PPU variants

Beyond the well-studied 2C02G, we know of the following PPU revisions, both made by Ricoh and other manufacturers:

RP2C02	Extremely rare. Serial numbers suggest fewer than 6000 were made. (pictures not archived)
RP2C02A	Sometimes erroneous sprite pixels appear in X=255. Some modern PCBs generate almost exclusively glitchy pattern fetches.
RP2C02B	Sometimes erroneous sprite pixels appear in X=255.
RP2C02C	differences unknown.
RC2C02C	In ceramic package. [1]
RP2C02D	differences unknown.
RP2C02D-0	differences unknown.
RP2C02E-0	OAMDATA and palette RAM are not readable. Various OAM evaluation bugs
RP2C02G-0	Writes to OAMADDR cause corruption of OAM. Leaving OAMADDR at a value of 8 or greater causes OAM corruption when rendering starts.
RP2C02H-0	no known differences relative to 2C02G.
RP2C03B	RGB PPU. Color emphasis bits set the corresponding channel to full brightness. Otherwise believed identical to 2C02B. OAMDATA and PPU palette are not readable.
RP2C03C	RGB PPU. Believed to be same core as 2C02C.
RC2C03B	Tweaked palette; otherwise believed identical to RP2C03B.
RC2C03C	Believed identical to RP2C03C.
RP2C03G	No photos on the internet, probably does not exist. If exists, likely the same core as the 2C02G with the same bugs and features as the 2C02G.
RP2C04-0001	Scrambled palette and new colors, otherwise like 2C03.
RP2C04-0002	"
RP2C04-0003	"
RP2C04-0004	"
RC2C05-01	PPUCTRL and PPUMASK swap locations. Five LSBs of PPUSTATUS return a constant. Otherwise like 2C03.
RC2C05-02	"
RC2C05-03	"
RC2C05-04	"
RC2C05-99	Approximately equivalent to some 2C03. Used in the Famicom Titler. Behavior of five LSBs of PPUSTATUS unknown. (picture of Titler PCBs)
RP2C07-0	PAL-B PPU. Vblanking is 71 scanlines long. OAM evaluation can never be fully disabled. Red/green color emphasis swapped. OAMADDR works correctly, not corrupting OAM contents.
RP2C07A	Some subtle differences in PPU that make this work better with Kevtris's HDNES, but otherwise believed identical to 2C07.
RP2C07A-0	Differences unknown from RP2C07A
UA6528	UMC-made clone of 2C02E (or something earlier) [2]
UA6528P	UMC-made variant for PAL-N (Argentina) [3] System crystal is 21492337.5 Hz (exactly 6×229.2516×15625)
UA6538(P??)	UMC-made variant for playing NTSC games in PAL countries. Emits PAL-B video. Vblank IRQ intentionally emitted 50 scanlines later than 2C07. See also Clock rate.
UA6541	UMC-made clone of 2C07 [4]
UA6548	UMC-made variant for PAL-M (Brazil) System crystal is 21453671… Hz (exactly 6×227.25×4500000÷286)
UM6558	UMC-made variant of UA6538 for SECAM countries. Emits 8-bit "Color Data" digital bus, for conversion into SECAM by UA6559 IC. Color palette noticeably off. System crystal is 21312500 Hz (exactly 4×341×15625). Maybe supports both 50 and 60 Hz operation?
UM6561xx-1	UMC-made NES-on-a-chip for NTSC. PPU half believed identical to UA6528. Revisions "xx" AF, BF, CF, F known.
UM6561xx-2	UMC-made NES-on-a-chip for PAL-B. PPU half believed to be identical to UA6538. Revisions "xx" AF, BF, CF, F known.
TA-02N	??-made clone of UA6528.
TA-02NP	??-made clone of UA6538. Pins 14-17 are background color in, like normal 2C03[5]
TA-02NPB	It has unique feature of switching video region: pin 16: Video mode: 50Hz (GND) / 60Hz (5V) pin 17: Crystal select: 26.xxx MHz (GND) / 21.xxx MHz (5V) pins 14 and 15 were grounded but when connecting them to +5V, PPU displayed solid color screen and game did not boot. This one can be found in Rinco console (all Rincos have ability to switch video system, but some of them are build using blobs) and black one called "eastern computer"
T1818	??-made NES-on-a-chip, NTSC timing. Believed to exist, but evidence currently scant.
T1818P	??-made NES-on-a-chip.[6]. Requires external 2KiB RAMs for CPU and PPU. UA6538 timing.
PM02-1	Gradiente-made variant for PAL-M (Brazil). [7]
VT01	V.R.Tech-made clone of UA6561. Only seen as chip-on-board. Supports composite out; RGB out; or 2bpp STN LCD, either greyscale or red/cyan checkerboard, 240 px wide, 80/120/160/240px tall. The chip was extended significantly in VT02 and newer NOACs.
KC-6078	Found in MT777-DX famiclone, behaves exactly like UA6538

If you know of other differences or other revisions, please add them!

See also

• Clock rate
• NES_2.0#Byte_13_.28Vs._hardware.29
• https://forums.nesdev.org/viewtopic.php?p=150127#p150127