NES and Famicom Clones, PAL vs NTSC, and the Dendy

This page collects everything region- and clone-related for the 8-bit Nintendo family in one place, because it all turns on the same handful of facts: which CPU, which PPU, and which crystal. My NES front-loader and Famicom family pages cover repair of the Nintendo machines themselves; they point here for anything about PAL vs NTSC timing, region conversions, or clone hardware.

I have not had a clone console on my bench. The clone material below leans heavily on Adrian Black’s bench work (Adrian’s Digital Basement), who has reviewed, opened and modified a whole run of them, and on the NESdev wiki for every clock and timing figure. Where a claim is Adrian’s opinion or a single observation on one unit, I say so. His videos are linked throughout and are worth watching, because the scope traces and side-by-side comparisons do not survive being written down.

The short version

• An NES is two custom 40-pin chips (the CPU, which also makes the sound, and the PPU, which makes the picture) plus a crystal and a pile of ordinary logic. Every clone is a different answer to “what do we do about those two chips.”
• The region lives in that trio. NTSC, PAL and Dendy consoles differ in crystal frequency and in the divider baked into each chip. There is no 50/60 Hz switch on a real NES, and the lockout chip has nothing to do with speed or colour.
• American games on a PAL NES run about 17 percent slow, play their music about a semitone and a quarter flat, and can break wherever the game counts CPU cycles. The Dendy clone kept the NTSC CPU-to-PPU ratio, so the same games run slow at 50 Hz but otherwise correctly, and often with less slowdown than on an NTSC machine.
• Clone chips are real hardware, not emulation, but not exact. The UMC CPUs have two pulse-channel duty cycles swapped, so sound effects come out with a different timbre, and clone PPUs differ in palette and in some register behaviour.
• A replica board built around genuine Ricoh chips is 100 percent compatible by construction. That is the kind Adrian recommends.
• Almost every cheap clone Adrian opened had the same analog faults: composite video too hot because nobody terminated it, noisy switching power, and power switches wired after the regulator. They are all fixable.

The three chips that decide everything

The NES CPU (Ricoh RP2A03 on NTSC, RP2A07 on PAL) and PPU (RP2C02 NTSC, RP2C07 PAL) both run off one master crystal. The PPU divides it down for the pixel clock and derives the colour carrier from it; the CPU divides it by a fixed number set inside the chip. That fixed divider is the whole story of the region split. Figures below are from the NESdev cycle reference chart:

Two consequences matter for repair:

• The crystal and the PPU are a matched pair. The PPU builds the video timing and colour burst from the crystal, so an NTSC PPU on a PAL crystal produces a signal nothing can lock to. Adrian tried exactly that on a clone board and got no picture at all (video, 28:00).
• The CPU is more forgiving than you would expect. It will run on whatever clock it is fed, just at a different speed, and the sound pitch moves with it because the audio unit lives inside the CPU. Adrian ran an NTSC CPU from a separate signal generator, unsynchronised with the PPU, with no trouble at all (video, 2:00).

PAL vs NTSC: what actually goes wrong

Nintendo designed the NES for 60 Hz Japan and North America and then adapted it for 50 Hz PAL television. Adapting it meant a different crystal (the PAL colour carrier is higher), a different PPU that draws 312 lines instead of 262, and a CPU that divides by 16. Nintendo could have used 15 and kept the NTSC relationship between CPU and PPU, but the divider circuit in the chip only produces even periods, so it went to 16 (NESdev).

So a game written for NTSC, played on a PAL NES, meets three separate problems:

• Everything is slower. Most games advance one step per video frame, so at 50 frames a second instead of 60 the whole game, music tempo included, runs at five sixths speed.
• Music is flat. Games set pitches as counts of CPU cycles, tuned for the NTSC clock. On the slower PAL CPU every note drops by about 7 percent, a bit over a semitone. PAL releases that were properly converted retuned their music; plenty of European releases were not converted well, and some games never got a PAL release at all.
• Anything timed in CPU cycles lands in the wrong place. A scanline is 106 9/16 CPU cycles on PAL instead of 113 2/3, so a status-bar split or a raster effect that the game times by counting cycles lands on the wrong line.

