View source for Power and cabling warnings

This page lists every community report of hardware that burned, melted, blew, shorted, shut down or arrived damaged, with the exact failure as reported. How to power and cool the card properly is on [[Cooling and power]]; load-related crashes without damage are on [[Instability under load]].

== What the card's power input is ==
* The card takes one 8-pin '''EPS12V''' connector: four 12 V and four ground pins, the standard CPU pinout. It is not a PCIe 8-pin.<br />{{Source|handle=wakeupmbrc1|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1558163640446619709}}
* Connector ratings quoted in the thread: PCIe 6-pin 65 W, PCIe 8-pin 150 W, EPS about 330 W; another member gives PCIe 8-pin as 175 W and the CPU 8-pin as 300 W rated, 400 W peak. A PCIe 8-pin also has its grounds on the opposite side from an EPS 8-pin.<br />{{Source|handle=endsjeremy|date=2026-10-11|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1558695475685752835}}<br />{{Source|handle=civiloid|date=2026-10-17|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1561156518555029505}}
* Two safe ways to feed it: a direct EPS (CPU) cable from the PSU, or an adapter from '''two separate''' PCIe 8-pin cables to one EPS.<br />{{Source|handle=c133|date=2026-10-19|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1561551536994132139}}<br />{{Source|handle=jjb01957|date=2026-10-19|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1561553832176255036}}
* Direct EPS cables work, but the 4+4 CPU latch is double width and does not clip onto the card; people trim it with a band saw or a Dremel.<br />{{Source|handle=analyticeth|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1558175554165940368}}<br />{{Source|handle=ropuls|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1558183124129808405}}<br />{{Source|handle=civiloid|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1558217306167844906}}<br />{{Source|handle=c133|date=2026-10-19|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1561564052306526299}}
* '''Check the PSU-side pinout before first power.''' Modular PSU cables are not standard between brands, and the wrong pinout will likely kill a 170HX.<br />{{Source|handle=c133|date=2026-10-19|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1561552472714117223}}
* Size PSUs and cables for transients, not for the power cap: spikes to 380 W were seen on a card capped at 250 W.<br />{{Source|handle=n2dcrypt|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556771516803321877}}

== Cables and adapters ==
=== PSU burned: one PCIe cable per card ===
{{Status|Warning}} A DeepCool PX1000P burned after about 48 hours of mining at only about 220 W per card. The owner had fed each card from a single PCIe 8-pin cable; he had not known that one PCIe cable carries about 150 W and that each card needs two. Another member's reply: a good PSU protected the card, a bad one burns the card. The PX1000P itself is rated well on SPL's PSU tier list, so the cabling, not the unit, was the cause. The owner moved to EVGA 1300 W and 1600 W units with their own breakers.

{{Source|handle=con3po7016|date=2026-10-11|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1558659981165854730}}<br />
{{Source|handle=globalapathy|date=2026-10-11|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1558667430551625728}}<br />
{{Source|handle=endsjeremy|date=2026-10-11|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1558695475685752835}}

=== Never feed the 2x PCIe-to-EPS adapter from one cable and its pigtail ===
{{Status|Warning}} Each PCIe input of the adapter needs its own cable back to the PSU. Do not use the 6+2 pigtail on a cable that already feeds something, and do not plug one cable in twice. The owner who burned the PSU above: "DONT DO THAT, I MADE MAGIC SMOKE LIKE THAT". A 3090 fed from one cable plus its pigtail made the PSU shut down under load. Others have shown it is unsafe regardless of PSU wattage. One owner had run a card from a Corsair RM1000e this way for a while without trouble; the answer was that crappy wires, bad contacts and high power limits combine, and that separate cables per card are not optional.

