High-Speed SIO: Warp Speed, UltraSpeed, XF551 and POKEY Divisors Explained

BASIC screen with a table of POKEY divisors 40, 16, 10, 9, 8, 6, 3, 1 and 0 and their NTSC and PAL bit rates, from 19040/18866 to 127840/126674

Every stock Atari disk drive talks at about 19,200 baud. That is fast enough for a floppy, but slow next to modern SD and network devices that could easily go five times faster. From 1982 on, drive upgrades added their own high-speed modes, each with its own way of asking for it. Today SIO2PC software, SD drives and FujiNet all emulate one or more of those protocols. SIO Explained covers the basic bus and mentions the XF551 and US Doubler methods. This guide goes further: how the bit rate is set, what each protocol actually does, why interleave matters as much as baud rate, and how to switch it all on.

What you'll learn: the divisor formula, the main high-speed protocols, what the computer side needs (DOS, OS patch or firmware), hardware and device settings, and Altirra's high-speed emulation.

What you need: Atari BASIC to run the examples. To use high speed for real you also need a high-speed-capable drive or SIO device and software on the Atari that speaks its protocol.

1. Divisors and bit rates

The Atari's serial port is clocked by POKEY. For SIO, the OS joins audio channels 3 and 4 into one 16-bit counter running from the 1.79 MHz clock, by writing $28 to AUDCTL (53768, $D208). The value loaded into AUDF3 and AUDF4 is the POKEY divisor, and the bit rate it gives is:

bit rate = clock / (2 × (divisor + 7))
clock    = 1,789,772.5 Hz (NTSC)   or   1,773,447.5 Hz (PAL)

A high-speed routine switches POKEY like this:

AUDF3   = $D204
AUDF4   = $D206
AUDCTL  = $D208
        org $2000
sethi   lda #$28        ; ch.3 clocked at 1.79 MHz, ch.4 clocked by ch.3
        sta AUDCTL
        lda #10         ; divisor 10 = US Doubler rate
        sta AUDF3
        lda #0          ; high byte of the 16-bit divisor
        sta AUDF4
        rts

This BASIC program prints the rates for the divisors you will meet:

10 REM POKEY DIVISOR TO BIT RATE
20 PRINT "DIVISOR","NTSC","PAL"
30 FOR I=1 TO 9:READ D
40 N=INT(1789772.5/(2*(D+7)))
50 P=INT(1773447.5/(2*(D+7)))
60 PRINT D,N,P:NEXT I
70 DATA 40,16,10,9,8,6,3,1,0
Divisor NTSC bit/s Who uses it
40 19,040 Standard SIO ("19,200 baud")
16 38,908 XF551, Happy 1050, Indus GT Synchromesh, SIO2PC "38,400"
10 52,640 US Doubler and its clones
9 55,930 Speedy 1050, Speedy XF
8 59,659 SIO2PC "57,600"
6 68,837 Indus GT SuperSynchromesh, 1050 Turbo
1 111,860 SIO2PC "115,200"
0 127,840 The fastest POKEY can go

These are bit rates. The gaps between bytes also count, so real throughput is lower. Transmitter and receiver must agree within about 5%, and the 1% difference between NTSC and PAL clocks can push a marginal device over the edge. Early XF551 firmware's high-speed mode failed on PAL machines for exactly that reason. The Speedy 1050 has the opposite problem: through firmware 1.6 its receive loop is slightly too slow for an NTSC computer, so high-speed writes fail with error 139. The fixed 1.7 firmware is part of the MegaSpeedy add-on.

With the screen on, about 60–70 kbit/s is the practical ceiling. Above that, display DMA and the vertical blank interrupt steal enough cycles that the 6502 starts dropping bytes. Divisor 0 really wants the screen off or very tight code.

2. The protocols

High speed was never standardised, so each drive family asks for it differently.

