Disassembly and cleaning
More actions
Ex-mining CMP 170HX cards usually arrive dirty inside, with stock paste gone hard and thermal pads sometimes missing, even when the shroud and the outer fins look clean. This page covers protecting yourself, what to expect inside, opening the card, getting the stock cooler off without breaking parts, and cleaning it. Paste, pads and phase-change material are on Thermal interface: paste, pads and phase-change, water blocks on Water cooling, and the x16 capacitor mod, which also needs the card open, on PCIe x16 capacitor mod. Testing a newly arrived card is on Acceptance checklist.
Before you start: protect yourself
Wear gloves: sharp edges and dirty boards
Warning On 2026-08-07 @robbbbbb7973 cut his finger cleaning a card and the dirt on the board infected the cut. @_arcanum_, who has opened many cards, says gloves are a must, especially when getting the board out of the backplate, and called the retaining wings on the EPS connector a badly designed part of that step. @zapirate and @impactic urged seeing a doctor about an infected cut. Wear gloves for the whole teardown.
Source: @robbbbbb7973, 2026-08-07
Source: @_arcanum_, 2026-08-07
Source: @zapirate, 2026-08-07
Source: @impactic, 2026-08-07
Dust from mining warehouses: wear a mask and do not breathe it
Warning @machinezer0 (2026-08-29) fell ill after blowing dust out of cards bought from a US seller, while nobody else in the house got sick; he suspected mould spores and signed off "Buyer beware". @humandrone8721_60165 keeps a turbo blower and FFP2 masks for cleaning. On 2026-09-24 @djpandaman found brown dust on two of his cards from China, and @spectakkular warned that dust on cards bought from China may contain asbestos, saying asbestos was only banned in new Chinese warehouses in 2011. Nobody has tested the dust. Clean with care, wear a mask, and do not blow the dust out indoors where you will breathe it.
Source: @machinezer0, 2026-08-29
Source: @humandrone8721_60165, 2026-08-29
Source: @meatsus, 2026-08-30
Source: @djpandaman, 2026-09-24
Source: @spectakkular, 2026-09-24
What you will find inside
Open every used card on arrival, even if it looks clean outside
After about a dozen cards from Chinese mines, @_arcanum_ (2026-07-31) recommended opening every card even if the shroud and outer fins look fine: many were heavily dusty inside, with no corrosion seen so far. @danske1 found caked dust falling out in clumps, and @rrickk420 found his card dirty and planned a full repaste. In October @civiloid's cards looked clean outside but were extremely dirty inside with pads falling apart; @_syn_thetic_ found many thermal pads absent and the worst paste he has seen. @rogersmith2026 and @spale777 call cleaning, repadding and repasting a must, and @armand0e suspected his poor temperatures came from skipping it. Plan to clean, repad and repaste every card you open.
Proven 8 GB and 10 GB
Source: @_arcanum_, 2026-07-31
Source: @danske1, 2026-07-31
Source: @rrickk420, 2026-07-31
Source: @scallion_9883, 2026-07-31
Source: @rogersmith2026, 2026-08-12
Source: @spale777, 2026-08-21
Source: @armand0e, 2026-08-21
Source: @civiloid, 2026-10-11
Source: @_syn_thetic_, 2026-10-22
Rust, brick dust, cement and mould on arrival
@machinezer0 (2026-08-18) received cards with fully rusted screws, some smelling of black mould, and @segmond's cards showed rust and smelled of burnt electronics. @icych had to clean cement off cards from a mining rig, and @_arcanum_'s cards from a Chinese mine had brick dust inside and a gritty shroud ("basically part brick"). @bayley_w thinks the rust probably came from storage after the cards were e-wasted, or from condensation where miners cool with undried outside air; @ju5t5omedude posted photos of farms cooled with evaporative water curtains like those on chicken farms, which matches the corrosion on @cmdrwaka324's brackets. @smit1237 has seen mining rigs half-flooded, which explains the rusty brackets on his cards; his 100 cards from a Chinese farm near a hydroelectric plant had heavily rusted metal parts, yet only one failed. @meatsus said prior mining use is not a concern in itself, and @lycorascence added that this holds if the cards were cooled and not thermally cycled. On 2026-09-02 @_arcanum_ said some of his cards arrived in such poor condition they might as well have served a decade in a mine, and @kellertreppe123 joked that his card came as a "sahara dust ... fine sanded cooling fan edition". Whether mining shortened the cards' life is on Card failure rates and longevity.