{{Source|handle=xeax|date=2026-10-18|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1561334034926280838}}<br />
{{Source|handle=endsjeremy|date=2026-10-18|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1561361091450372137}}<br />
{{Source|handle=c133|date=2026-10-19|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1561551536994132139}}<br />
{{Source|handle=con3po7016|date=2026-10-19|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1561553067290136629}}<br />
{{Source|handle=froggao|date=2026-10-19|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1561578260234768486}}<br />
{{Source|handle=wakeupmbrc1|date=2026-10-19|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1561578277397598312}}

=== 6-pin to 8-pin adapter melted at 170 W ===
{{Status|Warning}} One of six 6-pin to 8-pin adapters melted at 170 W load within weeks. The owner runs HP server PSUs with hard-wired PCIe 6+2 leads, two 6+2 per card into an EPS adapter. A 6-pin is rated 65 W; 6+0 and 6+2 plugs carry the same three 12 V terminals, while EPS12V has four. Use two full 8-pin leads per card or build proper EPS cables.

{{Source|handle=neapopa|date=2026-10-17|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1561122626586415135}}<br />
{{Source|handle=civiloid|date=2026-10-17|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1561156518555029505}}<br />
{{Source|handle=wakeupmbrc1|date=2026-10-17|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1561161829311324191}}

=== 12VHPWR connectors (not a CMP report) ===
{{Status|Warning}} An RTX 5090 run at no more than 70% power on a 1600 W Seasonic Noctua Prime needed its 12VHPWR cable reseated twice when it was close to burning a wire; another member had one +12 V terminal of his 12VHPWR connector turn yellow. Relevant if you adapt 12VHPWR leads for a multi-card rig.

{{Source|handle=endsjeremy|date=2026-10-17|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1560810497115426928}}<br />
{{Source|handle=wakeupmbrc1|date=2026-10-17|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1560810856726401025}}

== PSUs and power distribution ==
=== Refurbished server PSU blew up ===
{{Status|Warning}} Three refurbished HP 1200 W PSUs bought on eBay fed 10 GPUs capped at 160 W. Peak GPU draw was almost 2000 W, about 55% of rated capacity. After a 30-minute thermal bench on all cards and then GLM 5.3 tasks with very long prefill, one PSU blew with a loud bang after about two hours. Its fans had never spun up much and the units were very hot when pulled. The GPUs and the rest of the system survived. Caps in those PSUs are known to go bang; a CRPS tester lets you test used server PSUs away from your GPUs.

{{Source|handle=reala7096|date=2026-10-20|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1562183774475919512}}<br />
{{Source|handle=sojun80|date=2026-10-20|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1562205111365804134}}<br />
{{Source|handle=endsjeremy|date=2026-10-20|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1562220361901281392}}

=== PSU thermal shutdown ===
{{Status|Warning}} A Lian Li 1200G ran very hot at about 600 W and shut down after a few hours; its fan spins very slowly (a quiet curve). A USB fan aimed at the PSU grille keeps it stable. Another member says that is not normal and suspects a bad fan.

{{Source|handle=phoenixone7|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556813394701713419}}<br />
{{Source|handle=squishysquare|date=2026-10-06|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556822393408856136}}

=== Worn breakout boards ===
{{Status|Warning}} Constant reboots and PSU shutdowns under load. The owner bought six PSUs believing they were failing; the cause was ZSX-AMPS server-PSU breakout boards that went faulty after 3-4 years. He replaced them with X11-AMPS boards.

{{Source|handle=finishtilt14|date=2026-10-15|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1560224891343736936}}

=== Second-hand ATX PSUs ===
{{Status|Warning}} A member who sold hundreds of PSUs in the mining era says 9 of 10 second-hand ATX PSUs from eBay are degraded and may only manage about 50% of nominal continuously. His setup: a small new quality PSU for the motherboard, decommissioned server PSUs for the GPUs.

{{Source|handle=zorg33|date=2026-10-06|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556939640894918788}}

=== Whole rig losing power on big prompts ===
{{Status|Warning}} Four cards on a PLX board fed by two 750 W PSUs lost power together under heavy prefill; nine cards on three 750 W PSUs dropped together about once a day. Details in [[Instability under load]].

{{Source|handle=replacedriver47|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556769048279126047}}<br />
{{Source|handle=reala7096|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556769461439168573}}

=== PLX switch board shorting on CPU power ===
{{Status|Warning}} A replacement PLX board looked fine but shorted as soon as its CPU power connectors were plugged in, so the PSU would not start (protection tripped; the PSU was not damaged). It was the second defective board from the same seller, who had all positive reviews and good communication.

