S612 to S5000: Which Akai Sampler Do You Actually Want?

An Akai S950 rackmount 12-bit sampler, front panel and floppy drive

The Akai S612 arrived in 1985 with 128KB of memory. That buys one sample, about a second of it at the top rate, stored on a 2.8in Quick Disk that held one sample per side. Thirteen years later the S6000 shipped with 256MB and 128 voices. Same badge, nothing else in common.

Which is why the Akai S950 vs S1000 question, the one every forum thread opens with, is really a question about what you want the machine to do. One is a character box with 750KB. The other is a 16-bit workhorse that was in every studio in 1992 and is cheap now because everyone already has one. They are not competing.

Akai S950 vs S1000, and the rest of the family, in one table

Model Year Bits Sample rate RAM, standard to max Voices Filter Disk Buy this if
S612 1985 12-bit linear 4 to 32kHz 128KB, fixed 6 Variable analogue low-pass MD280 Quick Disk You want a machine fighting its memory
S900 1986 12-bit linear 7.5 to 40kHz 750KB, fixed 8 Analogue switched-capacitor, no resonance 3.5in DD, 800KB S950 grit, found cheap
S950 1988 12-bit linear 7.5 to 48kHz 750KB to 2.25MB 8 As S900, plus an envelope in the OS 3.5in DD or HD, 1.6MB Drums, jungle, anything that should sound like 1993
S1000 1988 16-bit linear 22.05 or 44.1kHz 2MB to 32MB 16 18dB/octave digital, no resonance Floppy; SCSI optional card Clean multisamples, long loops, little money
S1100 1990 16-bit linear 32, 44.1 or 48kHz 2MB to 32MB 16 As S1000 Floppy, SCSI standard S1000 sound plus SCSI, digital out, effects, SMPTE
S1100EX 1990s 16-bit linear As S1100 Its own, to 32MB 16 more As S1100 None You own an S1100 and ran out of voices
S2000 1995 16-bit 22.05 or 44.1kHz 2MB to 32MB 32 2-pole resonant digital Floppy, SCSI standard Budget. It is the cheap one and it is fine
S3000 / S3200 1992 16-bit 44.1kHz 2 or 8MB to 32MB 32 Resonant digital low-pass Floppy; S3200 adds SCSI, digital I/O, SMPTE Resonant filter and 32 voices, cheaply
CD3000XL Mid 1990s 16-bit 44.1kHz 8MB to 32MB 32, 26 when streaming 12dB/octave resonant, 24dB with IB304F Quad-speed CD-ROM, floppy, SCSI You have a shelf of Akai library CDs
S5000 1998 18-bit A/D, 20-bit D/A 44.1 or 48kHz 8MB to 256MB 64, 128 with the board 26 resonant types, 2 and 4-pole Floppy, SCSI, Zip; reads DOS and WAV Hardware that behaves like a computer

One thing that table settles. Everything from the S2000 onwards has a resonant filter and nothing before it does, which is the real dividing line in the range, and it leaves the S1000 and S1100 as the only machines that are both 16-bit and non-resonant. That is why they sound like themselves.

"I'm in the market of buying a vintage Akai S sampler and my initial options are the S950 and the S1100"

A real opening line from a real thread, and a good one: those two sit at opposite ends of the same decision. The S950 is 12-bit, 8 voices, 750KB, and its appeal is what it does to audio. The S1100 is 16-bit, 16 voices, 32MB, and its appeal is that it gets out of the way. If you cannot say which you want, you want the S950: the S1100's virtues are all available in software and the S950's are not.

The S900 vs S950 difference, in one sentence

Higher top sampling rate and more memory, and that is most of it. The S950 samples to 48kHz where the S900 stops at 40kHz, takes two EXM006 boards to 2.25MB where the S900 is stuck at 750KB, and uses HD floppies at 1.6MB against 800KB. It adds timestretch, crossfade looping and an envelope over the filter. Owners with both, on Gearspace, mostly report the sonic difference as negligible at matched rates; one put it as "at 40k I can't tell the difference". The filter hardware appears identical in both, an analogue switched-capacitor low-pass with no resonance, the S950 adding only OS control over it. That comes from owner reports, not Akai.

