Melting GeForce RTX 4090 Power Cables: A Timeline Of Events

Republished By Plato

Followers: 0

Nvidia’s glorious GeForce RTX 4090 wowed reviewers and set a new bar for just how stupidly fast a graphics card could be. Unfortunately, the launch of the $1,600 GPU has been marred by multiple reports of melting 12VHPWR connectors used in the cards damaging both the connector and the GPUs at times.

The new 12VHPWR connector is a compact power connector that combines the capability of multiple older 6- and 8-pin connectors into one tiny plug. It was originally adopted with the GeForce RTX 3090 Ti Founders Edition and is now used in the GeForce RTX 4090 Founders Edition as well as custom versions of the RTX 4090 built by the likes of Asus, MSI, et cetera. The 12VHPWR connector was also used in newer RTX 40-series graphics cards like the RTX 4080 and RTX 4070 Ti, though not in AMD’s Radeon RX 7000-series GPUs.

Update: The latest updates we’ve added to the saga of the melting 12VHWPR connectors revolves around Intel slightly revising the ATX 3.0 power supply spec to recommend specific types of internal connectors for added reliability.

With this fast-moving, confusing, and also very serious situation, PCWorld has decided to round up the facts you need to know to help separate fact from fiction. Nvidia officials have declined to comment while it investigates, but the latest development seem to point to bad 12VHPWR adapter cables. There’s also the possibility that not fully inserting the cable may cause increased resistance and enough heat to melt the connectors. We’ll update this story as new information is released.

Sept. 10 Hassan Mujtaba of WCCFTech reports of an alert issued from PCI-SIG to members of a “thermal variance, which could result in safety issues under certain safety conditions.” The member alert advises vendors to “work closely with their connector vendors and exercise due diligence in using high-power connections”

Sept. 14 The full email and additional details from the PCI-SIG are reported by Stephen Burke of Gamers Nexus and notes that “failures have been observed in certain cable routing conditions from PSUs and test boards that generate side load on the interface.” Burke said the report—apparently created by Nvidia—from PCI-SIG showed three different manufacturers have been tested with 10 sample assemblies with failures manifesting from 10 hours to 30 hours with melting. It’s worth pointing out that the internal report seems to refer to the connection on the PSU side—not the GPU side. In general, however, ATX 3.0 power supplies PCWorld have seen indicate the cables to be identical on both ends.

Brad Chacos/IDG

Sept. 22 VideoCardz’s editor WhyCry reports that GPU maker Zotac’s guidance on the new 12VHPWR connector is rated for 30-insertion cycles which raises alarms as to the lifespan and durability of the new connector. VideoCardz later amends its report to say that while 30 cycles appears very low, many Molex connectors introduced over the last 20 years have had similar mating cycles.

Oct. 24 The first report of a melted 12HPWR connector is posted on the Nvidia sub-reddit. The GPU appears to be a Gigabyte 4090 Gaming OC using an Nvidia-branded 12VHPWR adapter cable. Both Nvidia and Gigabyte reach out to the owner who reports a replacement card has been received. A second report of a melted dongle is received as well on that day with damage to the adapter cable and an Asus RTX 4090 TUF Gaming OC Edition occurring. The Reddit post immediately goes viral on the high-profile graphics card with many assuming the new connector to be at fault.

Oct. 24 A few hours after the initial melting report on Reddit, renowned power supply reviewer and the principle behind PSU certification company Cybenetics, Aristeidis Bitziopoulos, attempts to replicate the melting 12VHPWR connector by subjecting it to 600 watt loads for more than 90 minutes. He is unable to damage the cable while seeing only a small thermal variance. It should be noted that the test used a native 12VHPWR cable on an ATX 3.0 power supply rather than Nvidia’s adapter. Bitziopoulos concludes the 12VHPWR connector doesn’t seem to be an issue in his testing.

Oct. 24 Overclocker Buildzoid of Actually Hardware Overclocking, posts a video criticizing the new 12VHPWR connector noting that the new connector drastically reduces the number of pins and wires carrying power.

