Plug a Super Nintendo into a 4K OLED and the image looks like a smeared, bleeding watercolor. You are reaching for the wrong tool. The choice between AI upscaling and a hardware dongle comes down to one hard fact: DLSS, FSR, and PSSR are temporal solutions that read motion vectors, depth buffers, and frame data from a running 3D engine. A SNES or PS2 never generates that data. A hardware dongle such as the Marseille mClassic works on the finished HDMI signal, one frame at a time, using spatial filtering. This is not brand loyalty. It is signal physics.

Understand the wall first. DLSS, FSR, and PSSR are not upscalers in the way you think of a scaler. They are reconstruction engines. DLSS 4 uses a transformer-based model. FSR 4 is machine-learning-based and hardware-accelerated. PSSR runs on the PlayStation 5 Pro's custom AMD GPU. All of them demand a real-time constraint: they must see the frame buffer, the motion vectors from the render pipeline, and the depth buffer to guess where pixels are going. Without that data, they have nothing to reconstruct. No pre-HDMI console outputs any of that. The moment the image leaves the console as composite, S-Video, component, RGB SCART, or VGA, it is a finished raster. The temporal upscaling trick, using previous frames to inform the current one, is dead on arrival because there are no previous frames in the engine's memory. You are asking a scaler to invent data from a signal that was never designed to provide it.

What The Dongle Actually Does

The Marseille mClassic solves a different problem, and be precise about what that problem is. It is a plug-and-play HDMI dongle that applies context-adaptive anti-aliasing and a mild upscale. It is a spatial upscaling solution, meaning it processes a single frame at a time. No temporal upscaling, no motion vectors, no sub-pixel reconstruction. The chip inside is a Marseille ASIC. It takes the 1080p signal coming in over HDMI and applies edge-directed interpolation to smooth out jagged lines. The claim is 1440p output, but the reality is that it outputs a 1440p frame with a 1080p active image area upscaled, not a native 1440p render. That is a full resolution tier of difference. The zero-lag claim is also marketing gloss: independent testing with a Time Sleuth shows approximately 1 to 2 ms of processing latency. Tiny, but not zero. The gap exists because the context-adaptive algorithm needs an analysis window to decide what is an edge and what is noise.

Where It Helps, Where It Fails

Where the device genuinely shines is cleaning up the 480i mess of a PS2. Its context-adaptive anti-aliasing is a spatial filter operating on a single frame. It will not fix a bad composite signal, but it will tame the shimmer on a 3D game running at 480p over component. The dongle does not do the heavy lifting of a RetroTINK 4K, which uses an FPGA for sub-millisecond scaling and offers motion-adaptive deinterlacing that reconstructs full resolution from 480i without flicker. The dongle's deinterlacing is simpler, and it shows. For a PS2 game that switches from a 240p menu to 480i gameplay, the device will re-sync, and that can take a second or two. The RetroTINK 5X Pro handles that frame lock speed far better, re-syncing in well under a second.

Latency In The Real World

Now, the question of latency. Real-time AI upscaling latency is a different animal. DLSS 4 and FSR 3.1 operate inside the game engine, so they add latency on the GPU side, but they do not touch the display path. The Marseille dongle sits in the signal path, so its 1 to 2 ms of processing latency is added on top of the display's own input lag. A TV in Hong Kong with Game Mode on still adds 10 to 15 ms at 1080p and 30 to 40 ms at 4K. The dongle's latency is negligible compared to the panel processing. The OSSC, by contrast, is a line multiplier with no framebuffer, so its lag is measured in microseconds, not milliseconds. The Marseille device cannot match that, but it also does not need to. It is for someone who wants to plug a Wii or a PS3 into a modern display and not think about it. The Wii Dual internal HDMI mod is a better choice for a Wii, tapping pure digital video before the analogue encoder, but that requires opening the console.

Spatial Versus Temporal

Here is where the hardware limits become a feature, not a bug. Spatial upscaling versus temporal upscaling is not a contest. It is a fork in the road. Temporal upscaling needs a frame buffer and motion vectors, which is why DLSS 4 supports 75 games at launch, FSR 3.1 supports 60, and PSSR supports 55 on the PS5 Pro. Those numbers change annually, so check the developers' lists. The Marseille dongle works on anything that outputs HDMI, which means any console from the PS3 and Xbox 360 onward, and anything before that if you run it through a line doubler first. Its proprietary upscaling with sharpening will not fix a bad source, and its aspect ratio is preserved unless the source forces it. Feed it a 240p signal over composite and it will look like a 240p signal over composite. The device is not magic. It is a filter.

The Ceiling Of The Market

The RetroTINK 4K is the ceiling. It uses an FPGA for sub-millisecond scaling, supports motion-adaptive, bob, and weave deinterlacing, and simulates CRT behavior with black frame insertion, scanline generation, phosphor emulation, and HDR CRT beam simulation. It accepts composite, S-Video, component, RGB SCART, and VGA. The OSSC Pro adds a framebuffer and adaptive line multiplication. These are the tools for a Hong Kong apartment gamer who wants one display for a Super Famicom and a PS5. The Marseille dongle is not in this class. It is for the person who wants to plug a Wii into a 4K TV and play Mario Kart without the face melting. Its 1440p output is a full resolution tier below the RetroTINK's clean 1080p or 4K, and its deinterlacing is simpler. If you have a PS2 with a library of 480i games, the dongle will flicker where the RetroTINK 5X Pro will not.