Slowdown, oddly, is not the PAL problem. The PAL CPU is slower per cycle, but a 50 Hz frame is long enough that it still gets about 33,250 CPU cycles per frame against about 29,780 on NTSC, so a game that lags on NTSC usually has more room on PAL.

The reverse, a PAL game on an NTSC console, fails the other way: faster, sharp, and a game that relied on PAL’s long vertical blank to update video memory can run out of time and corrupt its graphics. The colours are generated by the PPU in the console, not by the cartridge, so an out-of-region game does not get “wrong colours” as such; the one real colour difference is that the PAL PPU swaps the red and green colour-emphasis bits, which only matters to games that use emphasis tinting.

The lockout chip is a separate matter. A 72-pin PAL NES refuses NTSC cartridges through the lockout chip region code (and PAL-A refuses PAL-B), which the CIC section of my front-loader page covers. Disabling the lockout lets the cartridge boot; it changes nothing about speed, pitch or timing.

The Dendy, and why it was the smarter design

Taiwanese makers built Famicom clones for 50 Hz markets, and the best known of them was TXC’s Micro Genius, sold in Russia by Steepler under the name Dendy. “Dendy” has since become the generic name for this whole family of PAL famiclones (NESdev). The chipset is UMC’s UA6527P CPU and UA6538 PPU.

Its designers were not trying to copy the PAL NES. They were trying to run the huge library of Famicom games, written for NTSC timing, on a PAL television. They did it with two changes:

• The CPU divides the PAL crystal by 15 instead of 16. That gives 1.77 MHz, within about 1 percent of the NTSC CPU, and keeps exactly 3 PPU dots per CPU cycle and 113 2/3 CPU cycles per scanline, just like NTSC. Code that counts cycles to time a raster effect still lands on the right line, and the music stays within about a sixth of a semitone of NTSC pitch.
• The PPU adds 51 idle lines after the picture and before the vertical-blank interrupt. A game sees the same short 20-line vertical blank it expects from NTSC, but it gets a lot of extra time to compute each frame before that interrupt arrives: about 35,460 CPU cycles per frame, the most of the three.

The result is that NTSC games on a Dendy run at 50 Hz, so they are slower overall, but they run correctly, and because of those extra lines they tend to have less slowdown than on a real NTSC machine. Adrian could not make the Super Mario Bros. 2 Shy Guy test spot slow down at all on a stock Dendy, where it occasionally slows on NTSC (video, 15:00), and he walks through the vertical-blank budget at 9:00 in the same video.

The Dendy’s cost is the chip quirks covered below, most audibly the swapped duty cycles.

Swapping CPUs, PPUs and crystals between regions

Converting a PAL NES to NTSC

A real conversion swaps all three: CPU 2A07 to 2A03, PPU 2C07 to 2C02, and the crystal 26.6017 MHz to 21.4773 MHz. Swap fewer and you get a mismatch, not a partial conversion. Things to know before starting:

• The cheapest donor set of NTSC CPU, PPU and crystal is a dead NTSC front-loader or a Famicom, which carry the same NTSC silicon. Adrian points out that Japanese auction sites are full of rough Famicoms that are good for exactly this (video, 13:00).
• Socket all three. Adrian socketed the crystal too, which makes it trivial to go back (video, 13:00).
• One community source reports that the PAL board’s audio-mixing component values are chosen for the PAL CPU and are not ideal for a 2A03. I have not confirmed it; expect to check the audio level afterwards.
• An NESRGB board (version 4 and earlier) needs its region jumpers moved to NTSC, or you get glitches and colour inversion and blame the wrong part.
• The lockout chip is untouched by the conversion. A converted PAL console still refuses American cartridges until you deal with the lockout.
• Many PAL consoles have diodes in the controller ports that make some NTSC controllers read as dead. That is a separate, easy fix covered on the Famicom family page, and worth knowing before you condemn a port.