Proven
Source: @cmdrwaka324, 2026-08-17
Source: @ju5t5omedude, 2026-08-17
Source: @scallion_9883, 2026-08-17
Source: @machinezer0, 2026-08-18
Source: @segmond, 2026-08-18
Source: @bayley_w, 2026-08-18
Source: @ju5t5omedude, 2026-08-18
Source: @icych, 2026-08-18
Source: @_arcanum_, 2026-08-18
Source: @meatsus, 2026-08-20
Source: @lycorascence, 2026-08-20
Source: @smit1237, 2026-08-20
Source: @smit1237, 2026-08-26
Source: @_arcanum_, 2026-09-02
Source: @meromgreen, 2026-09-02
Source: @meatsus, 2026-09-02
Source: @artistosu, 2026-09-02
Source: @stcord, 2026-09-02
Source: @kellertreppe123, 2026-09-02
Factory pads are good; odd-looking factory soldering is usually harmless
@smit1237, who has serviced many mining cards, says NVIDIA did not cheap out on the CMP 170HX thermal pads, unlike ASUS 3090s where 100 of 100 serviced cards had leaking pads after a year (2026-08-26). @mazzanfar7661 opened a new 8 GB card to add the x16 caps and found two capacitors bridged together, a second leaning with too much solder, and flux-like residue. @skirtchaser identified them as power-rail filter caps (ground on one side, the rail on the other), probably fine; @derick2k said a bridge does not matter if both pads are ground; @skirtchaser noted that oil from thermal pads can look like flux; @_arcanum_ posted a factory-looking reference photo, and on 2026-09-29 the card tested fine. @zagitta found a large blob of epoxy or similar on one capacitor of a newly bought card with the warranty sticker still over a screw (presumably factory); the card works and @meatsus said it is no concern. @neapopa found a card with a glued connector (which one is not stated) that came off with acetone.
Proven
Source: @smit1237, 2026-08-26
Source: @neapopa, 2026-09-18
Source: @mazzanfar7661, 2026-09-24
Source: @skirtchaser, 2026-09-24
Source: @derick2k, 2026-09-24
Source: @xtj7, 2026-09-24
Source: @djpandaman, 2026-09-24
Source: @_arcanum_, 2026-09-24
Source: @zagitta, 2026-09-30
Source: @meatsus, 2026-09-30
Signs of prior servicing: seals, swapped brackets, missing screws
@cmdrwaka324 (2026-08-17) received 5 clearly well-used cards from China with tarnished PCIe brackets; only one of the 5 brackets had engraved text, and @scallion_9883 suggested the brackets were replaced or the engraving polished off. All of @cmdrwaka324's cards carried a red and white seal with Chinese text over a backplate screw. @_arcanum_ thinks such cards were serviced before sale: brackets replaced, opened for dusting, maybe repasted, and some missing the screw on the EPS power connector retention bracket. On every card he has opened, thermal pads sat only on the FETs and the controller. @quark2world's card was missing some thermal pads and had parts fitted in the wrong places, and @danske1 found screws missing on the back plate bracket. @cressida1881 (2026-08-31) got one card with an NVIDIA-marked bracket and one with a blank bracket (the card itself was fine); @ins_mniac guessed the bracket was swapped because of rust, and @bagstah had seen two such cards, not rusty, one with a "not for resale" sticker. Expect a missing screw or two, and check that every pad is in place before you close the card.
Proven
Source: @danske1, 2026-07-31
Source: @cmdrwaka324, 2026-08-17
Source: @ju5t5omedude, 2026-08-17
Source: @scallion_9883, 2026-08-17
Source: @_arcanum_, 2026-08-17
Source: @cmdrwaka324, 2026-08-17
Source: @quark2world, 2026-08-17
Source: @cressida1881, 2026-08-31
Source: @ins_mniac, 2026-08-31
Source: @bagstah, 2026-08-31
Opening the card
Front shield first, 14 screws, one may hide under a sticker
The shroud does not split until the front shield is off (@cute_sc2). @ju5t5omedude counts 14 screws and notes that the bracket wedges into the case, so it has to come off; @_arcanum_ describes 10 Torx T8 screws plus 4 on the bracket. @cmdrwaka324's sequence:
- Remove the screws on the back that hold the cover. One may be hidden under a QC or seller sticker.