Power, Firmware And Sharpening Traps

One thing to watch: the Marseille device draws its power from the HDMI port. On under-spec ports, that can cause signal dropouts. The fix is a powered HDMI splitter, but that adds a potential point of failure. The firmware can be updated via USB through the Marseille firmware tool, and newer TV firmware can break the HDMI handshake, leaving you with a blank screen until you power cycle the dongle. The Hong Kong Retro Game Association holds meetups where you can physically test this gear. Go and verify a claim before you spend the money. A scaler with overly aggressive sharpening adds ringing halos around sprites that were not present on a CRT, and the dongle's sharpening can do that if you max it out. Back it off a notch or two.

Modern Consoles Need Not Apply

If you are on a PlayStation 5 or Xbox Series X looking for better anti-aliasing, stop reading. You need display-side motion blur and VRR guides, not a dongle. The PS5 Pro uses PSSR, Sony's machine-learning-based, hardware-accelerated upscaler, and it works well on the games that support it. The Xbox Series X uses a mix of FSR 2.x at the system level and DirectML for developers. None of that applies to a SNES. The Nintendo Switch uses bicubic or bilinear scaling internally, and the Switch 2, when it arrives, will use NVIDIA DLSS for 4K docked output. That is a temporal solution with engine access. A Super Famicom cannot do any of this. It outputs 240p over composite or RGB SCART, and the only way to make that look clean on a 4K OLED is a line multiplier with a framebuffer, not an AI upscaler.

What To Buy, In Order

The actual advice is brutal. If you have a CRT, keep using it. Nothing looks better on a CRT than a 240p game. If you are going to a flat panel, the minimum viable solution is a RetroTINK 2X Pro, which does sub-frame latency line doubling for under a hundred US dollars. The next step up is the RetroTINK 5X Pro, which adds motion-adaptive deinterlacing and handles 480i properly. The Marseille dongle is the middle path: it costs less than a RetroTINK 5X, requires no separate power supply, and works on any HDMI source. But it is not a RetroTINK. The GBS-Control is a cheaper alternative for the tinkerer, but its motion-adaptive deinterlacing exhibits chroma artifacts on fast-moving red objects. Test it before you buy. The OSSC is a line multiplier with no framebuffer and no composite input. If you have a NES or a SNES over composite, you need a RetroTINK, not an OSSC.

Connectors That Can Destroy Your Gear

One final trap: RGB SCART versus JP-21. Both use the same physical 21-pin connector, but the pinout is completely different, and plugging one into the other can damage equipment. JP-21 is the Japanese standard with a different voltage on a different pin. If you buy a scaler from Japan, check the input. The same applies to component versus RGB SCART: component is the North American consumer standard, and RGB SCART is the European one, and they require different scaler inputs. A scaler that accepts both is rare and costs more. The Marseille dongle only accepts HDMI, so you need an external transcoder for any analogue source, which adds cost and complexity.

Late Night, Out Of Stock, No Excuses

The failure case: you are in a Hong Kong apartment at 1am, the RetroTINK 4K is out of stock, and you need to play now. Do not buy a cheap scaler from a street market. It will add 20 ms of lag and crush the colours. Do not use the TV's built-in Game Mode as a substitute. A Samsung or LG TV in Hong Kong with Game Mode on still adds 10 to 15 ms of lag at 1080p and 30 to 40 ms at 4K. Instead, use the Marseille dongle if you have it, or find a used RetroTINK 2X Pro online. The 2X is not perfect, but it is lagless, and it will hold you over until the 4K comes back in stock. The Apple TV is not the answer. Its upscaler is designed for video, not games, and it adds input lag. A cheap HDMI converter will also fail you. These are DVD player chips, not game scalers.

Line Doubling Versus Framebuffer Scaling

The line between line doubling and framebuffer scaling is the line between zero lag and a frame of lag. Line doubling repeats each line exactly with no processing, which is why the OSSC has microseconds of lag. It is not buffering anything. Framebuffer scaling captures a full frame and resamples it, which adds at least one frame of lag. The RetroTINK 5X uses a framebuffer, but it is a fast one, so it stays under a millisecond. The Marseille dongle does not buffer a frame. It processes the signal as it passes through, which is why its 1 to 2 ms latency is so low. But it cannot do non-integer scales without introducing shimmer, so it is stuck at its 1440p output. The math of pixel grids does not change: you cannot map 240p to 4K evenly without either uneven pixels or a scaler that accepts the imperfection.

The Narrow Job It Does Well

The Marseille dongle is a good product for a narrow job. That job is cleaning up a 720p or 1080p HDMI signal from a Wii, a PS3, an Xbox 360, or a Switch in docked mode. It is not for a SNES, a PS2, or any console that outputs analogue. Its context-adaptive anti-aliasing smooths edges on 3D games, and the sharpening helps on a soft signal, but it will not restore detail that was never there. The RetroTINK 4K is the ceiling. The OSSC Pro is the purist's choice. The Marseille device is the plug-and-play compromise. The Wii Dual HDMI mod is the best upgrade for a Wii specifically, giving you a pure digital signal, but it requires soldering. For the Hong Kong apartment gamer with one display, the answer is a RetroTINK 5X Pro for retro consoles and a modern console with its own upscaler for the rest.

AI Upscaling Technologies and Hardware Dongles Solve Different Retro Gaming Problems