• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

ZSNES is coming back

ZNES brings back fond memories, but SNES emulation in 2026 is in a pretty good spot. SNES9X for a good all around emulator, an BSNES for extreme accuracy. I'm all for emulator choices, especially if it offers graphic enhancements through GPU. N64 and Playstation emulator could use some work. There still isn't a definitive good N64 emulator, but more games are getting decompiled and may make this less important. Playstation emulators today don't have the plugin system of EPSXE and Pete's plugins. I had Playstation games looking amazing on my PC... over 15 years ago. Duckstation was promising, until the author decided to hate on Linux.
 
Neat, while I don't remember the SNES emulators I used ever really having problems being CPU base, there may have been some "fancy" things that it simply couldn't render properly. But as long as the emulator has the ability to have save states I'm a happy camper.
 
sfsuphysics said:
Neat, while I don't remember the SNES emulators I used ever really having problems being CPU base, there may have been some "fancy" things that it simply couldn't render properly. But as long as the emulator has the ability to have save states I'm a happy camper.
Click to expand...
The all on the CPU ones actually are exceedingly accurate and are more or less "perfect" at this point. This is about trying to move it beyond what the OG hardware could do, not thing that the CPU based ones couldn't render properly. You see emulators of 3D systems that can do things like that, where they can increase resolution, replace textures with high rez texture packs and so on. This is to try and bring something like that to the SNES.
 
Zsnes is also textbook example of an UI that not only looks good but is also easy to understand and easy to use. It doesn't have tons of unnecessary settings just for the sake of it. Very user friendly. Even if the accuracy of the emulation was anything but accurate, it just worked without user tinkering beyond setting up key bindings.

While I do use Retroarch (a prime example of complete ASS user interface design) these days, I am down on using Zsnes again. The mods it it may come with could be very interesting.
 
I'm just wondering if it's possible for these emulators to use DLSS/FSR for better graphics?
 
AVATARAT said:
I'm just wondering if it's possible for these emulators to use DLSS/FSR for better graphics?
Click to expand...
Maybe one like Microsoft auto SR, Nvidia RTX super res or older FSR 1.x already work, any of those that work from the image output alone could work, but they tend to be trained on 3d, would be interesting if they really beat the upscaler developed over the years specially for pixel arts type that emulator use, they could be hard to beat at this point.

Lack of motion vector (and others input they require to work well like depth map, UI Mask and what not) would be an issue for the FSR/DLSS 2.0 and up... a more interesting one would be what a DLSS 5.0 type could do for old legacy title, 100% of the gpu budget could go into it, cpu good enough to do all the traditional render stuff.
 
Last edited:
AVATARAT said:
I'm just wondering if it's possible for these emulators to use DLSS/FSR for better graphics?
Click to expand...
Maybe but DLSS and FSR don't technically produce better graphics. They do but only when TAA is used instead, which I don't think emulators use TAA. Very few emulators make use of FSR, and nothing makes use of DLSS. You need accurate motion vectors to make use of any of these technologies, which no emulator does. Yuzu and it's forks do make use of FSR1 for upscaling, and Citron makes use FSR2, but Citron maybe discontinued.
 
We have better scalers for 2D sprite graphics already, some are already included in different emulators and have been for decades. Things like 2xSai, SuperEagle HQx and so on. All of them are very effective at removing jagged edges from sprites. However is the end result better than raw pixels? Kinda...ish... but it also mangles up all the details or make the image look like watercolor painting. I'd rather use a proper CRT filter, instead of mangling the picture up it even improves the details due to CRT's unique features.
 
Last edited:
DukenukemX said:
Maybe but DLSS and FSR don't technically produce better graphics. They do but only when TAA is used instead, which I don't think emulators use TAA. Very few emulators make use of FSR, and nothing makes use of DLSS. You need accurate motion vectors to make use of any of these technologies, which no emulator does. Yuzu and it's forks do make use of FSR1 for upscaling, and Citron makes use FSR2, but Citron maybe discontinued.
Click to expand...
Have to remember that DLSS works on 3D data, not 2D. I mean yes, it is processing the final 2D image but it is looking at data about the 3D scene (like depth information) and it is trained on data from 3D renders. It has been trained to identify things like edges in polygons, not issues with 2D pixel art. It also is, as noted, a temporal upscaler meaning it needs data on previous frames and motion vectors.