{{Source|handle=lubakmanqk|date=2026-10-22|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1562912202896179281}}

A Dell R740 system board is suspected of not powering one of its GPU power plugs; no fix posted. {{Status|Reported}}<br />{{Source|handle=chonsworth|date=2026-10-11|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1558775876743856149}}

Open question: what happens to the cards when the PSU feeding an external expansion board is cut while the host stays up, and whether they come back stable when it is powered again. Nobody has answered. {{Status|Theory}}<br />{{Source|handle=fnerdman|date=2026-10-26|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1564244395887435847}}

== VBIOS and power limit ==
=== The 300 W VBIOS has burned a 170HX ===
{{Status|Warning}} Asked about running the 300 W BIOS, a member pointed to a picture of a burned 170HX already posted in the Discord and advised against it; from memory, the PCB is missing power-delivery parts compared with the full board layout. The 300 W power-limit change sits outside the signed section of the VBIOS. The 300 W VBIOS exists for the 8 GB card only. Note that 8 GB cards ship with either a 250 W VBIOS (92.00.67.00.01) or a 300 W VBIOS (92.00.6D.00.0A) on the same board part; see [[Which card do I have]]. Cross-flashing a 250 W card to the 300 W VBIOS is a same-board, same-ID operation: it raises the HBM clock (1458 to 1728 MHz), the SM ceiling and the power wall, but adds no memory and no SMs; one owner measured +8% single-user decode. If you do it, do it on bare metal, one card at a time, with a stock dump saved and an SPI clip at hand: only a full power-off resets the card reliably, and a bad image needs the clip.

{{Source|handle=zagitta|date=2026-10-12|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1559239172202565735}}<br />
{{Source|handle=squishysquare|date=2026-10-18|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1561332714840260660}}<br />
{{Source|handle=xanthefin|date=2026-10-18|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1561336153515364444}}<br />
{{Source|handle=c133|date=2026-10-20|link=https://discord.com/channels/1518634379604136178/1528826330454429706/1562161732137386127|repo=https://github.com/PixelML/club-170hx}}<br />
{{Source|handle=darkmaniac7|date=2026-10-25|note=lab notes}}<br />
{{Source|repo=https://github.com/Schaka/170hx-journey}}

=== Going past the stock power design ===
{{Status|Warning}} For full-SM 10 GB cards that power-throttle, a shunt mod was suggested, preferably after soldering in the unpopulated power phases (DrMOS, chokes, capacitors). Nobody has reported doing it. A 600 W-class card would need the rear power connector replaced with XT60 or similar. A "high power mode" register found on 8 GB cards is deliberately unreleased because nobody knows what it does.

{{Source|handle=wakeupmbrc1|date=2026-10-16|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1560717138744971326}}<br />
{{Source|handle=globalapathy|date=2026-10-20|link=https://discord.com/channels/1518634379604136178/1518695328008638516/1562173726177366188}}<br />
{{Source|handle=matstatman|date=2026-10-20|link=https://discord.com/channels/1518634379604136178/1518695328008638516/1562174089920258139}}

== Cooling-related damage ==
=== No fan, no card ===
{{Status|Warning}} The card is passive. Without airflow it drops off within minutes even at idle (85 C at about 40 W idle; off the bus at 95 C). If a blower fails, "either it will burn down or cmp will die".

{{Source|handle=analyticeth|date=2026-10-07|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1557491900934791409}}<br />
{{Source|handle=nikhilr1696|date=2026-10-07|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1557491750145232997}}<br />
{{Source|handle=programmerboi|date=2026-10-16|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1560492431768813611}}

=== Water-block standoffs shorting a board (Radeon example) ===
{{Status|Warning}} A Radeon (not a 170HX) with exploded copper was posted. Diagnosis: water-block standoffs without dampers, poorly machined, cut through the solder mask and shorted the copper layer; with a PSU that does not trip fast, the spot reaches 1500 C or more in tens of milliseconds, the copper evaporates and the fibreglass burns. Many vendors leave out $2 fuses. Check your block's standoffs before building.

