wider audience. I hope Shinen has a PC port lined up.

Because DLSS is not free on a mobile SoC with a pathetically low tensor core count. The render time budget might be too much and that GPU time might be better spent on generating native pixels than having the tensors compute fake pixels. It's just speculation though, I hope there are some GDC presentations that can shed some light on the render cost of DLSS upscaling on the S2.

There's never been a PC port of their Nintendo exclusive games
 
When's this patch supposed to drop?



can't wait!

Dynamic 1440p Docked
Dynamic 1080p Handheld

I stopped playing it because it genuinely strained my eyes.
but now I wonder how close to 1440p this will get on average. if the resolution is above 1080p at all times, you gotta wonder why they even bothered using DLSS in the first place... if it constantly drops to 720p or so it's more understandable I guess... but it still can't look worse than the DLSS modes even at 720p 🙃
 
Last edited:


can't wait!

Dynamic 1440p Docked
Dynamic 1080p Handheld

I stopped playing it because it genuinely strained my eyes.
but now I wonder how close to 1440p this will get on average. if the resolution is above 1080p at all times, you gotta wonder why they even bothered using DLSS in the first place... if it constantly drops to 720p or so it's more understandable I guess... but it still can't look worse than the DLSS modes even at 720p 🙃

2026? Is that a typo?
 
I'm excited to play this after the patch. These devs have historically been great at listening to feedback and making updates, so I'm sure we'll get this to a great place eventually.

Honestly, the biggest takeaway from the video for me was that they were able to use DLSS to upscale to 4K. DF previously believed that would be impossible for a 60fps game, since their testing on similar hardware had a frame time cost of ~18ms to upscale to 4K.

That means that the Switch 2 has some kind of bespoke implementation of DLSS. It'll be really interesting to hear more about that eventually.
18ms of frame time cost for DLSS... Exactly the same as the LCD panel latency that Nintendo chose. So maybe the slow-ass 120hz screen was a conscious decision?
 
18ms of frame time cost for DLSS... Exactly the same as the LCD panel latency that Nintendo chose. So maybe the slow-ass 120hz screen was a conscious decision?
That's not how that works… And also it's clearly not 18ms, since it's a 60fps game, which is what I was pointing out.
 
Frame time is unrelated to the amount of time a frame has to be held for upscaling.
I'm not talking about how long a frame "needs to be held", I'm talking about the frame time cost to process the DLSS upscaling, which was initially estimated at ~18ms by Digital Foundry using what they considered "similar hardware".

Obviously the DLSS implementation on the Switch is a different process, because Shin'en were able to utilize that process within a 16ms frame time budget. That was all I was saying before.

I know talking frame time cost can get confusing, especially when there are so many other unrelated things measured in milliseconds (which is where mhirano was getting confused, since his aforementioned 18ms was referring to the response time of the Switch LCD screen… which also isn't entirely correct, but that's a different discussion.)
 
I'm not talking about how long a frame "needs to be held", I'm talking about the frame time cost to process the DLSS upscaling, which was initially estimated at ~18ms by Digital Foundry using what they considered "similar hardware".

Obviously the DLSS implementation on the Switch is a different process, because Shin'en were able to utilize that process within a 16ms frame time budget. That was all I was saying before.

I know talking frame time cost can get confusing, especially when there are so many other unrelated things measured in milliseconds (which is where mhirano was getting confused, since his aforementioned 18ms was referring to the response time of the Switch LCD screen… which also isn't entirely correct, but that's a different discussion.)

I don't think the DLSS cost itself is necessarily different with the Switch 2 version of DLSS.
a big part of the frametime difference is probably down to devs not following the Nvidia recommended way to implement it in PC games.

on PC, games usually follow these recommendations, which means they render post processing at full resolution. on Switch 2 they clearly don't.
Cyberpunk's DLSS implementation was initially hard to make out, because as soon as the camera moves motion blur is applied, and around the edges of the screen chromatic aberration is applied. both of these effects run at half res or lower of the target resolution, while on PC they would run at full res.

Fast Fusion also clearly runs post processing at a very low resolution.
here's a screenshot I took:

ioqo5prj.jpg

you can see that the front of my ship, where post processing is applied, is insanely pixelated, while the back of my ship where post processing isn't covering it is cleaned up by DLSS.
if they followed the Nvidia PC recommendations, my entire ship would look as clean as the back of it does here.

the tree on the far left also is fully pixelated, as the motion blur is overlapping it. DLSS doesn't usually look like that.
 
Last edited:
I don't think the DLSS cost itself is necessarily different with the Switch 2 version of DLSS.
a big part of the frametime difference is probably down to devs not following the Nvidia recommended way to implement it in PC games.

on PC, games usually follow these recommendations, which means they render post processing at full resolution. on Switch 2 they clearly don't.
Cyberpunk's DLSS implementation was initially hard to make out, because as soon as the camera moves motion blur is applied, and around the edges of the screen chromatic aberration is applied. both of these effects run at half res or lower of the target resolution, while on PC they would run at full res.

Fast Fusion also clearly runs post processing at a very low resolution.
here's a screenshot I took:
ioqo5prj.jpg


you can see that the front of my ship, where post processing is applied, is insanely pixelated, while the back of my ship where post processing isn't covering it is cleaned up by DLSS.
if they followed the Nvidia PC recommendations, my entire ship would look as clean as the back of it does here.

the tree on the far left also is fully pixelated, as the motion blur is overlapping it. DLSS doesn't usually look like that.
I wouldn't guess that this screenshot came from 2025 if I didn't know the context:

EROYbtZFmvYqOnS5.png
 
I wouldn't guess that this screenshot came from 2025 if I didn't know the context:

EROYbtZFmvYqOnS5.png

yup.

it's BY FAR, the worst DLSS implementation I have ever seen.
I have played games in ultra performance mode on PC before, just to see how it looks... and no game, even in 1440p ultra performance mode (480p intenral res) no game I have ever done this jokingly looked this bad.