FSR1 could be used, that is a simple spatial upscaler though I don't know that the results would be any good.
 
Sycraft said:
Have to remember that DLSS works on 3D data, not 2D. I mean yes, it is processing the final 2D image but it is looking at data about the 3D scene (like depth information) and it is trained on data from 3D renders. It has been trained to identify things like edges in polygons, not issues with 2D pixel art. It also is, as noted, a temporal upscaler meaning it needs data on previous frames and motion vectors.

FSR1 could be used, that is a simple spatial upscaler though I don't know that the results would be any good.
Click to expand...
I'm just going by what emulator authors have said when it comes to implementing DLSS and FSR. Yuzu and it's forks as well as RPCS3 have implemented FSR because it doesn't require motion vectors. Including DLSS is a frequent request for emulators such as RPCS3. The administrator of the RPCS3 forum didn't have great things to say about DLSS. Ani is also a developer of RPCS3.

> Furthermore, AMD doesn't disclose commercially important parts of its code. This algorithm is technically and morally obsolete.
FSR 1.0 is fully open source, it's a compute shader and the source is included in RPCS3, the algorithm is based on Lanczos, which is widely known and nothing new nor obsolete. Lanczos/FSR algos are still the best forms of spatial upscaling, they are also the best scalers for video upscaling/downscaling. Lay off the hallucinating AIs and start doing your own research so you don't end up with blatantly wrong information.

> It's time to consider an alternative.
The spatial upscalers you can implement are bilinear, trilinear, bicubic, FSR1 and NIS. Considering the first three are rudimental scalers that change image output a lot, and the latter two are based on the same Lanczos alternative, there is zero reason to use NIS over FSR1, but you can enable it on the NVIDIA control panel, it works without explicit support from the app. Unless you come up with a better spatial scaling algo, there are no alternatives.

> And if it so shitty why does it handle Antialising better?
RPCS3 has no AA override yet besides SSAA, which is objectively the best form of AA but also the most GPU demanding, it only emulates MSAA and it's up to the game to configure the parameters or not use it at all as on a real PS3. Other forms of AA are planned as enhancements but not on a short term as there are more urgent issues to fix. Use SSAA by rendering at a higher resolution than your screen so the rendering is supersampled and then downscaled to monitor size.
 
DukenukemX said:
> Furthermore, AMD doesn't disclose commercially important parts of its code. This algorithm is technically and morally obsolete.
FSR 1.0 is fully open source, it's a compute shader and the source is included in RPCS3, the algorithm is based on Lanczos, which is widely known and nothing new nor obsolete. Lanczos/FSR algos are still the best forms of spatial upscaling, they are also the best scalers for video upscaling/downscaling. Lay off the hallucinating AIs and start doing your own research so you don't end up with blatantly wrong information.
Click to expand...
He's correct that FSR is spatial only, which is why its results aren't all that good. The good ones are temporal. However It would take a much more intense engine rewrite to make it able to expose the motion vectors that the temporal ones need, if it even was possible. For something like an SNES it is just a non-starter. Like I said FSR1 could be used, but I dunno how good it would be.

That said, he's also clearly one of the anti-AI angy types so I'd take with a grain of salt his opinions on DLSS as it is objectively the best temporal upscaler. But a temporal 3D upscaler wouldn't be useful in ZSNES.

That said, someone probably could make a neural network like upscaler for low rez pixel art for something like an SNES. It would be pretty intense though because realistically what you'd have to do is have artists redo the artwork in higher resolution for all the games and then train it on that. With enough training you could probably get pretty good results... but then if you are doing that you could just use ZSNES to swap in teh high rez art they've drawn and skip the whole complex, time consuming, and expensive "train a neural network" step.