{{Source|handle=froggao|date=2026-10-16|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1560792288031015004}}<br />
{{Source|handle=wakeupmbrc1|date=2026-10-16|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1560794717560512592}}<br />
{{Source|handle=endsjeremy|date=2026-10-16|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1560796797780099093}}

=== EPS retention plate bending the PCB ===
{{Status|Warning}} A water block's EPS plug cover clamps the power cable, but the cable was too rigid for the opening and bent the PCB slightly. Do not use the retention plate; any PCB flex is too much.

{{Source|handle=gnasherrr|date=2026-10-25|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1563738173782360166}}<br />
{{Source|handle=_syn_thetic_|date=2026-10-25|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1563740001664110813}}

=== Aluminium radiator in an AIO kit ===
{{Status|Warning}} The AIO version of a dual-card water-block kit came with an aluminium radiator while the block is copper; the listing describes the block but not the radiator metal. Do not mix it into a copper or brass loop.

{{Source|handle=_syn_thetic_|date=2026-10-25|link=https://discord.com/channels/1518634379604136178/1545982693903306813/1563733230820130839}}<br />
{{Source|handle=globalapathy|date=2026-10-26|link=https://discord.com/channels/1518634379604136178/1545982693903306813/1564381243691507803}}

=== Liquid metal ===
{{Status|Warning}} It gets everywhere, needs careful management and still dries up; it has been seen shorting components.

{{Source|handle=globalapathy|date=2026-10-11|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1558676132033855549}}<br />
{{Source|handle=gpmickey|date=2026-10-11|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1558676651678629999}}

=== Delidding ===
{{Status|Warning}} The die-to-IHS interface is paste (apparently a phase-change material), not solder, and accidental delids have happened. The stock cooler will not mount properly without the IHS: if you cannot compensate for the lid's thickness you will cook the die, and the stock heatsink cannot be tightened further without damaging the VRM.

{{Source|handle=gnasherrr|date=2026-10-18|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1561442200003416125}}<br />
{{Source|handle=skirtchaser|date=2026-10-18|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1561497441117933568}}<br />
{{Source|handle=civiloid|date=2026-10-18|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1561522776815439892}}

=== Water-block pillars on bare footprints ===
{{Status|Warning}} The card is a depopulated A100 board with far more empty component footprints than a real A100. Every unpopulated footprint within reach of a water block's contact pillars must be covered with thermal pad, or the metal pillars short across the exposed copper and kill the card.

{{Source|handle=darkmaniac7|date=2026-07-30|note=lab notes}}<br />
{{Source|handle=markshowa|date=2026-10-18|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1561324784598384784|repo=https://github.com/amoghmunikote/170th-Street}}

=== BMC fan mode after a power loss ===
{{Status|Warning}} On a Supermicro GPU chassis, "Full" fan mode with a manual duty survived a power loss as Full mode but lost the duty, so every boot ran the fan wall at 100% until the OS rewrote it a few minutes later; with the power-restore policy at "previous", the box also started by itself when AC came back, at night. Fix: leave the BMC in its automatic mode at rest, hold Full plus a manual duty only while the OS runs, release it in a shutdown hook, and set power restore to always-off.

{{Source|handle=darkmaniac7|date=2026-10-27|note=lab notes}}

Host drive overheating under model loads (an NVMe that dropped off the bus): see [[Instability under load#Host faults that look like card faults|Instability under load]].

== Physical damage and condition on arrival ==
=== Shipping damage ===
{{Status|Warning}} A card with proof of life and a working unlock before shipping arrived in a box with a hole, bracket bent about 30 degrees, and was not detected on PCIe at all.

{{Source|handle=globalapathy|date=2026-10-18|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1561380398091735112}}

=== Damaged PCIe edge contacts ===
{{Status|Warning}} A card from Alifond arrived with damaged edge contacts; it still passed stress tests after repeated insertion. Long-term effect unknown.