I have played games on the Steam Deck with FSR2/3 in balanced mode that looked better than this in motion.
I have played quite a lot of Gori Cuddly Carnage on the Deck with XeSS quality mode, which should be a lower internal res than Fast Fusion (XeSS uses different scaling factors than DLSS and FSR)... and it looked way cleaner
 
Last edited:
I'm not talking about how long a frame "needs to be held", I'm talking about the frame time cost to process the DLSS upscaling, which was initially estimated at ~18ms by Digital Foundry using what they considered "similar hardware".

Obviously the DLSS implementation on the Switch is a different process, because Shin'en were able to utilize that process within a 16ms frame time budget. That was all I was saying before.

I know talking frame time cost can get confusing, especially when there are so many other unrelated things measured in milliseconds (which is where mhirano was getting confused, since his aforementioned 18ms was referring to the response time of the Switch LCD screen… which also isn't entirely correct, but that's a different discussion.)
Sorry but I was joking. I know all upscaling tech come with processing delay (lag) and the just joked that the intrinsic delay of the BAD Switch 2 panel was to keep up with the slowness of the rendeing pipeline
 
I don't think the DLSS cost itself is necessarily different with the Switch 2 version of DLSS.
a big part of the frametime difference is probably down to devs not following the Nvidia recommended way to implement it in PC games.
There's also a theory going around that the Switch 2 is using a custom implementation of DLSS, specifically a lower-cost (and thus lower-performing) version. Hard to know if it's a bespoke lightweight version or not, though. I'll be really interested in hearing from devs once more kits go out.
 
you can see that the front of my ship, where post processing is applied, is insanely pixelated, while the back of my ship where post processing isn't covering it is cleaned up by DLSS.
if they followed the Nvidia PC recommendations, my entire ship would look as clean as the back of it does here.

the tree on the far left also is fully pixelated, as the motion blur is overlapping it. DLSS doesn't usually look like that.
Please return with pictures from the "Pure Update" that launches tomorrow. I'm very interested in seeing the difference. :)
 
the Update is out! currently downloading...
the download is instantly slow, and I have no idea why...

let's see how it fares

EDIT: EHhhhhh... well, it sure is dynamic! some tracks look fine... others look horrible. the jungle map really pushes the system, and in handheld mode I pixelcounted around 480p in one shot) the industral map ran at around 800p.
it's not really easy to pixelcount the docked screenshots, as it's a 4k output, compressed down to a 1080p jpg. this smears over edges. but the industrial map seems to hover just above 1080p if I had to make an estimate.

sadly, the Switch 2 only allows you to take highly compressed 1080p screenshots, even in docked mode...
aqr7e8f3.jpg

zb5aymcz.jpg

emr4gqpk.jpg

pc9i9eo9.jpg

j6dgah2v.jpg

vypqod4f.jpg

zmrm7pm4.jpg

cur4b2pb.jpg

(in this screenshot here I counted around 480p 😬)


k3tstp7p.jpg

(I count around 765p~800p in this shot)

to note here is that this mode has no antialiasing at all, it uses real time cubemaps for reflections and has no ambient occlusion.
 
Last edited:
the Update is out! currently downloading...
the download is instantly slow, and I have no idea why...

let's see how it fares

EDIT: EHhhhhh... well, it sure is dynamic! some tracks look fine... others look horrible. the jungle map really pushes the system, and in handheld mode I pixelcounted around 480p in one shot) the industral map ran at around 800p.
it's not really easy to pixelcount the docked screenshots, as it's a 4k output, compressed down to a 1080p jpg. this smears over edges. but the industrial map seems to hover just above 1080p if I had to make an estimate.

Thanks for sharing! Now, I don't have the Digital Foundry eyesight for spotting all manner of techniques used, but as a simple layman I think your new jungle shots are preferrable to the first one with pixels galore.
 
I've only tried the new update in handheld mode and tbh, it looks worse then performance mode.

Definitely doesn't improve image fidelity.
 
Thanks for sharing! Now, I don't have the Digital Foundry eyesight for spotting all manner of techniques used, but as a simple layman I think your new jungle shots are preferrable to the first one with pixels galore.

I prefer it... but it's still imo visually worse than Fast RMX in edge cases.

the jungle map looks terrible in handheld mode still for example
 
Last edited:
I've just tried the update and pure mode and ehh... this aint it. It's worse than any of the DLSS modes. It's definitely not 1440p anyway. Ah well.
 
Last edited:
The IQ looks absolutely disgusting.

They should've done more to hit a decent resolution, it's not like the game is doing anything complicated or groundbreaking here.

Easy skip.
 
Thought I was going crazy for a moment there...

The update is absolutely G'damn awful and definitely worse than the DLSS modes, like, fook me it is bad, the aliasing is just up the whazoo, back to balanced mode I go.
 
This update proves that their 'abusing' of dlss was the right way to go.
I wonder what the people who were claiming it would be better without dlss are saying now.
 
No it's a fun game that looks nice. The game is super fast and uses motion blur which is what the image above is showing with a 400 x zoom in.

Just haters hating here again.

Game is light on content though.
The game is indeed very fun, but you don't need to zoom in to see the IQ is a mess. The underlying rendering is very good, but the low resolution and upscaling artifacts are obvious as soon as you load up the game.
 
Last edited:
Top Bottom