shmups.system11.org

Probably not worth the insane hike in price for a device that is already looking to be 300-500$ to double that in order to support content which is already sufficiently supported for most users.

Here's some examples:



I've scaled it using a few best/worst scenarios. I use 1200p for the intermediate step because that's a 5x scale and I know the OSSC can output it (and I'm speculating that the RT5X will too)

1) Original 240p source
2) 9x Bilinear scale from 240p to 2160p
3) 5x NN scale from 240p to 1200p and 1.8x bilinear scale from 1200p to 2160p
4) 5x NN scale from 240p to 1200p and 1.8x NN scale from 1200p to 2160p
5) 9x NN scale from 240p to 2160p

So if we examine these:

1) This represents the worst possible approach to scaling a 240p source to a 4K TV. I assume that this is what you'd get if you directly fed a 240p source to a 4K TV that had really shitty scaling.

2) This represents the best possible approach to "1080p" scaling (pure integer nearest-neighbour) combined with the worst possible approach to scaling a 1080p source to a 4K TV. This is meant to represent what happens if you feed a pixel-perfect OSSC or RT5X image to a 4K TV with a shitty scaler. This still look very sharp, and consider that you'll probably be viewing this on a TV from much farther away than your computer monitor.

3) This represents the best possible approach to "1080p" scaling combined with a 4K TV with the sharpest possible scaling (pure nearest-neighbour). The pixel blocks are uneven size, but becuaes they're so large, you can't really tell that some blocks are 9 pixels wide and some are 8 pixels wide. This is meant to represent what happens if you feed a pixel-perfect OSSC or RT5X image to a 4K TV with the sharpest possible scaler.

4) This represents the best possible approach to 240p -> 4K scaling, taking you right from 240p to 4K via a pure integer nearest neighbour scale. This is meant to represent a hypothetical 4K OSSC or RT9X, as in, a native 4K retro scaler.

Personally, 3/4/5 are all massive improvements over 2, I can barely tell the difference between 2 and 3/4 at TV-like distances, and I can't tell any difference at all between 4 and 5.

I have three 4k TVs and they all scale 1080p wonderfully, and even 720p, for that matter. Each of them measures less than 1 frame of lag when scaling in Game mode as well, verified with the Time Sleuth. I would imagine most TVs will do fine with 1080p to 4k, assuming you're not buying a bargain-bin $300 Walmart 4k TV. If you buy anything mid-range or above, I'd imagine the OSSC Pro and Tink5X will pair wonderfully with them. Do my three TVs scale perfectly? Probably not, if you wanted to get a magnifying glass and examine from pixel to pixel at 6 inches distance. But, at a normal viewing distance, they look and perform wonderfully!

Guspaz, is it widely advertised which scalers different TV companies use, and/or which exact models use what exact scaling method (if some models are different than others in the same companies line-up)?




I'm not too sure about this. I have the highest-end 4K TCL as of ~2 years ago which always competed with models almost double its price. Definitely NOT a $2500+ Samsung OLED or a like $4000 Sony, but 120 backlit zones and other bells and whistles it still hangs in there (even 2 years later) and is definitely not a bargain basement TV.

I'm telling you the above because I'm typing on it right now with a Mac computer I do professional design work on all the time (when I'm not on the Mac Pro with 5K), and my personal Mac is older so it's limited to 1080p output. I see every day how my "highish end" (ok, at least 'mid to mid-high' tier) TV upscales 1080p... and it's perfectly fine, but not the greatest (and I sit maybe 9 feet away). I also see every day how this same TV does on a true 4K/60 Xbox One X and it's kinda a big difference.



I also think 4K upscaling *might* make a marked improvement to my eyes (and again, not owning the equipment I can only rely on my previous quasi-related experience) – again, want to stress especially in those newer but still retro NON-240p/2D consoles like my bane the OG Xbox (and other early 3D as well), because I saw a marked improvement when I used an Extron DSC 301 HD to letter and pillarbox the OSSC's 2x480p=960p signal within a 1080p frame, instead letting my TV do the 960p > 1080p upscaling, so I'm not too hot on the TV's built-in scalers after seeing this, even if this example isn't integer like 1080 to 2160 (this was also right at the time I had a 1080p and a 4K TV so I could experience how both handled it).

Anyway, just wanted to give two more examples of why I think 4K scaling would be better, but like I said I totally get how that's not in the cards, pretty much due to cost alone (and seems like cost vs. benefit is skewing the wrong way as well).