[embedded content]

Oct. 25. With failures now reported at three, Nvidia officials tell the Verge’s Tom Warren that “we are investigating the reports” and are in contact with the owners of the impacted cards.

Oct. 25 Former HardOCP editor Kyle Bennett reports AMD’s upcoming RDNA3 GPUs will not use the 12VHWPR connector in its reference designs. Neither Bennett, nor his sources at AMD indicate when the design decision was made to skip 12VHPWR.

Oct. 25 Showing what a distraction the 12VHPWR has become, AMD’s Scott Herkelman publicly confirms the new Radeon cards will skip 12VHPWR and receives responses such as “That is a HUGE relief, happy with that news.”

Oct. 26 The official Reddit megathread listing showing documented failures now numbers five damaged 12VHWPR connectors.

Oct. 26 Jason Langiven, aka JayzTwoCents, who has long been critical of the connector being “dangerous,” attempts to replicate the failure on a native 12VHWPR cable and is unable to induce a failure on the cable under heavy loads.

Brad Chacos/IDG

Oct. 27 Igor Wallossek of IgorsLab.de conducts a tear down and failure test of a 12VHPWR power adapter and concludes that the issue doesn’t appear to be the 12VHPWR design itself nor the much-touted insertion cycle concern raised previously. Instead, Wallossek concludes it is the design of Nvidia’s adapter itself, which he describes as “inferior quality (and) can lead to failures and has already caused damage in single cases.” Wallossek said he believes bending and kinking of the adapter can cause weak solder joints and bridges to break and increase the resistance causing the melting.

Oct. 28 Ronaldo Buassali of TecLab.net.br posts his own failure tests, including swinging a power supply using just the connector and subjecting to a stress test of 1,532—well beyond its rated sustained wattage of 600 watts.

[embedded content]

Oct. 30 Stephen Burke of Gamers Nexus attempts to replicate the melting failure by intentionally damaging a 12VHPWR adapter similar to what IgorsLab.de had reported and subjected it to a 99 percent load for 8 hours with no melting observed. Burke also notes that his five adapters all appear to be constructed the same—and yet differently than the adapter IgorsLab had. Burke said his five 12VHWPR adapters use wires labeled for 300 volts versus the 150 volts the adapter Wallossek had. Burke concludes that we just don’t know what the issue is, but it is a real problem on some adapters—but not all of them. He also mentions a theory being floated that the smaller connector may not easily seat as well as the larger traditional power connectors. He also points out that contrary to what many consumers believe, a native connector that plugs directly into a power supply may also fail the same way if the native cable is constructed the same as the failed adapters. Burke also asks owners of RTX 4090 cards to report which cable adapters they have.

Just keeping everyone updated: Out of about 130 emails so far to the 4090cable inbox, we’ve received 7 that are 150V rated wires (and therefore potentially indicative of different supply), so 5%. That rating doesn’t instantly mean it’s bad. Replying to a few for info

— GamersNexus (@GamersNexus) October 31, 2022

Oct. 30 With news that there appear to be different 12VHPWR adapters being provided, Stephen Burke of Gamer’s Nexus reports via Twitter that of the 130 emails he has received, 7 percent of owners report they have the 150V cabling that was used in IgorsLab’s adapter cable. Burke notes that while the cable marking may say 150V, that only means it uses the same apparent spec cables—and does not indicate they are may have the low-quality solder joints that IgorsLab found. Burke also notes that of the 130, “not many are burned.”

Oct. 30 Andreas Schilling of Hardwareluxx.de conducts his own poll of forum members who have purchased RTX 4090 cards. He reports that 12 have a 4-pins-to-12VHPWR adapter marked “300V.” One has a 3-pins-to-12VHPWR marked “150V,” and two people have 4-pins-to-12VHPWR marked “150V.”

