Chiphell - 分享与交流用户体验

标题: 外媒发现 660W 持续负载烧 12VHPWR 原装线材，未来4090注意 [打印本页]

作者: lgblee 时间: 2022-9-24 12:15
标题: 外媒发现 660W 持续负载烧 12VHPWR 原装线材，未来4090注意
本帖最后由 lgblee 于 2022-9-24 13:12 编辑

https://www.nichepcgamer.com/arc ... elting-problem.html

测试环境

660W 状态下持续工作
（注意 12VHPWR 接口是设计成符合额定 600W 峰值 660W 的工作条件，这里 660W 持续工作是超载运用，但肯定有显卡厂商既要 4090 超载工作，又只给一个接口，相信到时买这些单接口 4090 超载运行的玩家并不会只有少数）
12VHPWR 线材每隔 30mm 弯折一次
室温 26 度

结论

接头端子过热融解，3个品牌、不同设计的 4090 下均有发生
具体是工作 2.5 小时后产生高温热点，10-30 小时后出现融解
3个品牌中，最少 10 个个体出现融解或高温热点
未弯折的 12VHPWR 线材不融解也不产生高温热点，但插拔次数多于 40 次的除外

作者: PolyMorph 时间: 2022-9-24 12:18
米冲高关键年，你发这是不是不想活了

作者: yoshiki1985 时间: 2022-9-24 12:18

不好的预感！

作者: wangzhiqiang 时间: 2022-9-24 12:25

无所谓，反正不打算买4060ti以上的显卡

作者: 让我静一会 时间: 2022-9-24 12:30
虽然我看不懂日文但我通篇也没看到说测试样本是4090，然后我去GN看了下原文也没说是4090，最后说造成这个情况的可能原因是端子没插好。

作者: czdmz 时间: 2022-9-24 12:31
所以这个12VHPWR 看来并不成熟么

作者: BFG9K 时间: 2022-9-24 12:35
660W是16PIN端子600W+PCI-E插槽60W
而且这个新闻之前被pci-sig辟谣了，是插歪了才会烧

作者: lgblee 时间: 2022-9-24 12:35

让我静一会发表于 2022-9-24 12:30
虽然我看不懂日文但我通篇也没看到说测试样本是4090，然后我去GN看了下原文也没说是4090，最后说造成这个情 ...

其他能 660W 持续工作的显卡又只用单个 12VHPWR 应该不存在吧

作者: aaaa12345 时间: 2022-9-24 12:40

PolyMorph 发表于 2022-9-24 12:18
米冲高关键年，你发这是不是不想活了

赤祼祼的恐吓

作者: Jimmy_Qiu 时间: 2022-9-24 12:43
从原文看，很可能用的是电子负载在660W连续负载下测试的。
Molex这个端子还是靠多个点接触，跑6A多真觉得不太稳。

要是显卡功耗继续增加下去，PCI-SIG还是考虑XT90这种端子吧

作者: frontwing 时间: 2022-9-24 12:43
坐等看黄世仁的乐子

作者: lgblee 时间: 2022-9-24 12:47
本帖最后由 lgblee 于 2022-9-24 13:04 编辑

Jimmy_Qiu 发表于 2022-9-24 12:43
从原文看，很可能用的是电子负载在660W连续负载下测试的。
Molex这个端子还是靠多个点接触，跑6A多真觉得不 ...

也有可能，但是原PDF说到了 3 manufacturers 肯定不会是一般的负载器，至少实际生产出来的显卡回路是要有的。

更新：仔细看一下，3 manufacturers 没说是什么东西的厂家，可能是显卡/负载器，可能是电源，还可能是线。

作者: yangzi123aaa20 时间: 2022-9-24 12:52

lgblee 发表于 2022-9-24 12:35
其他能 660W 持续工作的显卡又只用单个 12VHPWR 应该不存在吧

看文章是用直流电源直接测试的，而且是把线头那里进行了比较大的弯折
当然，这玩意也宣告民间定制线基本死亡了

用3080这种东西上定制线导致烧线的都不少，4090还敢上怕是瞬间起火

作者: 让我静一会 时间: 2022-9-24 12:53
本帖最后由让我静一会于 2022-9-24 12:59 编辑

lgblee 发表于 2022-9-24 12:35
其他能 660W 持续工作的显卡又只用单个 12VHPWR 应该不存在吧

如果我没理解错左边的那一坨负载设备并不是显卡

作者: lgblee 时间: 2022-9-24 12:54

BFG9K 发表于 2022-9-24 12:35
660W是16PIN端子600W+PCI-E插槽60W
而且这个新闻之前被pci-sig辟谣了，是插歪了才会烧 ...

