HP DPS-800GB Breakout No PS_ON Signal

Identify your PSU off the side label — `HP DPS-800GB`, `DPS-800GB A`, `DPS-800GN`, `HSTNS-PD05`, `379123-001`, `380622-001`, `399771-001`, `403781-001`. Pull the canonical card-edge pinout from HP's service note for that exact PN, or from a reputable breakout vendor's pinout PDF (Parallel Miner publishes per-model PDFs on each product page). Write the pinout on a sticky note. Most 'DPS-800GB won't turn on' tickets resolve at this step alone, because the user assumed the `DPS-1200FB` pinout transferred — it doesn't. Spend `5 minutes` here; save an hour of trial-and-error wiring.

Continuity-test the breakout PCB against the datasheet pinout. PSU unplugged from AC, ohmmeter on continuity, probe each breakout pad to the corresponding card-edge pin. Verify the four critical signal pins on a `DPS-800GB`: `PS_ON` (pin `33`, active-low enable), `PSKILL`/`PRESENT` (pin `36`, chassis-presence latch), `+12V_SENSE` / `+S` (broken out, *not* jumpered to `COM`), `-12V_SENSE` / `-S` (broken out, *not* jumpered to `+12V`). Any disagreement: stop, mark the bad pads with Sharpie, plan rework in Tier 2.

Verify `PS_ON` and `PSKILL` are both at `0.00 V` to `COM` after any breakout jumpers are in place. AC connected, mains switch up, no load attached. DMM common probe on `COM` (pin `34`), test probe on `PS_ON` (`33`): must read `0.00 V`. Same for `PSKILL` (`36`). If either reads above `0.1 V`, the breakout is using a `10 kΩ` resistor pull-down where the supervisor wants a hard wire jumper. Note both readings; they tell you exactly which trace to fix in Tier 2.

Confirm `+5 Vsb` is alive before chasing the main rail. DMM common on `COM`, test probe on the breakout's `+5VSB` pad. Should read `4.95-5.10 V` the moment AC is applied (no `PS_ON` needed — `+5 Vsb` is always-on). If `+5 Vsb` is absent, the PSU primary side is dead and no breakout rework will fix it; set the unit aside. If `+5 Vsb` is present and stable, the rest of the diagnostic is breakout-side.

Apply a known load above the supervisor's minimum threshold before declaring victory. A single Bitaxe Gamma at `~25 W` is below the `DPS-800GB`'s spec'd minimum-load region. Pair two Bitaxes (`~50 W`), wire an S9 hashboard (`~400 W`), or hang a `15 Ω 25 W` ceramic dummy resistor across the bus to burn `~10 W` of margin. Without real load, the supervisor may oscillate and you'll chase a phantom fault.

Bond `PS_ON` (pin `33`) to `COM` (pin `34`) with a hard wire jumper. Open the breakout enclosure. Find the `PS_ON` and `COM` silkscreen labels. Solder a `22 AWG` solid-core wire bridging the two pads. Hard short — not a `10 kΩ` resistor, not a 'near zero' pull-down. The `DPS-800GB` supervisor wants `0.00 V` on `PS_ON`, and any voltage above `~0.1 V` keeps it in standby. After rework, AC-cycle: fan should spin within `1-2 seconds` and `+12V` should read `12.0-12.3 V` no-load.

Bond `PSKILL` / `PRESENT` (pin `36`) to `COM` (pin `34`) with a second hard jumper. This is the chassis-presence latch — the cause of every 'PSU enables for `30 seconds` then dies' report on this PSU family. Solder a second `22 AWG` jumper between `PSKILL` and `COM` on the breakout. Confirm with DMM: must read `0.00 V` to `COM`. This jumper is mandatory; do not skip it because 'the rail came up' — a rail that drops at `~30 seconds` is more dangerous than one that never came up, because mid-hash brown-outs damage hashboards faster than a clean cold-start.

Bond `+12V_SENSE` (`+S`) to `+12V` and `-12V_SENSE` (`-S`) to `COM` at the breakout. If the breakout silkscreen has `+S` and `-S` pads broken out, jumper `+S` to the `+12V` bus and `-S` to the `COM` bus with `22 AWG` wire. Without sense bonded, OVP trips within seconds of enable. Minimum: jumpers at the breakout. Better for cable runs over `30 cm`: run thin (`28 AWG`) sense leads alongside the heavy 12 V cable to the load (Bitaxe barrel-jack pads, S9 hashboard input). Sense pins draw `<1 mA`; gauge is irrelevant for current.

Replace the heavy load cable with appropriate-gauge silicone-jacket wire. The `DPS-800GB` outputs `~83 A` peak at `12 V`. Single Bitaxe (`~3 A`) or paired Bitaxes (`~6 A`): `14 AWG` plenty. Bitaxe Hex on XT30: `12 AWG`. S9 build (`~33 A` per `PCIe 6-pin`, `~100 A` total): `10 AWG` minimum, `8 AWG` for runs over `30 cm`. Crimp connectors with a real ratcheting crimper (Klein, Iwiss, Engineer PA-09 — not pliers). Solder-tinning before crimping is contraindicated; cold-flow loosens the crimp under thermal cycling.