[embedded content]

Nov. 1 Ronaldo Buassali of TecLab.net.br posts a longer video of testing from the original live stream with additional explanations of how he tested the 12VHPWR. Unlike most of the testing so far, which used actual GeForce RTX 4090 cards, Buassali physically removes the 12VHPWR connector from the GPU and wires it up for stress testing. This lets Buassali push the connector assembly well past the 600 watts called for, including loads of 900 watts, 1,200 watts and 1,500 watts. Buassali’s conclusion? The 12VHPWR connector itself is “well sized, so much so that it supported much more than its specification.” However, Buassali concludes that even though the connector can handle more than it’s rated for, a poorly inserted connector that creates resistance could indeed be behind the melting of the connector. Buassali also doesn’t rule out a batch of bad cables, but that implies a manufacturing issue, not a design problem.

Nov. 2 Jon Gerow, director of R&D at Corsair and formerly of Jonnyguru.com, posts results from intentionally damaged 12VHPWR cable adapters under load and is unable to induce melting as well. Gerow was able to source multiple 12VHPWR adapter cables for destructive testing, and despite breaking off solder joints, he was unable to induce melting or a failure. He did note that some of the adapters weren’t constructed very well but even the worst of the batch passed stress testing without failing. Gerow concludes that some of the problems may have occurred when the owners didn’t fully seat the 12VHPWR adapter cables and also posts images of installed PCs where even a small gap of 1 mm could result in increased resistance.

Nov. 3 AMD formally announces its RDNA3-based Radeon 7900XT and Radeon 7900XTX and proudly notes that it did not use 12VHPWR connections. However, the company points out that the general perception that it changed its designs only after the melting problems cropped up a few weeks ago is not correct. AMD made the decision to stay with conventional 8-pin power connectors more than a year ago.

Nov. 4 A new post in the Nvidia subreddit, taken from a Facebook post of a Hong Kong-based RTX 4090 owner, is the first reported damaged 12VHPWR cable from a native cable plugged directly into a power supply. Previous to this report, all of the reported issues had only occurred in 12VHPWR adapter cables, not native cables. The following day, another person reports a melted connector using a native 12VHPWR cable from an ATX 3.0 power supply. This appears to dash hopes that a native plug would solve the problem.

Nov 7 The number of confirmed failed connectors now numbers 23 on the Reddit megathread, with issues spread among many graphics card makers. Oddly, there are no Nvidia Founders Edition cards listed with failures. There are also five unconfirmed cases listed from other board makers as well.

Nov. 7 VideoCardz editor WhyCry reports that a person on Reddit has been told his or her Gainward GeForce RTX 4090 will be delayed until the middle of November as it waits for replacement 12VHPWR adapter cables. The email, sent from Australian PC company Techfast to a customer, said “While investigations are still continuing and Nvidia has not released a public statement, Gainward has told us that cables shipped with their cards will (are) being replaced. As a result, they are holding shipping of all cards until this has taken place. We also understand this cable replacement will not be limited to Gainward alone.” PCWorld reached out to Techfast who confirmed the authenticity of the email.

Nov. 8 Despite Australian PC builder Techfast confirming an email sent to customers saying some RTX 4090 cards would be delayed while Gainward ships replacement cables and implying other GPU makers would do the same, Gainward’s EU Facebook page says that is not the case and has issued a clarification. “There is incorrect information suggesting Gainward is delaying the RTX 4090 shipment to replace the cables lately. Here we would like to clarify that—Gainward is not holding any RTX 40 shipment to replace the cables, and has no plan to do so. The cables Gainward currently used on the RTX 40 have been inspected by NVIDIA team and found no issues. All Gainward shipment is being made as usual. Please don’t hesitate to grab the Gainward graphics cards if you plan to buy one!”

Nov. 11 UK hardware site KitGuru receives an update from Nvidia that it is continuing to investigate the situation. “”We continue to investigate the reports, however we don’t have further details to share yet. NVIDIA and our partners are committed to supporting our customers and ensuring an expedited RMA process for them,” KitGuru says via Twitter.

Nov. 13 There are now 26 confirmed GeForce RTX 4090 GPUs with melted connectors documented in the Nvidia subreddit including for the first time, a report of an Nvidia GeForce RTX 4090 Founders Edition with a melted connector.