- Remove the 4 screws holding the PCIe bracket.
- With the plate off, remove the single screw on the small black bracket at the power connector.
@johnnydaboy could not lift the PCB after removing 14 outer screws and the 4 backplate-bracket screws: a screw was still hidden under a seller sticker. @zanooda hit the same problem on 2026-09-14, and @m3gagluk pointed to the sticker.
Proven
Source: @_arcanum_, 2026-07-31
Source: @lubakmanqk, 2026-08-19
Source: @cute_sc2, 2026-08-19
Source: @ju5t5omedude, 2026-08-19
Source: @cmdrwaka324, 2026-08-19
Source: @johnnydaboy, 2026-08-31
Source: @zanooda, 2026-09-14
Source: @m3gagluk, 2026-09-14
Lifting the PCB past the locating pin by the PCIe edge
The PCB sits in a milled slot in the aluminium housing and is held by an aluminium pin (a dimple in the frame) next to the PCIe edge on the IO-bracket side (@danske1, @cmdrwaka324):
- Unscrew the small black bracket on the power connector.
- Lift the EPS connector's wires to tilt the board, and push the connector through the shroud.
- Lift and tilt the PCIe edge out of the housing slot. The metal lid has to come out at a slight angle to clear the hook of the PCIe connector without nicking small parts (@humandrone8721_60165).
Push from outside rather than pulling on cables: @meatsus broke one of his cards right at that spot (it still works). @c133 found getting the EPS connector out harder than the PCIe edge, and @galygious said the NVLink fingers sit just past the shroud edge, which makes lifting the EPS side out harder. A Chinese teardown video shows the step at about 1:30 (youtu.be/N1yhvXj_Do8).
Proven
Source: @meatsus, 2026-08-31
Source: @danske1, 2026-08-31
Source: @humandrone8721_60165, 2026-09-14
Source: @cmdrwaka324, 2026-09-14
Source: @c133, 2026-09-14
Source: @galygious, 2026-09-14
Removing the stock cooler
Twist, never pry: capacitors and inductors break off
Warning The stock paste can glue the cooler to the IHS. @meatsus snapped electrolytic capacitors off one card while removing a stubborn cooler whose paste was "literal loctite", and @_arcanum_ had the same caps come off his card and soldered them back on (2026-09-07). On 2026-09-24 @_arcanum_ knocked an inductor off another card taking the heatsink off, because the paste was basically glued on; he soldered it back and the card was fine so far, and @djpandaman expects some of his heatsinks to feel glued as well. @finishtilt14 (2026-09-16): the cooler is stuck on hard and the board bends easily. Do this instead (@meatsus): hold the card with the PCB over your palm and your fingers on the sides of the cooler, and use grip strength in a twisting, back-and-forth motion, pushing at the corners for leverage and feedback. Never pull or pry the halves apart: the IHS sits flush with the heatsink and small metal cubes leave no gap for a pry tool. @petrikrohn_26497 noted that Linus also twisted the heatsink off rather than pulling (LTT CMP 170HX video, around 9 minutes).
Source: @petrikrohn_26497, 2026-08-12
Source: @meatsus, 2026-09-07
Source: @_arcanum_, 2026-09-07
Source: @ins_mniac, 2026-09-07
Source: @endsjeremy, 2026-09-07
Source: @finishtilt14, 2026-09-16
Source: @marklintonca, 2026-09-16
Source: @meatsus, 2026-09-16
Source: @neapopa, 2026-09-16
Source: @_arcanum_, 2026-09-24
Source: @djpandaman, 2026-09-24
If the heatsink will not move
@periloustech (2026-08-05) worried about ripping the chip off with the heatsink, and @coolgod_72592 (2026-08-09) could not get the heatsink (held by 4 screws) off the die even after heating the card and using isopropyl alcohol. Tips from the thread: check that no screws were missed, and twist in the same plane as the die (@ju5t5omedude); use contact cleaner spray as a solvent, or soak the card in 99% alcohol in a covered flat pan for a few minutes (@ju5t5omedude); alcohol has high surface tension, so squirt 99% IPA into the gap with a needle (@sealclubber). @diabeetee could not twist his off either and pressed up gently on one corner of the heatsink instead. @quark2world suggests warming the heatsink to loosen it; warming the block side with hot air helps with fresh paste but is untested on old dried paste, and an ultrasonic cleaner was suggested but nobody has tried it. @danske1 notes the package has an IHS, so it is safer than a bare-die 3000-series card, but hardened paste can make the heatsink hard to remove.