  • XF551 style. The command frame goes at 19,200 baud. Setting bit 7 of the command byte asks for a high-speed data frame ($D2 instead of $52 for Read, $D0/$D7 for Put/Write, $D3 for Status). The Indus GT uses the same commands. The XF551 offers high speed only with double-density (360K double-sided) disks.
  • Happy (Warp Speed). The Happy 810 rev. 7 and Happy 1050 add separate high-speed commands: $70 Put, $72 Read and $77 Write, with the command frame again at normal speed. The Happy 1050 also emulates the US Doubler.
  • US Doubler (ICD's "UltraSpeed"). The computer sends the whole command frame at high speed. The drive notices and answers at that speed, with no special command codes. A Get High Speed Index command ($3F) returns the divisor to use, so software can support any speed with one code path. The catch is that many such drives listen at one speed at a time and alternate, so the first high-speed frame is often missed. Retries are part of the protocol.
  • The Speedy 1050 runs at divisor 9, about 56K. Indus GT firmware has Synchromesh (38K). The GTSYNC.COM utility uploads improved Synchromesh and SuperSynchromesh (69K) code to the drive. The 1050 Turbo uses bit 7 of AUX2 to request a high-speed data frame, at about 68,000–70,000 bit/s.

There's no reliable way to ask a drive which of these it supports. The US Doubler query is the closest thing to a standard, so it's a good first test. This program asks D1: for its index, sending a Status command first to make sure a drive is there:

10 REM ASK D1: FOR ITS HIGH SPEED INDEX
20 FOR I=0 TO 4:READ B:POKE 1536+I,B:NEXT I
30 DATA 104,32,89,228,96
40 POKE 768,49:POKE 769,1:POKE 774,7
50 POKE 772,64:POKE 773,6:POKE 777,0
60 POKE 778,0:POKE 779,0
70 POKE 770,83:POKE 771,64:POKE 776,4
80 X=USR(1536):IF PEEK(771)<>1 THEN PRINT "NO DRIVE 1":END
90 POKE 770,63:POKE 771,64:POKE 776,1
100 X=USR(1536):ST=PEEK(771)
110 IF ST=139 THEN PRINT "NAK: NO US DOUBLER-STYLE HIGH SPEED":END
120 IF ST<>1 THEN PRINT "ERROR ";ST:END
130 D=PEEK(1600):PRINT "HIGH SPEED DIVISOR ";D
140 PRINT "= ";INT(1789772.5/(2*(D+7)));" BIT/S (NTSC)"

Against atari800's default emulated drive it prints the NAK message, as a stock 1050 would. Command $3F is also the "type 1 poll" that an 850 interface answers. Leaving AUX1 at 0 and confirming the drive first, as here, avoids the 850 replying on a missing drive's behalf.

3. Speed is useless without interleave

The disk turns at 288 RPM, 208 ms per revolution. A standard single-density read takes about 91 ms end to end, so sectors are laid out 9:1 (1, 10, 2, 11, …) to catch roughly two per revolution. Send the data faster and the drive is simply ready early: it then waits for the same sector to arrive, and nothing is gained. Real gains come from formatting with a tighter high-speed skew that matches the faster transfer. XF551 high speed, for instance, lets double-density disks use 9:1 instead of 15:1. That is why high-speed formatters exist (US Doubler command $66 "format skewed", XF551/Indus $A1 "format with high-speed skew"), and why SpartaDOS X's FORMAT asks for Standard or High Speed skew. A high-speed-skewed disk reads slowly at normal speed, so don't format that way for a friend with a stock drive.

4. The computer side

The stock OS SIO routine only does 19,200 baud, so something on the Atari must supply high-speed code:

  • The DOS. ICD's SpartaDOS X loads SIO.SYS (after SPARTA.SYS in CONFIG.SYS) for US Doubler, Indus, Happy 1050 and XF551 high speed. INDUS.SYS programs Indus and Happy drives with working high-speed code. Current SpartaDOS X lists Happy/Warp Speed, US Doubler/UltraSpeed, Indus Synchromesh and Synchromesh II, XF551 and Speedy support. Programs can call SpartaDOS's own routine through the LSIO pointer at COMTAB−10 instead of SIOV ($E459). XF551 high speed is also supported by DOS XE, XDOS, MyPicoDos, Micro-SpartaDOS, SuperDOS, Turbo-DOS and BW-DOS. Because SpartaDOS X bypasses the OS SIO routine, emulator SIO patches don't speed it up.
  • A patched OS. Matthias Reichl's (HiassofT) high-speed SIO patch for the XL OS adds high-speed support to the OS's own SIO routine, so DOSes and loaders that call SIOV benefit without drivers of their own. Commercial equivalents included CSS's Ultra Speed OS (1986) and Atarimax's APE Warp+ OS (3× SIO with US Doubler emulation).
  • Firmware. Ultimate1MB machines offer high-speed SIO from their BIOS. Jonathan Halliday's Incognito firmware includes a PBI implementation of Reichl's high-speed code, which needs no OS patch and works with external cartridges, and also handles SIO2BT and PCLink at speed.

Some titles misbehave when loaded faster than stock. If one fails, drop back to 19,200 before suspecting the image.

5. Devices and settings

  • RespeQt supports up to about 6× SIO speed with a good FTDI adapter. You can set a baud rate or a "non-standard" POKEY divisor (users report divisor 10 shown as about 52,400). RespeQt switches back to 19,200 when it receives a command it can't understand at the higher rate, so the status bar only shows the fast rate once the Atari is actually using it. With the Software (SIO2BT) handshake the rate options are hidden.
  • APE expresses speed as a divisor: $28 standard, $10 about twice as fast, $08 about three times.
  • FujiNet has an HSIO index setting (hsioindex in fnconfig.ini, or SIO command $E3 to device $70 with values 0–10). Reboot the Atari after changing it, because most SIO routines read the speed only once.
  • SIO2IDE runs US Doubler and Happy high speed at 52,000 bit/s. The sio2flip Flipper Zero app emulates US Doubler and XF551 modes. A8PicoSIO uses separate PAL and NTSC rates to match POKEY closely.
  • XL/XE hardware: many users found low divisors unreliable until they removed the filter capacitors on the SIO data and command lines (with a resistor change, per the SIO2SD project's high-speed notes).

6. In Altirra

Altirra's Disk Drives dialog sets an emulation mode per drive, and each mode has its own high-speed rate:

Mode RPM High speed
Generic (default) 288 38,400 (XF551 style)
810 / 1050 288 none
XF551 300 38,400
US-Doubler, Happy 288 52,000
Speedy 1050 288 56,000
Indus GT, 1050 Turbo 288 68,000

Altirra models the transfer timing for all of these, but implements the drives' extended commands only for the 1050 and XF551. For sheer speed, the SIO patch intercepts OS disk calls and completes them almost instantly. Burst transfers speed up custom loaders that bypass SIOV, up to about 90,000 baud. Turn both off when you want to test real high-speed behaviour, or when a protected disk objects.

Troubleshooting

  • Works at 19,200, fails faster: lower the speed one step, shorten or remove other SIO devices, and consider the capacitor modification above.
  • Drive found but always slow: your DOS or OS has no routine for that drive's protocol. Load the right SpartaDOS X driver, or use a patched OS.
  • Every high-speed write NAKs (error 139) on NTSC: check for an old Speedy firmware.
  • Disk reads slowly on someone else's drive: it was formatted with high-speed skew. Reformat with standard skew for distribution.
  • 850 interface plus drives: if a $3F query gets an odd answer, the 850 may be replying. Confirm the drive first, as in the example.

Further reading

  • Altirra Hardware Reference Manual (Avery Lee), Chapter 10 Disk drives: high-speed protocols, Table 45 (transfer rates) and Table 50 (per-drive timings).
  • SpartaDOS Construction Set manual (ICD), Appendix G, which documents the US Doubler commands.
  • SpartaDOS X Reference Manual (ICD), chapter 8: SIO.SYS and INDUS.SYS.
  • The comp.sys.atari.8bit FAQ, section 3.2 (drive enhancements and their high-speed modes) and 7.1.2 (modified OS ROMs).
  • Our guides SIO Explained, SIO2PC and RespeQt and FujiNet Setup Guide.

Credits