Bench-load the PSU and verify rail under sustained load. With all four jumpers in place, connect the load. DMM on the breakout `+12V` and at the load end. Healthy `DPS-800GB` holds `12.0-12.3 V` from no-load to `~600 W`. Sag below `11.8 V` under load means cable gauge insufficient, sense pins jumpered to bus instead of load, or breakout PCB undersized. Run `30 minutes`, IR-thermometer the breakout PCB / card-edge connector / cable lugs. Above `60 °C` PCB or `45 °C` connector = undersized copper. Step up gauge or replace breakout.

For S9 home builds, parallel two `DPS-800GB` units with `I_share` bonded. A single `DPS-800GB` is rated `1000 W`; an S9 pulls `1300+ W` peak. Two paralleled units handle the S9 cleanly *if* the `I_share` pins are tied between them with a low-impedance lead under `30 cm`. Without the `I_share` bond, one PSU takes the entire load, hits OCP, drops out; the other PSU sees the bus rise and also drops out — the rig dies in milliseconds, hashboards see a brown-out edge. Consult the HP service note for your exact PN — `I_share` pin number varies between revisions.

Suppress the stock fan from running at full RPM. Stock `DPS-800GB` fan logic ramps to `~6500 RPM` (jet-engine loud) the moment the rail enables, because in the original chassis the fan was BMC-driven via PWM. On a breakout the fan-control pin floats, which the supervisor interprets as fail-safe-to-maximum-cooling. Three options: (a) PWM the fan pin from a microcontroller with thermistor feedback, (b) replace the stock fan with a Noctua `NF-A4x20` PWM fan direct off `+12V` (~`$20 CAD`, `~2500 RPM`, room-tolerable), (c) accept the noise. Option (b) is the home-miner default. Confirm thermal margin after mod by IR-thermometering the heatsinks every `30 min` for `2 hours` of sustained load.

Add a `2200 µF` low-ESR electrolytic across the breakout `+12V` and `COM` busses to suppress OVP under transient load. On some `DPS-800GB` revisions, an aggressive OVP comparator trips when load drops from full hash to idle in milliseconds — the bulk cap discharges into a sudden-no-load and the rail overshoots `12.6-13.0 V` for microseconds, latching OVP off. The bulk cap slows the transient. `$3 CAD` of capacitor saves the OVP trip and the brown-out it would otherwise cause downstream.

For DCENT_OS users on Antminer S9 / S9-class builds powered by paralleled `DPS-800GB` units: DCENT_OS exposes per-chip HW% and rail-voltage telemetry on the dashboard, letting you see in real time when one PSU is sagging vs the other. On a multi-PSU rig this is invaluable diagnostic visibility — far better than stock Bitmain firmware which only reports chain-level. The DCENT_OS dashboard `Vin` value mirrors what the hashboard's voltage sensor sees. (DCENT_OS is Antminer-only — does not run on Bitaxe / NerdAxe / NerdMiner / NerdQAxe; for Bitaxe use the AxeOS dashboard `Vin` readout instead.)

Verify wire-termination quality at the breakout terminal block. Crimped ring lugs on heavy `+12V` and `COM` outputs should be ratchet-crimped, not hand-crimped or pliers-crushed. Pull-test after crimp: a properly crimped lug on `10 AWG` should not pull off under `30 lb` of force. Bare tinned-copper wire only — solder-tinning before crimp is contraindicated. Heat-shrink boots on every lug to prevent accidental shorts in the breakout enclosure. Many 'good DPS-800GB' units have been cooked by a single loose `10 AWG` crimp under S9 load.

Stop and consult D-Central if you've burned through a Bitaxe or two trying to get the rail clean, you're planning a paralleled-`DPS-800GB` S9 home build and want the wiring vetted before power-up, or you want a known-good breakout PCB sourced and assembled rather than rolling your own. D-Central's ASIC repair service covers Bitaxe-side damage from rail issues; the shop carries breakouts vetted for the four-jumper layout. Book at https://d-central.tech/services/asic-repair/ — free quote, parts at cost, return shipping included for North American customers.

Decide repair vs. rebuild for the breakout itself. If the cheap breakout has visible PCB browning, lifted pads, solder-joint cratering, or you've reworked it twice and still see flaky behaviour, replace it. Quality `DPS-800GB`-specific breakouts (Parallel Miner, Deep In The Mines, Mining-Heaven, D-Central-stocked) are `$25-45 CAD` and get the four-jumper layout right out of the box. The 'save `$15` on the no-name breakout' math falls apart after the first fried Bitaxe. Document the failure mode for warranty claims if applicable, then upgrade.

Consider the turnkey alternative for new builds. Mean Well `LRS-1000-12` or `RSP-1000-12` is the cleaner alternative: proper screw terminals, no signal-pin gymnastics, certified for continuous duty, `~$120-160 CAD` new. The `DPS-800GB` path is for the budget-conscious operator who enjoys repurposing enterprise hardware and has the bench skills to do the four jumpers correctly. Both paths get you a `1 kW` `12 V` rail; the trade is `$60 CAD` and `2 hours` bench time vs. `$120 CAD` and a `5-minute` unbox-and-wire. Pick the path that matches your time-vs-money trade.