Reported
Source: @periloustech, 2026-08-05
Source: @coolgod_72592, 2026-08-09
Source: @ju5t5omedude, 2026-08-09
Source: @sealclubber, 2026-08-09
Source: @diabeetee, 2026-08-09
Source: @quark2world, 2026-08-12
Source: @danske1, 2026-08-14
Cleaning
The stock paste is hard like cement
@rogersmith2026 and @danske1 (2026-08-12 and 08-14) found the stock paste solidified so hard it was difficult to remove, and the cards caked in grime. @gnasherrr says the stock paste, especially on cards from China, is rock solid: Clorox wipes do nothing, and he had to dig it out with dental picks, with magnifying goggles helping. @skirtchaser (2026-08-16) found 99.9% IPA did not remove it, and @meromgreen could not fully clean the pads with IPA. Which solvent works is contested; see the next entry. @smit1237 found repasting tedious, with many pads to cut.
Proven
Source: @rogersmith2026, 2026-08-12
Source: @danske1, 2026-08-14
Source: @skirtchaser, 2026-08-16
Source: @meromgreen, 2026-08-16
Source: @smit1237, 2026-08-22
Source: @gnasherrr, 2026-10-22
Cleaning agents: brush and air, IPA, contact cleaner or brake cleaner?
Contested
- 2026-07-24: @meatsus advises no alcohol on the PCB (only on the IHS, if anything), a light brush for dust with no cotton or cloth, compressed air only on the radiator and never on the PCB, and no solvent on the radiator because it can dry into a dirt film. A YouTuber cleans cards with WD-40, which @lubakmanqk said deserves jail time. @ju5t5omedude uses contact cleaner spray and a toothbrush. Source: @meatsus, 2026-07-24
Source: @ju5t5omedude, 2026-07-24
Source: @lubakmanqk, 2026-07-24 - 2026-08-03: @_arcanum_ says the cards smell like an unfiltered mine. Wash the shroud with lots of hot water and detergent, clean the PCB with a soft brush and contact cleaner without removing the cooler, use an isopropanol bath when repasting, and do not rinse thermal pads with alcohol. @machinezer0 noted that some miners washed cards with plain water. Source: @_arcanum_, 2026-08-03
Source: @machinezer0, 2026-08-03 - 2026-08-16: @quark2world deep-cleaned his cards by soaking them in electrical contact-cleaner spray (the kind used to degunk car electrics; the first runoff came out grey and brown), mainly to remove brown and grey goo (rust, dirt, old paste and pad residue) beside the IHS, and finished with an air blower. One can (about 4 EUR for cheap ones) does 2 to 3 cards, and 30-litre canisters exist for soaking a whole rig. Source: @quark2world, 2026-08-16
Source: @deejay9983, 2026-08-16
Source: @fallengod9962, 2026-08-16
Source: @danske1, 2026-08-16 - 2026-08-16: @skirtchaser found 99.9% IPA would not remove the stock compound and @meromgreen could not fully clean the pads with IPA; @quark2world sprayed an electronics contact cleaner on the cooler and the whole card, and it dissolved the paste without issue. @roboalchemist. warns that brake cleaner eats plastic and recommends a contact cleaner such as MG Chemicals 409B Electrosolve. Source: @skirtchaser, 2026-08-16
Source: @quark2world, 2026-08-16
Source: @meromgreen, 2026-08-16
Source: @roboalchemist., 2026-08-16 - Same discussion: @meatsus warned that contact cleaner is too aggressive for PC electronics, and @artistosu recommends 99% IPA. @meromgreen notes that wipes are much cheaper and that the oily residue comes from old pads. @skirtchaser had earlier used non-chlorinated brake cleaner as a cheap PCB cleaner but warned that chlorinated versions (common in the US) attack plastics, and @roboalchemist. worried about residues damaging boards over time. Source: @quark2world, 2026-08-16
