What Intel actually says
Intel's design notes for chipset real-time clocks (RTCs) recommend a Duracell 2032, 2025 or 2016. All three are 3V cells and 20 mm across. They differ in thickness and capacity. Panasonic's figures put the 3.2 mm CR2032 at 225 mAh and the 1.6 mm CR2016 at 90 mAh. So Intel's own guidance allows a thinner, smaller-capacity cell.
Intel's current 800-series chipset documentation tells the same story from the other end. It describes an RTC that's functionally compatible with the Motorola MC146818B, with 256 bytes of battery-backed RAM. The power source is a 3V battery, or a system battery if the design uses one. Neither document mandates a particular cell.
The PC/AT started with 6V, not a coin cell
IBM's March 1984 PC/AT technical reference describes battery backup for the CMOS configuration table and the clock, built around a Motorola MC146818. The chip held the clock plus 64 bytes of CMOS RAM. Software reached it through I/O ports 70h and 71h, and modern PCs still use those ports.
The AT's battery was a 6V lithium pack on a cable. It plugged into a four-pin connector labelled J21, with 6V on pin 4. XDA's explanation is that the clock chip's minimum operating voltage was 3V. The battery also fed the chip through a pair of isolation diodes, and those diodes would drop voltage. A 3.6V cell could then sag below the minimum. The 6V pack gave the circuit headroom. That wiring explanation is XDA's account of the Motorola datasheet and IBM's design. I haven't independently verified it.
If the battery died, the machine flagged a 161 error. IBM's suggested workaround, as XDA quotes it, was to run Setup every time you powered on.
The CR2032 was already in PCs, just not on the motherboard
The original IBM PC and XT had no battery-backed clock on the motherboard. Expansion cards filled the gap. XDA cites AST's SixPakPlus, which had a clock-calendar with a user-replaceable lithium cell. A May 1984 AST bulletin reportedly listed Sanyo and Varta CR2032 cells as compatible replacements. So the cell was keeping time in PCs in the mid-1980s. That doesn't mean motherboards had standardized on it.
A decade of battery chaos
Clone makers used whatever suited them. XDA cites several examples:
- External packs. Elitegroup's 1988 ECS 386/32 manual says the board has no onboard battery and needs an external pack. Asus's 1991 ISA-386C shipped with a separate 3.6V battery.
- Rechargeable cells on the board. Some boards used nickel-cadmium "barrel" batteries. These are notorious among collectors for leaking. XDA quotes Michal Necasek of the OS/2 Museum describing "a dreaded rechargeable NiCd barrel battery" on a 386 board.
- Sealed clock modules. Dallas Semiconductor's DS12887 put a lithium cell and crystal inside the clock chip. It was rated for over 10 years without power. When it ran flat, you usually replaced the whole module. These modules lasted into the Pentium era.
- Early coin cells. IBM used a flat 3V coin cell on its PS/2 Model 90 and 95 in 1990. An August 1994 Aptiva handbook told owners to replace the battery with a 3V CR2032 only.
Why chipsets tipped the balance
The key change was where the clock lived. Once chipset makers built the RTC into their own silicon, motherboard makers could feed backup power straight into a battery pin. XDA cites the SiS 85C497 (July 1995), which listed a current draw of 2.02 µA at 3V. It also cites VIA's VT82C586A (October 1996), which has a VBAT pin for the internal RTC. Intel's PIIX4 followed in April 1997, with a separate 3V lithium battery rated for up to seven years.
A clock drawing a couple of microamps can run for years on a CR2032. A replaceable coin cell was the practical option.
XDA also tracks the shift across Asus boards. The April 1996 P/I-P55T2P4 still used a Dallas module. The December 1996 P/I-P55SP4V, on a SiS chipset, shows a CR2032 holder. The PIIX4-based TX97 (1997) and P2B (1998) do too. XDA notes the TX97 is a Baby AT board, which is why it argues ATX wasn't the cause. A 2001 Que A+ study guide said the CR2032 had become the most common removable battery on Pentium-class systems, while still listing the Dallas module.