I'm not sure if it has to do with maybe viewing PC content, because one of my 4k TVs doesn't look the greatest when I have my PC hooked up to it and running 1080p. But, all of my retro consoles look wonderful on my sets through the OSSC and Framemeister. I'd have to think viewing retro consoles on a 4k set at 1080p might look better/more acceptable than viewing what should be a razor sharp PC picture at 1080p on a 4k set. I suppose it could be subjective from person to person as well. What one person thinks looks good as far as scaling goes isn't what someone else might think looks good.

For me, I think the OSSC and all of its generic line multiplication modes looks really good. I don't really tweak or mess around with timings or anything else and I love the way my consoles look. Other people will spend hours going through and dialing in all of the other settings and can't stand looking at anything less.

using the OSSC with generic sampling hurts the picture more than having the TV rescale from 1080p or 1200p to 4K. But yes, it still looks really good.

I don't know if it's TV model dependent, I would assume it is, but I've found when I play around with the sampling phase and other settings, I see very little differences to picture quality (to my eyes). Whenever I use the 240p test suite on my Genesis, SNES, etc... All of the scrolling tests look smooth, the checkerboard test looks good and uniform (to my eyes), the color bar and color bleed tests all look good. I don't know if changing sampling is supposed to affect these tests or what. Idk, picture always looks as it should to me.

just proves the point that integer scaling to native res isn't this important.

I don't think a computer outputting 1080p scaled to 4K can really be compared to 240p scaled to 1080p scaled to 4K. The whole reason that it still looks so good for the 240p content is because the 1080p signal you feed the TV has very large blocks of identical pixels, while a true 1080p computer image would have distinct pixels.

What I'm most interested in seeing from 4K-native video processing devices is improved filters that more closely resemble real CRTs (like crt royale).
That stuff is definitely noticeably better on 4K output.

My thoughts exactly. I think it would/could have the horsepower to fix many other upscaling issues too. As in I think one of the larger problems in upscaling things like 480p systems is that it's not just upscaling "very large blocks of identical pixels" like Guspaz just mentioned, but way more complex things. A good example might be something like a gradient which in 480p might be like 4 different bands (lighter to darker or whatever), and when viewed on a display-of-the-day would look pretty seamless and blend into a nice looking gradient... but when upscaled on a 65" or whatever now look like HUGE totally separate bands of individual colors which don't blend together at all, when the original developers wanted them to be a seamless sweeping ~"light blue to dark blue" part of the image... and upscaling just kills that intention. On 240p this upscaling effect can be endearing and make us love the huge pixel graphics even more, but it rubs me the wrong way on some of those non-240p and non-2d content.

I think what you're describing has less to do with upscaling and more to do with moving from composite to RGB.
The newfound clarity is fantastic, but the loss of certain composite effects certainly impacts the look.
This is now more evident than ever after getting a MiSTer for Christmas and playing with the composite blending option in the Genesis core.
I used to be a razor-sharp pixel diehard and dithering patterns never bothered me much, but I'm starting to reconsider how I approach playing Genesis titles now after toggling the effect side-by-side and seeing how so many titles purposely used the effect.

The retro gaming video processor of the future (and this may be a few generations away) should eventually include detecting these patterns and blending them appropriately. For some effects, especially replicating types of displays (CRT/LCD/etc.), we'll need more than 1080p output.

I was hoping something beyond x2 upsample would work.

The Framemeister had something that could be adjusted to gain that dithering effect and was posted on the gaf boards quite a few years ago. If anyone can find it and manage to replicate it on the ossc, that would be a pretty big breakthrough.

the Framemeister has a slider to adjust the horizontal scaling. By increasing it to above 10 you get a composite-like filter effect, but it affects the whole image, not just those patterns in question.

If the OSSC Pro eventually implements composite input and a high quality comb filter, some may be satisfied with feeding composite video directly into the machine.

Also: looks like Sony is really going to push LG this year.

This is true. All you really need for 240p content is that good *First* step of integer scaling of 2 to 3 x before upscaling normally as you need those source pixels preserved large enough that upscaling can't deform them significantly. (For pixel games. 3D rendering is different ball game).

When it comes to true 1080p content upscaling to 4k, integer scaling still matters but it's context dependent on whether a game is 3D rendered or 2D.
And if 3D, how well that 3D is rendered in terms of undersampling issues. (Aliasing).
Games with a lot of undersampling when integer scaled to 4k, will produce a result similar to scaling low res pixel art to higher resolution without integer scaling. It's trying to interpolate values between the undersampling that it physically cannot and in the process will just make it blurrier without solving the problem to begin with. (The undersampling remains. It's just blurrier now) Therefore the best option is to preserve the undersampling as is and it will retain the sharpness and visual make up of what you'd see on a native monitor.