Nov 14 Igor Wallossek of Igorslab.de confirms there are at least two manufacturers of the Nvidia supplied adapter plugs with one of the adapters seemingly using a rougher casting making insertions more difficult. Wallossek, who said he has spoken with Nvidia’s director of engineering Gabriele Gorla, doesn’t blame end users, but does believe the rougher castings may possibly have lead some customers to to rock the 12VHWPR connector into the socket which may be why so many reports show melted connectors starting from one side of the connector with few to none melting from the center. The solution, Wallossek said, may indicate the PCI SIG needs to issue guidance on keep out zones to give consumers better access to the socket to directly inserting it rather than rocking the connectors in from one side. Wallossek also notes that of the two adapter cables used supplied by Nvidia, one design only grabs pins on two sides while the other design grabs pins on all four sides.

Nov. 16 In an exhaustive investigation relying on third-party failure analysis using X-ray examination as well as electron microscope exam, as well as reproducing the failures, Gamers Nexus’ EIC Stephen Burke concludes debris inside the connectors from manufacturing or from insertion as well as improperly seating the cables is largely behind most of the failures seen. Burke also cites numbers from a vendor that the risk of failure is from 0.05 to .1 percent at this point. Although improperly seating the cables while putting it under strain seems to get most of the blame, Burke also wonders if the design of the connector shouldn’t have anticipated the issues.

[embedded content]

Nov. 16 Tom’s Hardware reports that a GeForce RTX 4090 owner has filed a class-action in the US District Court for the Northern District Court of California over the melting connectors. Filed by Lucas Genova on Nov. 11, the suit alleges Nvidia “marketed and sold the RTX 4090 with a defective and dangerous power cable plug and socket, which has rendered consumers’ cards inoperable and poses a serious electrical and fire hazard for each and every purchaser,” according to the Tom’s Hardware report.

Nov. 17 Looking to pour a little salt into the wounds of its competitor, AMD officials have been touting their fortune of not implementing the new 12VHPWR connector in its upcoming Radeon 7000-series of cards.

Nov. 18 After weeks of mostly silence, Nvidia finally issues a statement that it has received 50 known reports of melted connectors and after analyzing the returned cables, has largely found improper insertion to be the likely cause, according to a report by Gamers Nexus. The company also poured cold water on building concern that use of third-party 12VHPWR cables would void warranties. The company told Gamers Nexus EIC Stephen Burke that it would honor warranties related to the issues and would expedite the RMA process. While improper insertion would indicate user error during installation of the the cards is to blame, Nvidia also told Gamers Nexus it is looking at ways to improve the connectors that was homologated by the PCI-SIG.

Nov. 18 The official Nvidia Reddit megathread closes out its logging of reported incidents which appear to drop off after Gamers Nexus video and Nvidia’s official notice. However, one Reddit user posts an image of a melted RTX 4090 12VHPWR connector in the PCMR subreddit on Nov. 24.

Dec. 1 With concern over melting 12VHPWR connectors greatly subsiding after reports that improper insertion is likely the primary cause, the PCI SIG releases a statement reminding people that its spec addresses interoperability, not safety, and each member is responsible for their own products. “PCI-SIG wishes to impress upon all Members that manufacture, market or sell PCI-SIG technologies (including 12VHPWR connections) of the need to take all appropriate and prudent measures to ensure end user safety, including testing for the reported problem cases involving consumers as alleged in the above-referenced lawsuit. Members are reminded that PCI-SIG specifications provide necessary technical information for interoperability and do not attempt to address proper design, manufacturing methods, materials, safety testing, safety tolerances or workmanship. When implementing a PCI-SIG specification, Members are responsible for the design, manufacturing, and testing, including safety testing, of their products.”

[embedded content]

Feb. 2023 Intel releases a minor update to the ATX 3.0 spec that recommends that power supply vendors use internal connections using spring-type rather than dimple-type connections. Intel said it did this as the suggestion of connection companies themselves as the spring-type design offers more surface area. The different connection recommendation, however, does not mandate it for all, and older dimple-style connectors can continued to be used if a power supply vendor chooses to. Existing power supplies using dimple-style connectors also do not need to be replaced—they work as expected if inserted correctly.