Post-Process Anti-Aliasing vs Native Anti-Aliasing in Retro Upscaling

Your Super Famicom's 240p signal hits the 4K OLED and the TV's scaler goes to work. What comes back is a face made of jelly. Every sprite edge shimmers, every line of text blurs. The smooth 2D pixel art you remember from Hong Kong magazine screenshots has become a smeared memory. The mClassic dongle, plugged into the HDMI port and powered by it, promises to fix this with post-process anti-aliasing. Before you spend the money, know exactly what the box is doing and, more importantly, what it cannot do.

Native Anti-Aliasing And The Frame-Time Cost

Native anti-aliasing in a 3D renderer happens during the geometry rasterisation stage. The GPU takes multiple samples per pixel, averages them, and stores the result in a G-buffer. Techniques like 4x MSAA cost 20-40% of your frame time because the hardware is processing sub-pixel data that never appears on screen. It is a technique for modern GPUs and for emulators running on PC hardware. No original console from the NES to the GameCube era has it. Those machines render at native 240p or 480i with no AA pass at all. The image is pure, unprocessed, and razor sharp.

Post-Process Anti-Aliasing And The Softening Trade-Off

Post-process anti-aliasing is a different animal. FXAA and SMAA analyse the completed frame, detect edges, and blur them. FXAA uses a single LDR colour framebuffer and a blur radius of about 0.5 pixels. SMAA does better, using morphological edge detection on luma, colour, and depth to search up to 32 pixels for an edge. Both run after the renderer has finished, which is why they work on any game. The trade-off is that they soften the image. Thin geometry like power lines and fences gets reduced to a broken, dashed appearance. This is the Digital Foundry 2012 finding, and it never gets old.

Marseille Inc. places the mClassic in this post-process family. It is a proprietary spatial anti-aliasing algorithm applied to the HDMI input signal. It accepts 480p, 720p, and 1080p input, and at 60 Hz it outputs up to 1440p. The 1440p output is real in the sense that the dongle sends a 1440p frame down the wire, but the active image area is 1080p upscaled. It is not a native 1440p render. The mClassic is a post-process anti-aliasing mClassic retro setup that lives in the HDMI cable between the console and the TV, after the scaler, and it has no access to the geometry being drawn.

• Claim: 1440p output, zero added lag
• Reality: 1440p frame, 1080p active image; 1-2ms processing latency measured with a Time Sleuth
• Power: HDMI port power; under-spec ports cause signal dropouts
• Input range: 480p, 720p, 1080p; 4K input passes through unprocessed

The Marseille marketing implies zero added lag. The published Time Sleuth results from the Shmups Forum show a different number. The mClassic adds approximately 1-2ms of processing latency. The gap is small but real, caused by the context-adaptive algorithm's analysis window. The dongle has to see the whole frame before it can decide where to blur. That is a physical fact. A competitive player feeling for the difference between 16.67ms and 18ms is not imagining it. For rhythm games and shmups, that is the difference between a perfect dodge and a hit.

For a modern comparison, look at the Nintendo Switch. SMAA in Switch titles is a per-title implementation because the hardware can handle it. FXAA and SMAA both work on a single LDR colour framebuffer, but SMAA preserves sub-pixel detail down to one pixel where FXAA loses it. The mClassic's context-adaptive algorithm is closer to SMAA than FXAA in intent. It detects edges, profiles them, and blurs only those pixels. It is not a full-screen blur. For 3D games from the PS2 and GameCube era, this is a net win. The jaggies on a character model's shoulder disappear. But the algorithm also softens 2D pixel art. A line of text that was one pixel wide on a CRT becomes a grey smudge. Purists who want razor-sharp pixels for Magical Pop'n or Rondo of Blood lose out.

Marseille mClassic processing method is a form of screen-space edge detection. It looks at the luminance and chroma of each pixel, finds an edge, and then applies a blur that is wider perpendicular to the edge than along it. This is what makes it context-adaptive. The algorithm is not a fixed 3x3 kernel. It is a filter that reads the local geometry of the frame and decides where the smoothing will do the most good with the least damage. The cost is the 1-2ms of lag measured on the Shmups Forum. The benefit is that it works on any input. Put it after an OSSC or a RetroTINK-5X Pro and it will smooth the line-doubled output. Placing it after the RetroTINK-4K is redundant because that device already does a full-frame motion-adaptive deinterlace and filter pass.