Moving only the CPU: over- and underclocking

Because the CPU tolerates any clock and the PPU does not, moving just the CPU between regions is a controlled speed change with the video untouched. Adrian did both directions on a pair of socketed boards (part 1, part 2):

• Dendy CPU (divide by 15) in an NTSC board: about 1.43 MHz, a 20 percent underclock at 60 Hz. It runs, but games that count cycles break badly. Super Mario Bros. 3’s status-bar split wandered, Kirby’s Adventure corrupted, Contra Force was unplayable.
• NTSC CPU (divide by 12) in a Dendy board: about 2.2 MHz, a 25 percent overclock at 50 Hz. Almost everything he tried ran, with slowdown gone, but the music is out of tune because pitch follows the CPU clock.
• External clock on the CPU’s CLK pin (pin 29) with the PPU on its own crystal: an NTSC CPU ran happily up to the 30 MHz limit of his generator (2.5 MHz CPU) at a true 60 Hz. Contra Force became close to lag-free. Music tempo stays correct, because games pace music off the 60 Hz frame, but pitch shifts. Cycle-timed effects drift further off as the clock rises, and one demo ROM detects the speed and refuses to run.

His conclusion, which I agree with: games tolerate overclocking far better than underclocking, and anything that counts CPU cycles will always break at a non-stock speed, so a practical “turbo” needs to be switchable per game.

Using clone chips as NES repair parts

The Ricoh CPU and PPU have not been made in decades, so a dead one means a donor pull or a clone. The UMC clones are real, working chips, but they are not exact, and which one you need depends on the console:

Sources: NESdev CPU variants and PPU variants.

Three traps:

• The “P” parts are Dendy parts, not PAL NES parts. Several repair writeups list the UA6527P and UA6538 as the PAL versions. In a genuine PAL NES they turn it into half a Dendy: the CPU runs at the wrong speed for the PPU, and PAL games are tuned for a CPU that divides by 16.
• Relabelled chips are common, especially “UA6527P”. NESdev documents chips sold under that marking with both a divide-by-15 and a divide-by-16 divider, some with correct duty cycles and some with swapped ones. Paint and sanding hide the real marking; acetone takes paint off, and the markings on the bottom of the package sometimes survive. Do not trust the label. Measure: divide the crystal frequency by the frequency on the CPU’s M2 pin (pin 31) and you have the divider.
• All known clone CPUs use the NTSC table for the sample channel, including the ones made for PAL systems, so sampled sounds in PAL games play at the wrong pitch on them.

The duty-cycle swap deserves a word, because it is the thing people hear. The two pulse channels can play four waveform shapes. On the UMC CPUs and on the NOAC blobs descended from them, two of those shapes are swapped, so a sound written as a thin, reedy pulse comes out as a fuller one and the reverse. Pitch is unaffected. Music mostly sounds right; sound effects often do not. Adrian heard it immediately on every Dendy-chipped machine (video, 22:00), and NESdev and the OpenTendo top-loader project both document it.

The clone PPUs have their own quirks: the UA6538 is brighter and more saturated than a Ricoh PPU, and on Adrian’s bench a UA6538 still produced horizontal and diagonal artifacts when moved into a real front-loader, so on that unit the noise came from inside the chip rather than from the clone board (video, 48:00). That is one chip, not a family verdict, but it is why I treat a clone as a working substitute rather than a resale-grade repair, as I say on my top-loader page.

What is inside a clone console

“NES clone” covers four very different machines. Knowing which one you have decides what you can fix and what you can expect.

1. Discrete clone chips

A board that copies the Nintendo design chip for chip, but with UMC or other clone CPUs and PPUs in place of the Ricoh parts. This is the classic famiclone, the Dendy and Micro Genius family, and it is still what you get in the cheapest AliExpress consoles. Adrian’s red AliExpress top-loader had its chip markings painted over; acetone revealed a UMC 6527 and 6538 on a PAL crystal, which is a Dendy chipset. His read on pricing: the PAL version of a cheap clone is cheaper because plentiful Dendy chips go in it, and the NTSC version costs more because it needs different chips.

These are the most repairable clones, because every chip is in a socket-friendly 40-pin package that can be swapped for a genuine Ricoh part, crystal included.

2. NES on a chip (NOAC)

The entire console, CPU, PPU, sound and RAM, on one die under an epoxy blob on the underside of the board. This is the Yobo, the Retro-Bit, the RetroN 1, 2 and 3 (see my RetroN page), and most plug-and-play consoles with built-in games. Nothing inside is serviceable; a fault in the blob is the end of the board.