弯折显卡线在ITX设备中超级常见，是有可能引发插歪，但传统8针年代定制线烧端子烧线都是比弯折质量过关的线还要更严重的接触不紧密问题造成。

作者: aaaa12345 时间: 2022-9-24 13:00
这里有多少N卡的拖不言而寓，每当这个时侯总是蹦出来顾左右而言他以混视听，别把这一片净土给污了，过来过去明贬暗撒来着

作者: chp1979 时间: 2022-9-24 13:06
米冲高这种梗怎么还真有人当真啊？

没活了可以咬打火机

作者: lgblee 时间: 2022-9-24 13:06

让我静一会发表于 2022-9-24 12:53
如果我没理解错左边的那一坨负载设备并不是显卡

如果那 3 manufacturers 不是显卡或负载器，而是电源或线的厂商，那这个标题确实有误。
那标题改一下 660W 负载烧线

作者: atiufo 时间: 2022-9-24 13:14
总想搞个大新闻，这根线只能拔插二十多次的新闻出来后，现在已经有人传成显卡上的母口只能拔插二十多次了，多插几次显卡直接报废

这贴今后传下去就是插12VHPWR接口有大概率会直接烧毁4090，线材以及电源，你看着吧

作者: BFG9K 时间: 2022-9-24 13:21

atiufo 发表于 2022-9-24 13:14
总想搞个大新闻，这根线只能拔插二十多次的新闻出来后，现在已经有人传成显卡上的母口只能拔插二十多次了， ...

16pin真这么不耐用的话那用物理规格和16pin一样的12pin的30系FE早就会有一堆烧卡的反馈了，然而并没有

作者: 真实讣告 时间: 2022-9-24 13:27
用在电脑上的Molex Mini-fit母头端子标称寿命都是30次，新的12VHPWR和过去4/6/8/24pin PCIe/EPS/ATX是一样的，这个测试本身是PCI-SIG早在RTX40系发布之前就搞的，下次搞点大新闻之前还记得提高一下自己的姿势水平

作者: texnis 时间: 2022-9-24 13:28
以后大功率显卡是不是应该在12V电源接口上加个温度传感器了？一旦温度急剧升高有烧融接头端子风险，立刻降频报警（显卡上装个报警LED和蜂鸣器，不依赖CPU软件）

作者: cast 时间: 2022-9-24 13:55
(, 下载次数: 164)

不懂日语，但这个图才是关键，看得明白，就自然明白，不管描述的是插头还是内金属插针，

看明白了，就知道如何处理折弯和线的拉扯处理，当然如用的是烂插头件，什么也救不了，

这么多年，当一打开别人装的机箱盖时.......................

作者: b44n8gsw99 时间: 2022-9-24 14:08
本帖最后由 b44n8gsw99 于 2022-9-24 14:11 编辑

烧显卡供电接口以前也偶有发生，原因不外乎几种，1：接口不规范，松动，接触不良造成的，2：材质，使用非铜制材质，接口产生很大的接触电阻，过热，3：电流过大，超过了接口的承载能力，4：接口温度过高，母口受热膨胀，时间久了就会接触不良

作者: zkdavids 时间: 2022-9-24 14:26
我电源小，顶多上个3070ti

作者: kaiwenwu 时间: 2022-9-24 14:26
molex有插拔次数200次的座子和插头，用料好一点就没有这种事情了

作者: kaiwenwu 时间: 2022-9-24 14:27

真实讣告发表于 2022-9-24 13:27
用在电脑上的Molex Mini-fit母头端子标称寿命都是30次，新的12VHPWR和过去4/6/8/24pin PCIe/EPS/ATX是一样 ...

但是molex也有寿命200次的

作者: hp6325 时间: 2022-9-24 16:06
提示: 作者被禁止或删除内容自动屏蔽

作者: Jimmy_Qiu 时间: 2022-9-24 20:34
等等，我又看了一下图，这不是Gamers Nexus的视频截图么。这日语内容恐怕也是翻译自下面链接的内容的。
直接去原链接看吧。
https://www.gamersnexus.net/news ... -melting-gpu-cables

作者: Jimmy_Qiu 时间: 2022-9-24 20:40

PCI-SIG, the group that governs PCIe standards, has issued an email to all of its members and their suppliers regarding melting 12VHPWR cables. That email was technically private, but so many groups were on the email that it inevitably got passed around and ended up in our hands too.

原文引用，是PCI-SIG发的email告知会员有此问题的。

The PDF explains that Nvidia has been testing 12VHPWR connectors to validate that prototype power supplies and cables can meet the specification of 55A continuous. During this testing, Nvidia found certain cable routing conditions led to excess heat and, in some cases, melting.