This context-adaptive anti-aliasing dongle is powered by the HDMI port. That is a weakness. An under-spec port on a cheap HDMI splitter or an ageing TV will not deliver enough current, and the dongle will drop the signal. Plug it directly into the TV and use a separate splitter for the console. The mClassic has a physical switch for modes. One mode applies the smoothing, the other is pass-through. A/B test the filter on a single game. There is no menu, no USB firmware update. If the algorithm has a bug, you cannot patch it. The dongle is what it is.

Here is the honest trade-off. For 3D games from the N64, PS1, PS2, GameCube, and Dreamcast, the mClassic is a genuine upgrade. The polygons are wobbly, the textures are blurry, and the filter makes them look like they are being rendered at a slightly higher resolution. For 2D pixel art, it is a downgrade. The pixel grid is a deliberate aesthetic choice. A one-pixel-wide line of dialogue is supposed to be exactly that. When the mClassic blurs it, the game loses its sharpness and ends up looking like an emulator with the smoothing filter left on by accident. If you play mostly RPGs and platformers from the 16-bit era, do not buy this. Save the money for an OSSC and a set of HD Retrovision component cables.

If you are in a Hong Kong apartment with one display and you own both a PS5 and a Super Famicom, the mClassic is a reasonable middle ground. It costs about as much as three new games and adds a small amount of lag that most players will never notice. But the failure case is real. The mClassic will not fix a 480i signal properly. The motion-adaptive deinterlacing it claims is not there. For PS2 games that are 480i-heavy, get a RetroTINK-5X Pro or a GBS-Control. The GBS-Control's motion-adaptive deinterlacing works but exhibits chroma artifacts on fast-moving red objects. The OSSC has no framebuffer and no post-process AA, so the mClassic after it is a fine pairing. The OSSC Pro adds a framebuffer and adaptive line multiplication, which is a different price class entirely.

Here is the physics. A frame at 60Hz lasts 16.67ms. Any scaler that buffers a full frame adds at least that much lag. The OSSC and GBS-Control with line doubling have zero lag because they repeat each line exactly with no processing. The mClassic's 1-2ms is a rounding error compared to the 10-15ms a Samsung or LG TV in Game Mode adds at 1080p, and 30-40ms at 4K. The TV's scaler is still active even in Game Mode. You cannot turn it off. So the chain is console to mClassic to TV, and the TV is the biggest source of lag in the room. If you are a competitive player who perceives milliseconds, the TV's Game Mode is not enough. Get a monitor that accepts 1080p natively with no scaling, or a 4K panel with a dedicated 1080p integer-scale mode. The mClassic does not help here.

One more detail. The mClassic outputs 4:4:4 chroma subsampling at 1440p, but only if the source is 1080p. Feed it a 480p signal from a DVD player or a PlayStation 2, and it has to upscale internally, using 4:2:2. Colour detail is halved. On a CRT this does not matter because the phosphor grid masks it. On a 4K OLED it is visible as colour fringing around red and white text. The HDMI standard is a defined, unchanging pipeline. What goes in is what gets processed, and the chroma path is lossy. Use component cables from HD Retrovision and let the mClassic handle only the AA, not the colour space conversion.

The mClassic suits the Hong Kong apartment gamer with one display, a PS5, and a retro console they want to plug in without a soldering iron. It suits the buyer of a pre-modded console who wants a gift that works out of the box. It does not suit the competitive shmup player who feels the 1-2ms and needs the OSSC's microsecond response. It does not suit the pixel art purist who wants to see every dot of a Super Famicom sprite. For that person, the OSSC with HD Retrovision cables remains the reference. The mClassic is a filter, and filters are always a compromise.

The mClassic passes audio through but does not embed analogue audio into the HDMI stream. Using a PS2 with component cables and the HDMI dongle means connecting the audio separately. The dongle has no breakout. The result: video goes through the dongle, audio goes straight to the TV, and the two are a few milliseconds apart because of the mClassic's processing. For most games this is fine. For a rhythm game where the sound drives the input, it is a dealbreaker. The competitive player will sync their audio to the video delay and then find the 1-2ms of AA lag on top. That is the moment they unplug the dongle and go back to raw 1080p.