Adrian ran the AccuracyCoin test cartridge on a 2005 Yobo and a 2006-board Retro-Bit. The Yobo passed 111 of 131 tests against 130 on a real NES, and every game he tried on it played. Its picture was over-bright and over-saturated and its sound mix was wrong (the noise channel far too loud), which cannot be fixed because it happens inside the blob. The Retro-Bit was worse: its PPU could not do colour emphasis or the dimmed colour set at all, and its maker had left the video buffer transistor and the regulator’s protection diode off the board entirely. His verdict on both: curiosities, not something to buy.

3. Replica boards with genuine Nintendo chips

A new motherboard that uses real Ricoh CPU and PPU chips, usually pulled from scrapped consoles. Compatibility is 100 percent by construction, because the two chips that define compatibility are the originals.

The Lava RSC is the one Adrian recommends and now uses daily. It is a Famicom-slot board with socketed genuine Ricoh chips and support logic that mirrors the Nintendo board, plus an FPGA that reads the PPU’s EXT pins, the same approach as the NESRGB, and rebuilds the picture as RGB, S-video and composite. He measured the FPGA output trailing the PPU by about 145 nanoseconds, a fraction of one scanline, and the Zapper works on a CRT. Things to know about it:

• It takes 60-pin Famicom cartridges. American and European cartridges need an adapter.
• The chips are salvage, so it is a lottery which revision you get. Adrian’s shipped with mid-1980s E-revision chips and showed sparkles at colour edges; a G-revision PPU cured it. The same PPU was fine in a real NES, and his theory, which he did not measure, is that a weak EXT output from that chip sits near the switching threshold of the Lava’s 3.3 V level shifters (follow-up).
• The palette cannot be loaded from the SD card; palettes live in firmware, and the SD slot only flashes firmware. Buy the cheaper Lite model.
• The 10-pin mini-DIN AV port uses the Sega Saturn pinout, so Saturn cables work. No cable ships with it.

The other kind of replica board is the bare aftermarket Famicom motherboard sold without CPU and PPU, meant to take the chips from your own console. Adrian moved a genuine Famicom’s chips onto one (video) and found: a PAL crystal fitted at the factory with the NTSC one loose in the bag (so no picture until he swapped it; check yours), composite video far too hot with no DC-blocking capacitor, and socketed chips that fouled the cartridge eject mechanism. His verdict was “absolutely not”; for an RF-only Famicom, a drop-in AV board is less work and less risk than a full transplant.

4. Real CPU, FPGA PPU

The newest kind, and the one in Adrian’s latest video: an AliExpress Famicom-slot console (board marked QINGLIU) with a genuine socketed 40-pin CPU and an FPGA standing in for the PPU, outputting HDMI, VGA, and composite, S-video and RGB on a DIN jack. It is sold with a Dendy CPU, or with a genuine Ricoh 2A03 for a little more.

Because the FPGA generates the video, it always outputs 60 Hz, and two DIP switches inside the case set the clock it feeds the CPU. Set to match the fitted CPU, a Dendy CPU runs at exactly NTSC speed. Set to the other positions with a genuine 2A03 fitted, you get a live overclock of up to about 2.38 MHz. Adrian’s test game, Contra Force, went from badly laggy to nearly smooth, at the usual cost of shifted music pitch. What it does not do, disappointingly, is lift the eight-sprites-per-line limit: the FPGA PPU flickers exactly like a real one.

It also shipped with a design fault. VGA and HDMI were driven from the same DAC with no buffer, and the maker had put termination resistors on the board so HDMI looked right with nothing in the VGA port. Plug a monitor into VGA and both outputs dimmed; take them off and HDMI blew out. Adrian’s fix was to remove the VGA connector and rebalance the resistors so HDMI alone is correct, and to remove one small capacitor in the composite encoder’s filter, which made composite much sharper. With those mods he rated it good; without them, not recommended.

5. Emulation

Consoles that read the cartridge into memory and run it on a software or FPGA emulator, such as the RetroN 5, are a different animal entirely. My RetroN page covers how to tell them apart.