The "late 1990s" label describes this broader trend, based on these board examples. It isn't a precise industry-wide cutover date.
How long does it last, and what depends on it?
Intel's RTC note gives an example calculation of about 3.88 years of minimum battery life. The figure rests on stated assumptions. It assumes a 170 mAh usable capacity and a 5 µA average draw, which works out to about 34,000 hours. Intel adds that standby power significantly extends battery life. On a desktop plugged into a live power supply, the coin cell only covers the time when power is fully removed. In practice it usually lasts far longer than the example.
The battery also matters less than it once did. XDA notes that most modern boards keep BIOS settings in non-volatile flash. The cell mainly keeps the clock and a small block of CMOS memory alive.
What this means for Windows users
- Windows has long had its own clock conventions. XDA cites Raymond Chen's 2004 explanation that Windows traditionally kept the hardware clock on local time. The reason was dual-booting with MS-DOS and Windows 3.1, which also use local time.
- Windows can now cope without a dedicated battery. XDA says Windows 10 added Secure Time Seeding in November 2015, which estimates the time from network connections. Microsoft's stated view is that PCs and big form-factor hardware have a dedicated clock battery, while smaller devices often don't. XDA also notes the feature has misfired in the past, with some servers' clocks drifting weeks or years.
- A dead battery usually means a wrong clock, not a dead PC. That's the common symptom on desktops. A clock reset to a default date, or a BIOS time prompt, points at the cell.
Machines that skip the coin cell
Framework is the clearest consumer example. Its knowledge base says no RTC/CMOS battery is required or included with the Framework Laptop 13 (AMD Ryzen 7040 Series) mainboards. Framework's RTC rework documentation adds that if both AC and the main battery are disconnected, the clock resets, and the OS can recover it once it connects to a network. The same page says 13th Gen and Ryzen 7040 Series work this way by default.
Framework's earlier 11th Gen boards were different. They require an RTC battery to boot, and on some systems a low voltage can push the CPU into a bad state. That is why Framework offered a substitute module. It's a cautionary example: a coin cell isn't always harmless to remove. This is design-specific, so don't copy it on a desktop board.
Dell is taking a different route on servers. Its support article says that since August 2024, iDRAC and BIOS firmware on 14th-generation PowerEdge servers changed how they handle coin-cell failure. With the updated firmware, removing the battery no longer resets BIOS settings, matching AMD-based PowerEdge behavior. The change reached 15G and 16G platforms from February 2025.
Dell also logs less noise now. A failed cell produces one informational event, UEFI0447, instead of three critical or warning events and an F1/F2 prompt. Dell says most customers likely use NTP and won't notice. It also says the CMOS battery sensor has been disabled and no longer appears in management interfaces such as iDRAC, Redfish and SNMP. Dell's guidance is specific to those servers and firmware versions. Don't generalize it to consumer boards.
Swapping the battery: practical notes
- Check your motherboard manual for the specified cell type and polarity before buying a replacement.
- Power the PC down and unplug it first.
- Expect to re-enter the date and time afterward. Check your BIOS settings too, since some boards reset them.
- Don't swap in a CR2016 just because it's the same diameter and voltage. It's half as thick, so a holder built for a CR2032 may not make proper contact.
- Don't put a non-rechargeable cell in a system designed for a rechargeable one. Framework specifically warns about this for its laptops.
The verdict
The CR2032 became the desktop default because integrated chipset RTCs made a cheap, replaceable 3V cell the obvious choice. It was never a specification requirement. The same RTC design that IBM shipped in 1984 still sits behind your motherboard's clock today. Some makers are now questioning whether the coin cell needs to exist at all. On desktops, which have no main battery to fall back on, that question has no practical answer yet.
References
- Your motherboard's CR2032 battery was never a standard, it just won by accident in the late 1990s XDA · 2026-11-17T21:31:21+00:00
- PowerEdge: Coin-cell (CMOS) battery changes | Dell Asia Pacific dell.com
- Does the Framework Laptop 13 (AMD Ryzen 7040 Series) include a RTC/CMOS battery on the mainboard? knowledgebase.frame.work