Myths: Difference between revisions

From NESdev Wiki
Jump to navigationJump to search
(→‎Old programs: link to list of behaviors deemed "quirky")
Line 41: Line 41:
'''Fact:''' No. Especially older NES programs tend to have been tested only on bad emulators. Your college-project emulator should match the behavior of an NES or at least that of an accurate emulator. Dirty iNES headers might break it.
'''Fact:''' No. Especially older NES programs tend to have been tested only on bad emulators. Your college-project emulator should match the behavior of an NES or at least that of an accurate emulator. Dirty iNES headers might break it.


A lot of emulators, especially prior to about 2005, were based on incomplete knowledge of how the NES works. Some old demos expect all internal memory ($0000-$07FF) to be $00. Since then, public knowledge of the quirks of the NES hardware behavior has grown, and emulators such as Nintendulator and Nestopia more faithfully reproduce the misbehaviors in sloppy or [[wikipedia:Cargo cult programming|cargo-cult-programmed]] code.
A lot of emulators, especially prior to about 2005, were based on incomplete knowledge of how the NES works. Some old demos expect all internal memory ($0000-$07FF) to be $00. Since then, public knowledge of the [[errata|quirks of the NES hardware behavior]] has grown, and emulators such as Nintendulator and Nestopia more faithfully reproduce the misbehaviors in sloppy or [[wikipedia:Cargo cult programming|cargo-cult-programmed]] code.
See [[Program Compatibility]] for a list of programs known to work and known not to work on an NES.
See [[Program Compatibility]] for a list of programs known to work and known not to work on an NES.



Revision as of 05:10, 23 May 2014

This page documents some myths about the NES hardware that originate in obsolete documents and emulators.

NTSC picture height

Myth: The NTSC NES picture is only 224 pixels tall.

Fact: The NTSC NES picture is 242 pixels tall: 240 lines of picture and 2 lines of vertical border. The PPU fetches and generates a signal for all 240 lines, even if TVs cut off the edges.

Ideally, NES games place nametable mirroring glitches in the overscan, and some emulators simulate overscan in order to hide these glitches. For example, PocketNES for Game Boy Advance hides the top 16 pixels, the bottom 11, and 8 on the left and right sides. The Mega Man Anniversary Collection (GameCube/XBOX/PS2) likewise hides the top/bottom 8 lines, plus an extra left/right clipping to hide the attribute color glitch on scrolling. Also, the Wii Virtual Console (NES) does not display the top and bottom 8 lines.

Scrolling registers

Myth: You always need to use the VRAM address register ($2006) to scroll.

Fact: The proper way to set the scroll position is to write the upper bits of the X and Y coordinates to PPUCTRL ($2000), and then bits 0-7 of the X and Y coordinates to PPUSCROLL ($2005). The NES will update the VRAM address register for you near the end of the pre-render scanline (261 on NTSC, 311 on PAL).

The only time you need to use $2006 to scroll, as described in Loopy's document "The skinny on NES scrolling", is if you're trying to change the vertical part of the scroll position during rendering time. This could happen if you turn rendering on late to free up more VRAM update time or if you are splitting the screen. Nintendo's Super Mario Bros. zeroes $2006 to set the scroll, but given that this was one of the first games with a scroll split, it might just be another example of imperfect hardware understanding. Launch titles tend to be full of these in part because the hardware itself may have been a moving target during the game's development. Another example is the horrible code in Tetris for Game Boy.

Mappers

Myth: The VRC6 is a very complex mapper even superior to the MMC5.

Fact: The VRC6 is a decent mapper able to do standard PRG and CHR bankswitching, a CPU cycle counter, and 3 extra sound channels. The MMC5 has extended video possibilities, a true scanline counter, and countless features that the VRC6 lacks, but only has 2 extra sound channels. Rumor has been made that the VRC6 was superior to the MMC5 because the MMC5 Castlevania III was censored, and (like other 72-pin games) didn't use extra sound. For the truth Castlevania III doesn't even come close to using all MMC5 capabilities, and it likely used MMC5 because it supported the VRC6's PRG ROM and CHR ROM bankswitching modes and was cheaper than Konami getting the VRC6 approved through NOA and NOE.

Largest game

Myth: Dragon Quest/Warrior IV (DQ4) is the largest NES game, having 1 MiB (1048576 bytes) of ROM.

Fact: All versions of DQ4 and the English version of Dragon Warrior III use SUROM. This board allows games to be as big as Mega Man VI, with 512 KiB PRG ROM and 8 KiB CHR RAM. This rumor was due to a 1MB overdump of DQ4 floating around.

Games larger than DQ4 include the following:

  • The largest licensed Famicom game is Metal Slader Glory (512 KiB PRG + 512 KiB CHR + MMC5).
  • The largest licensed NES game is Kirby's Adventure (512 KiB PRG + 256 KiB CHR + MMC3).
  • The largest unlicensed non-pirate NES game is Action 52 (1536 KiB PRG + 512 KiB CHR + custom mapper).
  • One large "Hong Kong Original" port is Final Fantasy VII (2048 KiB PRG + CHR RAM).
  • Some pirate multicarts are 4 MiB or larger. These cannot be represented in the the original iNES format, which has a practical limit of 2048 KiB PRG ROM and 1024 KiB CHR ROM. Use NES 2.0 instead.

Old programs

Myth: If a binary file has a .nes file extension, it will work as intended on an NES, and I should tweak my emulator to match how it is supposed to work.

Fact: No. Especially older NES programs tend to have been tested only on bad emulators. Your college-project emulator should match the behavior of an NES or at least that of an accurate emulator. Dirty iNES headers might break it.

A lot of emulators, especially prior to about 2005, were based on incomplete knowledge of how the NES works. Some old demos expect all internal memory ($0000-$07FF) to be $00. Since then, public knowledge of the quirks of the NES hardware behavior has grown, and emulators such as Nintendulator and Nestopia more faithfully reproduce the misbehaviors in sloppy or cargo-cult-programmed code. See Program Compatibility for a list of programs known to work and known not to work on an NES.

Old tutorials

Myth: GBAGuy's NES tutorial is worth following.

Fact: Old tutorials like these are full of cargo-cult programming because the authors apparently didn't fully understand the hardware. For example, this tutorial in particular treats the OAM address register as 16-bit (just like $2006) and attempts to initialize variables in system RAM using .db statements - a lot of the programs don't even work on a NES. NES 101 and Nerdy Nights are considered better. Even the webmaster of Patater.com now recommends Nerdy Nights.

PPU details

Myth: The sprite address ($2003) must be cleared on VBlank ending.

Fact: The PPU itself sets the sprite address to zero at the end of VBlank, but due to a design flaw it can result in minor sprite RAM corruption if it was nonzero beforehand - in particular, it can cause values from one 8-byte "page" of sprite RAM to leak into another due to its lack of proper memory refreshing. Some Chinese games actually rely on this behavior and will lock up otherwise.

Color emphasis

Myth: Enabling more than one color emphasis bit at once will damage the PPU, or at least cause the TV to lose sync.

Fact: Enabling multiple color emphasis bits is perfectly safe - in fact, some licensed games including Felix the Cat and Just Breed enable all of them simultaneously to dim the screen. On the other hand, enabling all emphasis bits results in an unreadable white screen on an RGB PPU, such as that in the Famicom Titler or the Famicom TV. Worse, setting the PPU into slave mode (by setting the master/slave bit in register $2000) is definitely capable of causing damage, as it results in extremely high current draw from the EXT pins (due to them trying to output +5V despite being wired to GND).