原文引用，从此段看，是NVDIA发现的问题，然后通知的PCI-SIG。

作者: Jimmy_Qiu 时间: 2022-9-24 20:46
本帖最后由 Jimmy_Qiu 于 2022-9-24 20:53 编辑

在email中PCI-SIG把锅甩给了制造商。
注意并非显卡制造商，而是连接器和线缆的制造商。
原文如下。

Please be advised that PCI-SIG has become aware that some implementations of the 12VHPWR connectors and assemblies have demonstrated thermal variance, which could result in safety issues under certain conditions. Although PCI-SIG specifications provide necessary information for interoperability, they do not attempt to encompass all aspects of proper design, relying on numerous industry best-known methods and standard design practices. As the PCI-SIG workgroups include many knowledgeable experts in the field of connector and system design, they will be looking at the information available about this industry issue and assisting in any resolution to whatever extent is appropriate.

作者: Jimmy_Qiu 时间: 2022-9-24 20:50
全文如下，当然愿意的点链接去看视频也行。不过也是英文而且在油管。
来源:https://www.gamersnexus.net/news ... -melting-gpu-cables

Melting 12VHPWR Cables

PCI-SIG, the group that governs PCIe standards, has issued an email to all of its members and their suppliers regarding melting 12VHPWR cables. That email was technically private, but so many groups were on the email that it inevitably got passed around and ended up in our hands too.

As a reminder, 12VHPWR is the official name for the new PCIe 5.0 12+4-pin cable used on some current and upcoming high-power graphics cards and is capable of handling 600W sustained on a single cable.

You may have already seen this story covered on some rumor sites citing that the problem is with adapters converting to 8-pin PCIe connectors, but that’s incorrect and may have been based on incomplete information. Here’s what the email had to say:

“Please be advised that PCI-SIG has become aware that some implementations of the 12VHPWR connectors and assemblies have demonstrated thermal variance, which could result in safety issues under certain conditions. Although PCI-SIG specifications provide necessary information for interoperability, they do not attempt to encompass all aspects of proper design, relying on numerous industry best-known methods and standard design practices. As the PCI-SIG workgroups include many knowledgeable experts in the field of connector and system design, they will be looking at the information available about this industry issue and assisting in any resolution to whatever extent is appropriate.”

“As more details emerge, PCI-SIG may provide further updates. In the meantime, we recommend members work closely with their connector vendors and exercise due diligence in using high-power connections, particularly where safety concerns may exist.”

The PDF explains that Nvidia has been testing 12VHPWR connectors to validate that prototype power supplies and cables can meet the specification of 55A continuous. During this testing, Nvidia found certain cable routing conditions led to excess heat and, in some cases, melting.

The conditions required for the excess heat were either subjecting the cables to severe bending or a high number of mating cycles (about 40). Cables tested in these scenarios exhibited hot spots at roughly 2 and a half hours, and melting at 10 to 30 hours. Connectors from multiple suppliers have failed.

This is with a continuous 55A of current (or 660W at 12V), which would not be a typical load condition, especially not in gaming. Nvidia did not observe any failures on connectors with low mating cycles and without any bend.

Photos provided in the PDF show some pretty gruesome melting, and it’s not in the same area each time. The failures occurred on different pins depending on the direction the cable was bent. This could be seriously dangerous.

The document includes per-pin measurements taken during the testing. As the cable was bent in various directions, severe current imbalance resulted from huge swings in resistance. We’ll use the last set of data as an example as it’s the most severe. The resistance in pins 3 and 4 measured high, especially pin 3, resulting in a measured 36.4A on a single pin, or 436W on a single pin, leading to a hotspot temperature of 180 degrees celsius. By the way, the current rating of stranded 16AWG is only between 5A to 7A, so we’re talking about 5 to 7 times the rated current.

The PDF goes on to hypothesize that the bending and side-loading cause the plug to improperly seat in the receptacle, perhaps deforming it. The testing conditions might seem extreme, with the cable being bent around at full load for hours on end, but this kind of thing is done to ensure a margin of safety as products get used and age in various circumstances.

Group members are encouraged to do independent testing and share the results with Nvidia, who also has volunteered to work with the manufacturers of the connectors to fix this issue. Nvidia and the PCI-SIG are trying to get ahead of a potential problem before it is allowed to become widespread.

Our opinion is that while the situation is serious and should be taken seriously, this isn’t likely to be a problem that you’ll encounter in your gaming PCs. The test conditions are intentionally extreme in a way that you likely won’t have in your own system, especially if you take care not to put too much strain on any of your cables and connectors. It’s never a good practice to shove, bend, or cram any cables.

作者: JP_ToKyo 时间: 2022-9-24 22:24
看來只能買雙16P的顯卡..單16P不安全...

作者: yoloh 时间: 2022-9-24 23:15
有没有像我这样的，从来不用超过200瓦的显卡的。
性能不够画质降一点就行了，玩游戏比较注重游戏性，不太在乎画质和帧数，只要不卡就行。

欢迎光临 Chiphell - 分享与交流用户体验 (https://www.chiphell.com/)