Now 3D content that isn't heavily undersampled and closer to the ground truth (Somewhat like upscaling video content) will fare much better when upscaled and will look good whether you integer scale it or not and it becomes a matter of preference. You can have false positives with this and modern 3D rendered games because of the heavy use of Temoral Anti Aliasing. Consoles are too light to really render games with high quality Anti-aliasing. (Though the PS5/XBSX are powerful enough that you could render every game from the last 2 generations with extremely high quality Anti Aliasing at 1080p and even 4k, 60FPS or 30 depending on user preference. Because you can already do the same with weaker existing PC hardware. But you'll never see it on consoles) TAA is a very lightweight technique costing a fraction of a frame in render time. And due to this TAA depending on the implementation can cause an obscene number of artifacts that don't exist in a raw undersampled image. Or even reverse Anti-Alias the image and cause aliasing to exist that doesn't without the TAA as well. TAA is a whole can of worms that I could spend days talking about.

But let's get back to the point,
i've prepared some 4k videos illustrating both 240p content and various upscaling scenarios to 4k.
And 1080p 3D rendered content heavily undersampled (No AA) and with essentially no undersampling (very high quality AA) integer upscaled to 4k and upscaled with a simple bicubic resample.(As most TV scalers AFIK would not be well represented by a simple Bilinear upscale as it is very low quality).


I would highly encourage downloading these videos and playing them at 4k on a 4k native screen if you can. Google Drive's video player rencodes the video to 1080p.
240p integer scaled directly to 4k https://drive.google.com/file/d/1rQydjD ... sp=sharing
240p integer scaled 3x3 to 960x720 and then upscaled via Bicubic to 2880x2160 https://drive.google.com/file/d/1H9dF4A ... sp=sharing
240p integer scaled 4x4 to 1280x960, bicubic upscaled to 1440x1080 and then that is integer scaled to 2880x2160 (Because why not? As you'll see the final results don't look much different) https://drive.google.com/file/d/1qyB0Do ... sp=sharing
240p upscaled directly to 4k via bicubic. https://drive.google.com/file/d/1SyUBOB ... sp=sharing
These were upscaled from a direct 1x scale 240p video capture from Duckstation (dithering disabled)and then encoded with an AviSynth script with MEGui to do the scaling in an offline render to show what scaling to 4k could look like. (Compression is middle of the road so does technically have some visible chroma loss on reds and some compression artifacts if you know what to look for)


1080p Content (A game with simpler 3D rendering is used, but more modern games will have more rampant undersampling problems. They are visible enough here to illustrate the point)
1080p no AA bicubic upscaled to 4k https://drive.google.com/file/d/1rq_uAP ... sp=sharing
1080p no AA integer scaled to 4k https://drive.google.com/file/d/1mFPgPL ... sp=sharing
1080p with AA integer scaled to 4k https://drive.google.com/file/d/1LVv10x ... sp=sharing
1080p bicubic upscaled to 4k https://drive.google.com/file/d/12N3uPk ... sp=sharing


This would be the main benefit of a direct 4k scaler, but that would involve a frame buffer of some kind i'd imagine. But high res oversampling of CRT filters improves at lower resolutions as well.
Running those same shaders at a higher resolution and downsampling to 1080p has a noticeable improvement over direct 1080p rendering.
https://www.screenshotcomparison.com/comparison/10075
More individual details up close will of course still look better at 4k native. But I don't imagine most people will be sitting 3 feet away from the screen, at normal viewing distance your eyes won't be able to discern details of individual pixels anyway. There is an ideal PPI to viewing distance ratio for every monitor based on resolution,size/PPI and distance where this happens. (And why a 1080p display with a good signal can still look just as fantastic as a 4k screen when viewed at the exact right distance.)

Does the OSSC Pro even have the capacity to apply filters like this? Would the better solution be a separate box that accepts the OSSC signal, applies the filter, and then outputs to the TV? I know a lot of people like the perfect pixel no scanline look but it'd be really great for a lot of consoles if we could easily replicate a consumer Trinitron, consumer shadow mask, PVM, BVM, etc. with all of our original consoles. It's the only part, IMO, where emulation wins. You can really, really dial in the filters and on an OLED you can get a very CRT-esque experience. One they get BFI working well enough on a fast/bright enough panel we'll be very close to replicating the CRT experience in a flat panel.

I really hope so. If the Pro + NTSC SNES or NES can work on Sony 4k tv's, then yes for me! The original OSSC + those systems are a pain in the butt on my X800D. I can get it to work in 5x 256x240 mode for like 30-40 seconds, then screen flickers for a few seconds. The Framemeister works great with both of those consoles. I plan on getting another Sony tv this year (X90J) as I love Sony's processing so I am hoping compatibility is as good as the Framemeister.