Faults nearly every cheap clone shares

Across the red top-loader, the bare Famicom board and the HDMI clone, Adrian kept finding the same analog mistakes. They are worth checking on any clone, and all of them are fixable:

• Composite video too hot. The buffer drives the jack with no series 75 ohm resistor, so the signal runs well above the standard 1 volt and the picture is washed out, with blown whites. A 75 ohm series resistor fixed the red clone (video, 40:00). An upscaler with automatic gain can hide this; a CRT will not.
• No DC-blocking capacitor on video. Nintendo’s own schematics have one; some clones do not, which some displays tolerate and some do not.
• Noisy switching power. Clones often replace the 7805 with a small buck converter, which runs cool, but the red clone’s had no output inductor and put crawling lines in the picture. An inductor of about 1 uH followed by ceramic capacitors to ground cured it. Feeding the board clean 5 V from a bench supply is the quick way to prove the supply is the culprit.
• Digital noise coupling into video. Holding Reset stops the CPU, so if the interference vanishes while Reset is held, it is board layout, not the supply. That one is not fixable short of a new board.
• Power switch after the regulator. The regulator runs, and wastes power, whenever the adapter is plugged in, even with the console off.
• Bridge rectifier input. Most clones rectify their input, so DC polarity does not matter. That is a convenience, but do not assume it for a genuine Famicom, which has no rectifier and dies on reversed DC (see the Famicom family page).

Identifying what you have

• Read the crystal. 21.477 MHz is NTSC. 26.60 MHz is PAL or Dendy.
• Read the two 40-pin chips. Ricoh RP2A03 and RP2C02 are genuine NTSC; RP2A07 and RP2C07 are genuine PAL; UMC and similar 65xx numbers are clones. If the markings are painted over, acetone will take the paint off. Adrian checked the Lava’s chips for authenticity by comparing them against revision photo references and checking that the round moulding marks were untouched, which a sanded and re-marked chip would lose.
• One blob and no 40-pin chips means a NOAC.
• Run the 240p Test Suite. It reports the CPU clock and region. About 1.79 MHz at 60 Hz is NTSC, about 1.66 MHz at 50 Hz is PAL, and about 1.77 MHz at 50 Hz is Dendy.
• Measure the divider. Master crystal frequency divided by the frequency on CPU pin 31 (M2) gives 12, 15 or 16.
• Listen. On a UMC CPU the sound effects have a different timbre while the music stays in tune.

Buying a clone

This is Adrian’s advice rather than mine, since I have not owned these, but it is consistent across every video:

• A replica board with genuine Ricoh chips (the Lava RSC Lite) is the only kind he recommends outright, and he considers it better than an original because of the video output. In North America, budget for a cartridge adapter.
• Cheap discrete-chip clones and NOAC consoles are curiosities. Most of them work with most games; none of them is an NES.
• The FPGA-PPU HDMI clone is good after mods and not before.
• Whatever you buy, check the crystal matches the chips you expect, and check the video level before blaming the console for a washed-out picture.

Sources and further reading

• NESdev wiki: Cycle reference chart (NTSC, PAL and Dendy clocks and timing)
• NESdev wiki: CPU variants (Ricoh revisions, UMC clones, relabelled UA6527P chips)
• NESdev wiki: PPU variants (UA6528, UA6538, UA6541 and NOAC PPUs)
• NESdev wiki: APU Pulse (duty cycles, and the famiclone swap)
• NESdev wiki: CPU pinout (CLK pin 29, M2 pin 31)
• Adrian’s Digital Basement: I found a new NES clone with unique features (video)
• Adrian’s Digital Basement: a cheap red NES clone with a real CPU and PPU (video)
• Adrian’s Digital Basement II: overclocking the NES CPU, part 1, with the Dendy explained (video)
• Adrian’s Digital Basement II: overclocking the NES CPU, part 2 (video)
• Adrian’s Digital Basement: Yobo and Retro-Bit NES clones (video)
• Adrian’s Digital Basement: Lava RSC famiclone review (video)
• Adrian’s Digital Basement II: the Lava RSC’s sparkly PPU in a real NES (video)
• Adrian’s Digital Basement II: the cursed Famicom on an aftermarket board (video)
• OpenTendo NES top-loader project (clone CPU duty-cycle note)