{{Source|handle=moliminous|date=2026-10-10|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1558528724582793281}}

=== Corrosion on ex-farm cards ===
{{Status|Warning}} A card bought from Alifond for about $2.5k arrived with a bent bracket, visible aluminium oxidation and a bracket screw rusted off; it works, but the owner believes it ran under water-curtain cooling. External corrosion means the board itself is probably rotting too. Three years of continuous running in uncontrolled humidity makes cards look like this.

{{Source|handle=kellertreppe123|date=2026-10-19|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1561782970262757537}}<br />
{{Source|handle=wakeupmbrc1|date=2026-10-19|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1561814886223450345}}<br />
{{Source|handle=endsjeremy|date=2026-10-19|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1561843312338407495}}

=== Missing thermal pads ===
{{Status|Warning}} Cards from Alifond arrived filthy, with thermal pads missing on the back of the PCB and on the side components under the main heatsink (the main die pads were there). The seller does the capacitor mod, so the cards are opened anyway, but they are not cleaned, repadded or repasted. Another buyer called the quality "horrendous". Clean, repad and repaste before use.

{{Source|handle=_syn_thetic_|date=2026-10-25|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1563953029353443489}}<br />
{{Source|handle=kellertreppe123|date=2026-10-25|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1563958215052632115}}

=== A capacitor fell off ===
{{Status|Warning}} (a 100 uF 16 V polymer part, one prong destroyed) while cleaning a very dirty card; the card still worked without it. Replacement parts are on [[Recovery]].

{{Source|handle=lia_oc|date=2026-10-21|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1562506436624257196}}

=== Insects ===
{{Status|Warning}} Ants got into a card through an unsealed window gap. Do not try to blow them out with fans or heat; take the card apart and clean it.

{{Source|handle=con3po7016|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1518691940646781070/1558152316488515625}}<br />
{{Source|handle=wakeupmbrc1|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1518691940646781070/1558152462303756471}}

== Soldering and flashing ==
=== The capacitor (x16) mod ===
{{Status|Warning}} is hand soldering of 24 tiny 0402 capacitors on a dense multi-layer board, and the one step in the usual unlock with real hardware risk. A failed attempt that damages nothing else is easy for a technician to fix, and the card still works at x4 meanwhile. Use flux and a clean tip, and tack one side of every cap before doing the other side. Link problems after the mod are on [[Card not detected or BAR allocation fails]].

{{Source|handle=rrickk420|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1558127650223030302|repo=https://github.com/Consensus-Protocol/cmp170hx}}<br />
{{Source|handle=_orrorin|date=2026-10-13|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1559584745367806053}}<br />
{{Source|handle=shirogeek|date=2026-10-13|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1559648337299382383}}

=== SPI flash programmer voltage ===
{{Status|Warning}} Flashing a different VBIOS means opening the card and programming the flash chip directly. Flash chips are 1.8 V or 3 V, and a 3 V programmer will fry a 1.8 V chip. Some CH341A programmers sold as 3.3 V actually put out 5 V; a CH347-based programmer selects the voltage in software. Level-shifter boards (including 1.8 V) exist. One member has fried a few chips. The VBIOS is signed, so a modified image is a separate problem; see [[Driver and unlock failures]].

{{Source|handle=wakeupmbrc1|date=2026-10-13|link=https://discord.com/channels/1518634379604136178/1518695328008638516/1559582893200904244}}<br />
{{Source|handle=_orrorin|date=2026-10-13|link=https://discord.com/channels/1518634379604136178/1518695328008638516/1559588668623163453}}<br />
{{Source|handle=ichundes|date=2026-10-13|link=https://discord.com/channels/1518634379604136178/1518695328008638516/1559694344384938076}}<br />
{{Source|handle=globalapathy|date=2026-10-14|link=https://discord.com/channels/1518634379604136178/1518695328008638516/1559750969539956807}}

== Related pages ==
* [[Cooling and power]]
* [[Instability under load]]
* [[Recovery]]
* [[Known bad combinations]]
* [[Platforms]]

{{ImportFooter}}
[[Category:Troubleshooting]]
[[Category:8 GB]]
[[Category:10 GB]]