I just found out about the Retrotink 5x Pro. I wonder if the jittery sync will be a non issue for it and will make it compatible with Sony sets (and other tv's that have problems with the original OSSC)?

It won't be a Framemeister killer unless it gimmick markets itself and gifts free ones to every influential reviewer with affiliate links that would pay well on a $350-600 device. So it could? Let's not forget that casual retro gamers - the largest group - still live in Composite video bundled cable world. Post videos about using ugly RF to work on 4K HDMI-only televisions. Buy emulated consoles that output HDMI already.

I think OSSC Pro should take Composite + S-Video like Framemeister does to broaden its appeal. You know what the #1 video scaler is on Amazon? $15 AV2HDMI (CVBS PAL+NTSC to HDMI with 720p / 1080p switch). Free shipping + 5% off coupon (!) and generous return policy. Some people complain in reviews that it doesn't even work or does weird scaling but not one complaint about input lag. I don't notice 2 frame lag myself.

I respect the OSSC Pro disclosure of the chips it's using. Not technically sure why an FPGA is needed but that raises the price compared to an AMD or Intel processor. I see the Cyclone 5 5CEFA5F23C8N costs $106.28.



I could ask the same question. BONKERS and Guspaz, impressive stuff, can you tell me if the upscaling + filtering you're using is possible in "no lag" time without a frame buffer? Or is it that the best scaling and filtering to create a good image requires a frame buffer (video memory) like the Framemeister, or non-trivial time to compute? The latter could be parallelized with FPGA.

This discussion makes me feel so old. Latest TV I have is still 2010 40" LCD with analog and HDMI inputs I made a post about. Everyone else living in 4K HDMI-only high life.

Obviously, people that buy cheap Chinese converters aren't in the Framemeister or OSSC Pro target niche demographic, so those users don't matter.

FPGA has multiple advantages for video processing at a reasonable price. Modern OS and drivers are built for security and user friendliness. If we could work on the metal, ARM might have a chance to compete, but how? We would need a custom OS and drivers!

Does the OSSC handle 240p (SNES, N64) better than the Framemeister?

It's lag-free (compared to the 24ms on the FM) and it can look better since you can fine-tune the settings. It's less compatible though with capture devices and depending on your needs (live gaming or capture) the advantages might not be this important to you (for example: you don't need fine tuned sampling and horizontal integer scaling when you're capturing to a 4:2:0 codec eventually anyway).

I'd say for 240p the current OSSC already looks significantly better than the FM for RGB sources even without any fine-tuning.

Agreed. The amount of noise I can end up seeing in certain solid colors on the Framemeister drives me insane. No such issues with the OSSC. I really like the Framemeister, don't get me wrong, but unless you absolutely need highest compatibility for capturing original hardware, I'd take the OSSC all of the way. Even newer capture devices are becoming more friendly with the OSSC's output modes. Like I mentioned previously, if the Framemesiter was currently a $250 device, it would still get my highest recommendation. But, considering you're looking at a $500+ entry point now days, versus the $150ish of an OSSC, the Framemeister is a really hard sell in my opinion. And furthermore, I can't even see the OSSC Pro or RetroTink 5X coming close to the $500 asking price of a Framemeister.

Honestly, the only couple of must have scenarios for the Framemeister, in my opinion, are 480i or having ultimate control over the Gameboy Advance player and being able to size and position the picture a million different ways. But, I'd still say the OSSC does 480i adequate enough and you can still use the GBI profiles for the Gameboy Advance player through component or an GC video adapter to make it look good enough for 95% of people.

I'm sorry but the OSSC is far superior to the Framemeister in this case.

Is it? I've never tried it. I just assumed the Framemeister would be preferred with being able to resize, zoom, etc... I know you're the expert on this stuff, so I'm not doubting you. I just didn't realize the OSSC would work as well, since it doesn't have the vast amount of picture tweaking options that the Framemeister does.

It became a no-brainer when 360p happened. That's a 6x effective scale in 1080p.

What is the best codec to capture gameplay in, and how come the Framemeister can't output in it? What is the best codec that the OSSC can do?

I still don't own an OSSC (I'm waiting for the Pro), and I use the Framemeister with the Firebrandx profiles to play Game Boy player games. It's not too complicated to figure out how to use my setup and change the settings, but I heard that the OSSC is way more complicated to use than the Framemeister in general, and the GBI is complicated enough on the Framemiester, so I'm wondering how much more difficult it is to use the GBI on the OSSC.

The guy in this video says that the OSSC is not as user friendly as the Framemeister, is that true:
https://youtu.be/YGuq4mJhyYY?t=1142

The basic GBI + OSSC setup is unusual for being almost plug-and-play.