Decades after IBM introduced the Multi-Color Graphics Array with its PS/2 line, a lone reverse engineer has pried open the silicon. The chips that powered low-end 1987 machines now sit exposed at the gate level. Their secrets challenge old assumptions about what this short-lived standard could actually do.
MCGA arrived on April 2, 1987, baked directly into the motherboards of the IBM PS/2 Model 30. The Model 25 followed months later. No standalone cards ever shipped. (Wikipedia)
It delivered all the modes of the earlier Color Graphics Adapter. Then it added two signature capabilities. One offered 320 by 200 resolution with 256 colors drawn from a palette of 262,144. The other gave 640 by 480 in monochrome. Refresh rates hit 70 hertz for the color mode and 60 hertz for the high-resolution black and white. A 15-pin analog connector, the same DE-15 later used by VGA, handled the output. (Wikipedia)
Hardware-wise the subsystem relied on two custom gate arrays. The 72X8300 managed memory control, sync generation, and character RAM. The 72X8205 handled address decoding, RAMDAC interface, and pixel formatting. A 256-entry by 18-bit palette chip, often an Inmos IMSG171, completed the video chain along with 64 kilobytes of multi-port video DRAM. (Ardent Tool)
IBM positioned the PS/2 family as a fresh start. Micro Channel architecture, new keyboards, 3.5-inch floppies. Graphics formed a key part of the pitch. Higher-end models received full VGA. The entry-level machines got MCGA, a cut-down variant that shared the 256-color mode and analog connector but dropped EGA compatibility and higher-resolution color. (Hackaday)
The strategy backfired. Third-party VGA cards flooded the ISA market. Buyers upgraded existing PCs instead of purchasing expensive PS/2 systems. MCGA lingered only in the low-end Model 25 and 30 before those machines disappeared by 1990. Epson produced the only notable clone in its Equity Ie. (Wikipedia)
Yet the 256-color 320-by-200 mode found fans. Mode 13h in programmer parlance became a staple for DOS games and the PC demoscene. It offered a straightforward linear framebuffer. Developers could ignore the complexities of planar EGA or the palette tricks of earlier standards. (Hackaday)
Fast forward to 2026. Andrew Schlae, operating under the GitHub handle schlae, decided the chips deserved closer inspection. He acquired decapped and delayered specimens. Then he traced every transistor. The resulting repository contains full schematics and layouts for both gate arrays, rendered in KiCad after scaling massive die photographs from SiliconPr0n. (GitHub – schlae/IBM_MCGA)
The work revealed more than expected. New capabilities surfaced that IBM never documented in the PS/2 technical reference.
One stands out. MCGA can genlock to external horizontal and vertical sync signals. These appear on pins 11 and 12 of the video connector, labeled as ID0 and ID1. Set bit 3 of register 0x12 to one and the subsystem switches into this mode. The original manual listed that bit as reserved and zero. An external PLL would be required to lock the internal clocks. Still, the hardware support exists. (GitHub – schlae/IBM_MCGA)
Other undocumented registers and bits emerged. Register 0x10 controls mode details including a compatibility flag that doubles horizontal timings for 80-column text. Bit 2 selects between 25.175 MHz and 14.318 MHz clocks. Register 0x1A contains unknown bits that may force 256-color output or modify border colors. Then come the manufacturing test modes accessed through registers at 0x19 and 0x18. These allow direct reads of RAMDAC and VRAM data, speedups on counters, and output disables for chip testing. (GitHub – schlae/IBM_MCGA)
The two chips themselves tell a story of IBM manufacturing in the late eighties. The memory controller appeared in both internal IBM processes and Seiko versions. The video formatter showed similar variation. One specimen required top-metal removal during decapping, complicating the imaging. Transistor counts reached 4342 for the larger array and 3312 for the smaller, built on a two-micron CMOS process with two metal layers. Layout details show long horizontal runs on metal one and vertical power distribution on metal two. (GitHub – schlae/IBM_MCGA)
Interest in the project surfaced quickly. Hacker News threads and X posts circulated the repository within hours of its activity spike. Retro computing enthusiasts already familiar with PS/2 hardware saw immediate value. The schematics open the door to potential Verilog models, accurate emulators, or even modern recreations using programmable logic. (GitHub – schlae/IBM_MCGA)
Compatibility quirks of the original implementation remain relevant for collectors. MCGA senses the attached monitor and adjusts. Monochrome displays receive summed RGB signals on the green pin. The subsystem refuses to run EGA software, forcing games to drop to CGA four-color modes or fail entirely. Some titles shipped with dual assets to handle the limitation. (Ardent Tool)
Disabling the onboard MCGA for add-in cards required toggling bit two of system register 65h. Early upgrade attempts from MCGA to full VGA demanded careful configuration. The IBM PS/2 VGA Display Adapter card eventually provided that path, delivering better EGA support as a bonus. (Ardent Tool)
Today the PS/2 platform occupies a curious place in PC history. It introduced standards that outlived the machines themselves. VGA became ubiquitous. The Model M keyboard achieved legendary status. MCGA, however, faded into footnotes. Its 256-color mode lived on in software long after the hardware vanished from new systems.
Schlae’s reverse engineering changes that narrative. What looked like a cost-reduced compromise now appears more capable. Hidden genlock support hints at video applications IBM never pursued. Test modes expose the care taken in factory validation. And the complete gate-level view gives the community raw material for preservation that no technical reference manual could provide.
The work also highlights how much still hides inside vintage silicon. Gate arrays from the late eighties, dismissed as obsolete, surrender new details when examined with modern tools and patience. Similar projects have illuminated processors, custom sound chips, and graphics controllers from the same era. Each adds texture to a history once considered settled.
MCGA never dominated the market. It never received the software library that VGA enjoyed. Yet its influence on early 256-color PC gaming deserves recognition. The fresh documentation ensures that influence can be studied, emulated, and perhaps extended with greater accuracy than before.
So the chips rest in their GitHub repository. Schematics. Die photos. Netlists waiting for Verilog conversion. A quiet rebuke to the notion that old technology holds no more surprises. IBM engineered these parts for machines that barely outlasted the decade. Their digital descendants continue to reveal themselves nearly forty years on.
WebProNews is an iEntry Publication