S1000 vs S1100: what the extra number buys

The S1000 is 16-bit at 22.05 or 44.1kHz, 16 voices, an 18dB per octave non-resonant digital low-pass, 2MB standard. SCSI is an optional card, not standard, which catches buyers out. Memory went to 8MB on EXM002 cards, then 32MB once the EXM008 boards arrived in 1990.

The S1100 keeps the engine and fixes the plumbing. SCSI standard, a digital output, an effects section built on a Motorola DSP56000, and SMPTE read and write, which is how it ended up in post rooms firing effects off timecode. It adds 32 and 48kHz sampling and records to hard disk, subject to an Akai formatting limit of 500MB per volume however big the drive. One source gives it 20-bit converters against the S1000's 16-bit, with a hotter output stage, though samples are still 16-bit; single source, unconfirmed.

What the S1100EX actually adds

Voices and outputs, nothing else. The S1100EX is an S1100 with the front panel, inputs and disk drive deleted, driven entirely from the host: 16 more voices, its own memory, eight more individual outputs and a stereo pair. Audio inputs on the back let it cascade, so a stack still lands on two mixer channels. Zoe Blade's notes put the ceiling at six per S1100. On its own it is a paperweight.

The S3200, the CD3000XL and the S5000

The S3200 is where Akai fitted a proper resonant filter and doubled polyphony to 32 voices, with SCSI, AES/EBU and SMPTE standard rather than optional. Sources disagree on its year: Reverb says 1992, Vintage Synth Explorer says 1996, probably the later XL revision bleeding into the entry. The CD3000XL is that XL-era engine with a quad-speed CD-ROM drive bolted in and a 12dB per octave resonant filter, 24dB multimode with the IB304F board fitted.

The S5000 is the forgotten one. 18-bit converters in, 20-bit out, 8MB to 256MB on ordinary SIMMs, 64 voices, 26 resonant filter types, two SCSI ports, and it reads DOS media and plain WAV files. No Akai format, no conversion step: the only one here you load straight from a PC.

The S612, and the AX60 trick

The S612 with its MD280 drive is a museum piece with one genuinely useful job: a 13-pin DIN voice output. Plug that into an AX60 and the sampler becomes a digital oscillator feeding the AX60's CEM3394 four-pole resonant filter, envelopes and VCA. A sampled waveform through a real analogue 24dB filter, in 1986.

12-bit vs 16-bit sampler: where the grit actually comes from

Search this and you get hi-fi articles explaining quantisation noise. Useful for choosing a DAC, useless here, because bit depth is the smallest of the four things doing the work.

Why do 12-bit samplers sound better? Mostly they do not

Bit depth sets a noise floor and nothing else. Twelve linear bits gives roughly 72dB of theoretical signal-to-noise, sixteen gives roughly 96dB. On a loud drum loop that is a faint grain under the decays. Truncate a 16-bit file to 12 bits in a DAW and you will strain to hear it. If bit depth were the magic, the plug-in would have killed this market years ago. Four other things do the work.

Bandwidth, not bit depth. On the S950 you buy sampling time by dropping the rate, and the machine closes an anti-aliasing filter as you do. It is not set at half the sample rate: one well-informed Gearspace poster works it out as roughly the rate divided by 2.5, so a 19.2kHz bandwidth setting means sampling at 48kHz, because no real filter passes 24kHz and stops 24.001kHz. It is a switched-capacitor type with something like 36dB per octave of slope, described by the people who modelled it for the RX950 plug-in as a sixth-order Butterworth. A wall that steep has its own ringing and phase behaviour, and it does far more than four missing bits.