The reason that it works well for modern 3D games is that we can easily train it on the games themselves. Basically just run the game with extremely high SSAA/resolution and that's your training set. You use existing game assets, just with good AA and resolution, which is something you can do no problem it is just too slow for realtime. With low rez pixel art, there's not a high rez version to train on, so you'd have to make it.
 
Sycraft said:
He's correct that FSR is spatial only, which is why its results aren't all that good. The good ones are temporal. However It would take a much more intense engine rewrite to make it able to expose the motion vectors that the temporal ones need, if it even was possible. For something like an SNES it is just a non-starter. Like I said FSR1 could be used, but I dunno how good it would be.
Click to expand...
There might be a chance if the games were decompiled and then native made for PC, but I'm not sure if there's a lot of interest in this?
That said, he's also clearly one of the anti-AI angy types so I'd take with a grain of salt his opinions on DLSS as it is objectively the best temporal upscaler. But a temporal 3D upscaler wouldn't be useful in ZSNES.
Click to expand...
I see him as a down to Earth developer who knows more about the inner workings of games than any of us could ever imagine. This is why I like listening to emulator developers because not only are they working on cutting edge stuff but they're very transparent. If emulator authors end up working for game studios then we may have games that well optimized.
That said, someone probably could make a neural network like upscaler for low rez pixel art for something like an SNES. It would be pretty intense though because realistically what you'd have to do is have artists redo the artwork in higher resolution for all the games and then train it on that. With enough training you could probably get pretty good results... but then if you are doing that you could just use ZSNES to swap in teh high rez art they've drawn and skip the whole complex, time consuming, and expensive "train a neural network" step.
Click to expand...
Ani did mention NVIDIA Image Scaling (NIS), but I think we're going too far for 2D games. If AI can be used for anything, then it's to create high rez textures, but there's a chance the community has already made high rez textures for SNES games and didn't use AI. Also, there are remakes for some SNES games. Some Remasters are better than others from that era.

View: https://youtu.be/ZXBt5-REXmA?si=dvWbxcJHrm__Mq1Z

View: https://youtu.be/zBeOSfX3rb4?si=lLtcZrgY6GA_QH1p
 
DukenukemX said:
There might be a chance if the games were decompiled and then native made for PC, but I'm not sure if there's a lot of interest in this?
Click to expand...
There is some of that going on with certain games. It's a lot of work so I imagine it is the kind of thing that'll only be done for really popular or significant titles. A lot of it seems to be focused around Xbox titles that didn't come to the PC, but there's been N64 work as well.

DukenukemX said:
I see him as a down to Earth developer who knows more about the inner workings of games than any of us could ever imagine. This is why I like listening to emulator developers because not only are they working on cutting edge stuff but they're very transparent. If emulator authors end up working for game studios then we may have games that well optimized.
Click to expand...
He can be smart and know a ton, and still have blind spots. In fact, working with researchers, I'd argue that the smarter you are and the more you know about a given topic, the more blind spots there are and potential biases. Dude clearly hates AI and AI upscaling. Well, sorry to break it to him but FSR 1 is not the best upscaler. The temporal AI upscalers are objectively way, way, better. That they aren't suited for use in emulation doesn't change that fact.

DukenukemX said:
Ani did mention NVIDIA Image Scaling (NIS), but I think we're going too far for 2D games. If AI can be used for anything, then it's to create high rez textures, but there's a chance the community has already made high rez textures for SNES games and didn't use AI. Also, there are remakes for some SNES games. Some Remasters are better than others from that era.
Click to expand...
I think another issue with SNES graphics and any kind of AI upscaling is just that they are so low rez. There's just not enough input data to really do a good job. I mean we see that with things like DLSS. Performance mode works well at 4k because it is a 1080p input. However, at 1080 performance doesn't work nearly so well because of the 540p input.


I dunno, maybe someone will introduce an AI upscaler for the new ZSNES that does a great job, but I feel like it is a pretty hard problem to solve.
 
Last edited:
You must log in or register to reply here.
Back
Top