I used the torapu/buttersoft mod for years and I was happy with the results during that time. A while back I swapped it out for mikejmoffitt's v1.2 3-CHIP sharpening board which is based on yoshiyukiblade's circuit discussed here:
viewtopic.php?f=6&t=70196
After comparing both I can say that mikejmoffitt's board is the better of the two and IMO the best option currently available. It doesn't have the non-linear issue that the torapu mod has and it can have an output that is basically on par with a 1-CHIP. There is a caveat though. In order to get 1-CHIP levels of the sharpness, the installed board needs to have its sharpness compensation values adjusted. Ideally this is done on a per console basis since the amount of blur varies from console to console, even with the same motherboard revision. This is actually one of the benefits of yoshiyukiblade's circuit. Since the sharpness is adjustable, each console can have its sharpness optimized, regardless of how blurry the stock output is. This has been my experience after having now installed 9 of these boards.
To better illustrate this I'll post some oscilloscope plots since the blur can be measured using an oscilloscope. I'll also include an accompanying lossless capture for each. These captures were taken using a RetroTINK-4K into a Datapath-E1S capture card.
Here's how the blur of a stock SHVC-CPU-01 looks:
And here's a SNES Jr. which has been RGB bypassed using borti4938's board:
Now take a look at the rise and fall times of each. A stock 3-CHIP has roughly a rise time of 100ns and a fall time of 250ns. The SNES Jr. (which is basically a 1-CHIP) has rise and fall times that are roughly 20ns. The much faster rise and fall times of a 1-CHIP yields a much sharper output. In short, we can say that the faster the rise and fall times are, the sharper the output.
Now here's the same SHVC-CPU-01 console with mikejmoffitt's board installed:
The rise and fall times have been sped up, roughly cut in half. This is noticeable in the capture as well where we can see the blur is significantly reduced. Depending on your setup, with say a CRT or a scaler with generic settings applied, this last bit of blur won't be so noticeable.
Now here's the same console after the sharpness compensation values have been optimized:
The rise and fall times have been sped up even further and they're now on par with the SNES Jr./1-CHIP. That remaining blur on Mega Man is now all but gone. If you look closely you can see there is a tiny amount of blur left but that's splitting hairs. I don't think anyone sitting a normal gaming distance away from their display would be able to notice it.
Here's the same SHVC-CPU-01 console with the torapu/buttersoft board installed for comparison:
The output with this mod has really fast fall times but slow rise times. As I understand, this combination of fast fall times with slow rise times means it's non-linear. This creates more of a filter effect over the blur rather than reversing the blur like the yoshiyukiblade circuit does. The output looks similar to using the OSSC's Reverse LPF feature.
I still think the torapu/buttersoft mod is a decent option despite it being non-linear. It's inexpensive if you DIY, easy to install, and easily reversible if you decide to go with another option in the future. But if I had to pick between the two, I would definitely go with mikejmoffitt's board.