Transposition by clock, not interpolation. A DAW holds one project sample rate and interpolates to change pitch. The S900 and S950 run a variable playback clock per voice: pitch a sample down two octaves and that voice is genuinely being clocked at a quarter of the rate. As one Gearspace poster put it, dropping a break an octave in a modern interpolating sampler makes the attacks "mushy", where on the S950 "it sounds crunchy and real". Pitch upwards and you get the reverse, folded content a DAW would have filtered out.

The interpolation the 16-bit machines do use. One detailed owner account says the S1000 and S1100 use eight-point windowed sinc interpolation and the S3000 moved to plain linear, which would make the later machines duller with stronger aliasing. Single source, unconfirmed, so treat it as a hypothesis that fits the reports.

The analogue path, and how you were forced to use it. Converters, op-amps and output stage from 1986 are not transparent. More important is what the noise floor forced on you. A 72dB floor means recording hot, right to the ceiling, often with a compressor in front and sometimes a little over. Hot, dense, slightly clipped source material is most of what people mean when they say 12-bit. That is a workflow artefact, not a chip.

So: the bandwidth limiting, the filter slope, the clock-based transposition and the output stage are real. "Twelve bits is warmer" is nostalgia. Buy a 12-bit machine for those four. If you only want something lo-fi to play with, a Casio RZ-1 does its own version for a fraction of the price.

Floppies, SCSI and getting samples in, in 2026

This is the part that stops sales. Every one of these loads its OS or its sounds from a 3.5in floppy, the drives are thirty years old, and blank HD disks are now a specialist purchase. All of it is solved.

Akai S950 Gotek install, and the settings that matter

A Gotek is a USB-stick floppy emulator that fits the same bay and cable. Stock firmware will not talk to an Akai properly; you reflash with FlashFloppy or HxC. The settings differ per model, which is why people conclude it is broken:

  • S950: jumper at S0, interface = akai-s950, host = akai.
  • S900 and S01: S0, interface = shugart.
  • S1000 and S1100: S0, interface = shugart, and critically pin02 = nc.
  • S2000 and S3000XL: S0, interface = ibmpc-hdout.
  • S2800 and S3000: S0, interface = ibmpc-hdout, pin34 = dens.

That pin02 line has a story. An owner documenting an S1100 install hit intermittent read and write errors, measured 3.68V where 5V belonged, and built a level shifter that did not help. Reading the FlashFloppy source showed pin 2 configured as an output: sampler and Gotek were both driving the same wire. Setting it to no-connection fixed it. Images are 1.6MB, 1,638,400 bytes exactly. On the S950 you reach the drive through the memory hatch underneath without removing the cover, and it cannot format a blank image itself, so put blanks on the stick from a computer first. HxC firmware works too and is the only route for the oldest machines, because it emulates Quick Disk: the S612 with MD280, the S700, the X7000.

SCSI: SCSI2SD, ZuluSCSI and BlueSCSI

Anything from the S1100 up has SCSI, and an SD card emulator beats a mechanical drive. ZuluSCSI, BlueSCSI and SCSI2SD all work the same way: build images on a computer with something like Translator, drop them on a FAT32 card as HD0.img through HD7.img, format the volumes from the sampler's own menu and copy the OS across so it boots from SCSI.

Two things bite people. Power: many samplers put none on the SCSI connector, so the emulator needs its own USB supply, powered and settled before the sampler boots or you get "no SCSI device connected". And termination: a SCSI chain terminates at both ends and exactly once at each. The S5000 has a switch for it. Most "my SCSI is flaky" posts are a termination problem.

Memory

The S950 needs EXM006 boards or the third-party equivalents still being made; two get you to 2.25MB and that is the ceiling. The S1000 takes EXM002 or EXM008 cards in four slots, sizes mixable. From the S2000 and S3000 generation it is ordinary 72-pin SIMMs. Buy an older machine already expanded: the boards now cost more than the price difference.