Source: @meatsus, 2026-08-16
Source: @artistosu, 2026-08-16
Source: @meromgreen, 2026-08-16
Source: @skirtchaser, 2026-08-16
Source: @roboalchemist., 2026-08-16 - 2026-10-11: @squishysquare's cleaning kit is isopropyl alcohol plus a toothbrush. Source: @squishysquare, 2026-10-11
- 2026-10-22: the ArctiClean kit (Amazon B001JYQ9TM) was suggested in the thread, not tried by the poster. Source: @_syn_thetic_, 2026-10-22
Source: @gnasherrr, 2026-10-22
Source: @segmond, 2026-10-22
Washing in a dishwasher or with water
Contested
- 2026-07-24: @dalzyyy ran fans and a heatsink (card not specified) through a dishwasher. The fans came out like new, the heatsink kept its "supergunk", and he baked the parts dry in a 3D printer chamber. @quark2world would let them dry overnight, and @impactic warned about the fan bearings. Source: @dalzyyy, 2026-07-24
Source: @quark2world, 2026-07-24
Source: @impactic, 2026-07-24 - 2026-08-04: @dalzyyy washed dirty, partly rusted dead cards in a dishwasher and let them bake dry. Source: @dalzyyy, 2026-08-04
- @skirtchaser: let them dry for a very long time, because water can creep under the IHS. Source: @skirtchaser, 2026-08-04
- @cute_sc2: washing raises the chance of corrosion problems later, and drying with heat raises it further. Source: @cute_sc2, 2026-08-04
- @quark2world: the IHS is what makes dishwashing dangerous; der8auer recommends electronics cleaning spray afterwards. Source: @quark2world, 2026-08-04
- @zapirate recommends a 99.9% IPA bath after washing and lets boards dry for 48 h. Source: @zapirate, 2026-08-04
- @sealclubber prefers a toothbrush, IPA and a cloth, or an ultrasonic bath with water and dish soap, and avoids water unless it can be fully dried. Source: @sealclubber, 2026-08-04
- @giaki3003 has used a dishwasher for over 10 years with no issues. Source: @giaki3003, 2026-08-04
Repad and repaste before closing
What owners used when closing the card
For a first card @meromgreen (2026-08-10) lists a printed fan duct plus IPA wipes, 2 mm pads and paste; @rogersmith2026 applied 2 mm pads and fresh paste, and @danske1 repasted with MX-6 and 2 mm Gelid GP Extreme pads. @quark2world: apply paste in an X pattern (pressure spreads it), pad the open contacts where parts are unpopulated, and buy a large tube because the IHS is huge. @meatsus advises making sure the chips on the back of the board are cooled. @spale777 prefers phase-change material (PTM), which should last the card's lifetime; alternatives discussed were Thermal Grizzly KryoSheet (@arjankai: no heat activation needed but pricier), Shin-Etsu paste on the large IHS (@_arcanum_, since heat flux is low on this big die), and @dalzyyy warned that paste pumps out. In October Upsiren putty instead of pads and Duronaut paste were suggested so the cards never need reopening, and @gpmickey planned to repad 12 cards while they were open for the cap mod. Materials, pad thicknesses and results are on Thermal interface: paste, pads and phase-change.
Reported
Source: @meromgreen, 2026-08-10
Source: @rogersmith2026, 2026-08-12
Source: @quark2world, 2026-08-12
Source: @danske1, 2026-08-14
Source: @quark2world, 2026-08-17
Source: @spale777, 2026-08-21
Source: @arjankai, 2026-08-21
Source: @dalzyyy, 2026-08-21
Source: @_arcanum_, 2026-08-21
Source: @danske1, 2026-08-22
Source: @gpmickey, 2026-10-11
Source: @_syn_thetic_, 2026-10-22
Source: @segmond, 2026-10-22
A full cleaning and putty/paste guide by @_syn_thetic_ is in progress (pointer only; it will be added here with the author's agreement); see Discord.