Faults, and what to check before you hand over money

  • Power supplies, S1000 and S1100. The serious one. A failing S1000 supply does not simply stop; owners report units damaged by supplies that went out of tolerance rather than off. A Meanwell RS35-5 is the usual replacement, and on an unserviced S1000 you budget for that first.
  • Display, S1000 and S1100. The LCD is lit by an electroluminescent panel driven by an inverter. The panel fades with hours and the inverter whines as it ages, which is a nuisance in an audio machine. An LED retrofit behind the same panel is the common fix.
  • Floppy drives, all of them. Head wear and dried grease. Replacing like for like is not simple: one S950 owner found the jumpers on two apparently identical drives completely different, and got nothing but "BAD DATA FROM DISK" until he sorted it. Fit a Gotek instead.
  • Battery-backed memory. The MPC60 keeps sequences alive on a CR2032. Coin cells from 1988 are as reliable as that sounds, and a dead one means your work goes at power-down.
  • Pads, MPC family. Dead or uneven pads are usually the sensor sheets, oxidised contacts on the pad cable header, or dry joints on the pad PCB multipin connectors. Sensors are still made. Test every pad softly: hard hits hide a failing sensor.

The checklist, in order: does it boot without a disk error; is the display readable in daylight; run a tone through every individual output, not the stereo pair, because the individual jacks are where dry joints live; play a sample two octaves down and listen to the converter rather than the room; wiggle the SCSI and MIDI connectors while it plays; and put an ear near the transformer for hum or inverter whistle. What that costs to put right is in our piece on recapping and servicing; if you are importing one, the duty and voltage side is in importing vintage studio gear into Africa.

The best Akai sampler for hip hop, and for everything else

For hip hop and jungle, the S950, and it is not close. The bandwidth limiting, the steep filter and the clock-based transposition are the sound, and the 750KB ceiling is a feature because it forces short, decisive chops. The MPC60 is the same era with a sequencer and pads attached, which makes it an instrument rather than a module. For 12-bit that does not sound like an Akai, the Roland S-330 and the S-550 stop at 30kHz and are thicker with it, and the Ensoniq EPS 16 Plus is 16-bit with onboard effects and a grain of its own. Choosing between them is the same exercise as the Juno-60 against the Juno-106: buy the one whose limitations you want.

For multisamples and anything where the sampler should disappear, the S1000. For that plus SCSI, digital out and timecode, the S1100. For a resonant filter and 32 voices, the S3200. For WAV files off a laptop, the S5000. For a modern production centre with the old engine underneath, the MPC4000, 24-bit and good for 512MB.

What we could not verify

  • S900 memory. Wikipedia gives 896KiB total with 768KiB usable; Vintage Synth Explorer gives 750KB. We used the lower figure because it matches the S950's published standard memory. They do not reconcile.
  • Two dates. The S3200 is 1992 on Reverb and 1996 on Vintage Synth Explorer, which we think is the XL revision. The S5000 was announced in 1998 and reviewed by Sound On Sound in January 1999, and sources split between the two.
  • S1000 and S1100 interpolation, and the S1100's converters. The windowed sinc claim, the S3000's switch to linear and the 20-bit converter figure all come from one owner write-up. Consistent with the machines' behaviour, unconfirmed.
  • MPC60 bit depth. A recurring forum claim says it samples at 16 bits and stores at 12. No Roger Linn or Akai statement either way, so we have not repeated it as fact.
  • S900 and S950 filter hardware. Owner consensus is that the board is identical and the S950 adds envelope control in software. No service documentation found.

Everything in that table is on the shelf here in Ballito except the S900 and the S2000, listed because you will meet them. The S950, the S1000, the S1100 and its EX, the S3200, the CD3000XL and the S612 with its MD280 are here, tested through every individual output rather than the stereo pair, along with the PEQ6, built for this exact era and giving you the recallable tone control the S950's filter never had. They are all 19 inch, so if you end up with three of them the housing question answers itself: a second-hand server cabinet racks the lot for about what a 12U studio rack costs. Ask which suits the job before buying the one with the best reputation. The answer is often the cheaper machine.

Sources

Tags: Akai Buying Guide Drum Machines Maintenance Roland Samplers
Back to blog

Leave a comment

Please note, comments need to be approved before they are published.