From Newsgroup: comp.sys.ibm.pc.games.action


An amusing, if informative tale*:

Ever since the 3xxx series, there have been stories about how Nvidia
cards are catching on fire. Well, not so much fire as melting and
scorching the on-card power connectors. Either way, the card didn't
come out very well from the experience and, while there have been any
tales of actual 'oh-my-god-my-whole-PC-is-on-fire!" experiences, I
think it's probably only a matter of time. These problems were
exacerbated with the release of the 4xxx and 5xxx line of nvidia cards (especially the 49xx and 59xx cards), which demanded even more power
from the connectors.

Nvidia, of course, did what any responsible corporation would do. The
first denied the problem even existed, then tried to lay the blame on
other components (it's the power supply at fault! it's the cheap
cables you're using! It's the user who plugged it in wrong!), before
admitting that maybe, in some very rare rare rare cases (which aren't
that rare, apparently) the connectors might fail... but here are some half-hearted work-arounds that, sure, may prevent the melting power
cable issue, but also hobble your GPUs performance, making that $1500
card run the same as something half the price.

In fairness, nvidia isn't entirely wrong that the problem lies (at
least partly) with the cables. The 12 volt 2x6 connectors used in
these cards can pull up to 450 watts of power, and that's a lot of
juice to go through such a tiny package. Each pin carries 6.5 amps of
current. That's well within the design rating of the cables and
connectors. The problem is that if one of those pins can't keep up
that current (maybe it's broken, or corroded, or whatever), the load
gets transferred to the other 5 pins... and that pushes them over the
design limits. Shortly thereafter, the magic smoke is released.

PSU and cable manufacturers have tried to work around the problem, but ultimately a better solution needs come from nvidia. They need to
ensure that no single pin can overdraw current. This would probably
require a redesign of the entire connector. Even better, they need to
start making GPUs that don't demand 450W of power. These solutions
would solve the melting issue far better than pushing it onto
PSU/cable manufacturers.

Because let's face it: with computer prices skyrocketing, and margins
on PC sales so razor thin already, the first thing that OEMs are going
to start cutting back on is quality of the PSUs and cabling. Which
means we're going to be seeing even more stories about nvidia cards
catching fire... and while it's the cables that are at fault, it's the
nvidia cards that are causing them to fail so spectacularly. And
nvidia has had years to rectify this issue. There's already a
class-action suit acting nvidia from 2022; nvidia may or may not
escape blame in that one. But any future suits will have ample history
of nvidia ignoring the problem and probably have a lot more success.

Meanwhile, ARM is really starting to amp up its own GPU capabilities.
Sure, it's not anywhere near the capabilities of a fully powered
5090RTX yet... but if ARM knows anything, it's getting a lot of
processing done with a lot less power draw, and that's becoming
increasingly important in the market.




* story https://www.pcgamer.com/hardware/graphics-cards/possibly-the-first-instance-of-asus-anti-melting-12v-2-6-power-cable-err-melting-shows-up-adding-more-fuel-to-the-fire-that-is-nvidias-connector/

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.sys.ibm.pc.games.action

At Fri, 12 Jun 2026 10:40:16 -0400, Spalls Hurgenson <spallshurgenson@gmail.com> wrote:

An amusing, if informative tale*:

Ever since the 3xxx series, there have been stories about how Nvidia
cards are catching on fire. Well, not so much fire as melting and
scorching the on-card power connectors. Either way, the card didn't
come out very well from the experience and, while there have been any
tales of actual 'oh-my-god-my-whole-PC-is-on-fire!" experiences, I
think it's probably only a matter of time. These problems were
exacerbated with the release of the 4xxx and 5xxx line of nvidia cards (especially the 49xx and 59xx cards), which demanded even more power
from the connectors.

Nvidia, of course, did what any responsible corporation would do. The
first denied the problem even existed, then tried to lay the blame on
other components (it's the power supply at fault! it's the cheap
cables you're using! It's the user who plugged it in wrong!), before admitting that maybe, in some very rare rare rare cases (which aren't
that rare, apparently) the connectors might fail... but here are some half-hearted work-arounds that, sure, may prevent the melting power
cable issue, but also hobble your GPUs performance, making that $1500
card run the same as something half the price.

In fairness, nvidia isn't entirely wrong that the problem lies (at
least partly) with the cables. The 12 volt 2x6 connectors used in
these cards can pull up to 450 watts of power, and that's a lot of
juice to go through such a tiny package. Each pin carries 6.5 amps of current. That's well within the design rating of the cables and
connectors. The problem is that if one of those pins can't keep up
that current (maybe it's broken, or corroded, or whatever), the load
gets transferred to the other 5 pins... and that pushes them over the
design limits. Shortly thereafter, the magic smoke is released.

PSU and cable manufacturers have tried to work around the problem, but ultimately a better solution needs come from nvidia. They need to
ensure that no single pin can overdraw current. This would probably
require a redesign of the entire connector. Even better, they need to
start making GPUs that don't demand 450W of power. These solutions
would solve the melting issue far better than pushing it onto
PSU/cable manufacturers.

Because let's face it: with computer prices skyrocketing, and margins
on PC sales so razor thin already, the first thing that OEMs are going
to start cutting back on is quality of the PSUs and cabling. Which
means we're going to be seeing even more stories about nvidia cards
catching fire... and while it's the cables that are at fault, it's the
nvidia cards that are causing them to fail so spectacularly. And
nvidia has had years to rectify this issue. There's already a
class-action suit acting nvidia from 2022; nvidia may or may not
escape blame in that one. But any future suits will have ample history
of nvidia ignoring the problem and probably have a lot more success.

Meanwhile, ARM is really starting to amp up its own GPU capabilities.
Sure, it's not anywhere near the capabilities of a fully powered
5090RTX yet... but if ARM knows anything, it's getting a lot of
processing done with a lot less power draw, and that's becoming
increasingly important in the market.

* story https://www.pcgamer.com/hardware/graphics-cards/possibly-the-first-instance-of-asus-anti-melting-12v-2-6-power-cable-err-melting-shows-up-adding-more-fuel-to-the-fire-that-is-nvidias-connector/

Disturbing to hear.

I have a new workstation sitting on a pallet* in the garage waiting
for me to unpack and set it up.

It contains:

Graphics: 32 GB NVIDIA Geforce RTX 5090

Now I'm afraid when I boot, it might halt-and-catch-fire.

* Yes, System76 ships their tower workstations on small pallets.
They are meticulously packed. (Not a solicited endorsement.)
--
-v System76 Thelio Mega v1.1 x86_64 Mem: 258G
OS: Linux 7.1.0 D: Mint 22.3 DE: Xfce 4.18 (X11)
NVIDIA GeForce RTX 3090Ti (24G) (610.43.02)
"Happiness is finding special characters ."
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.sys.ibm.pc.games.action

It's only tangentially related to the fires, but there's an
interesting analysis of the other problems facing nvidia here: https://www.youtube.com/watch?v=nwrDdpkw_oA

The short of it is that nvidia isn't betting so big on AI because they
believe that it's the Future Of Technology, but because their original
market --gaming hardware-- was struggling (that feeding the AI and
crypto markets was incredibly profitable didn't hurt either, but it's
not anything to build a long-term business strategy upon).

Basically, nvidia just hasn't been able to provide the same
performance gains year after year as it was prior to 2020. It used to
be that nvidia was able to release new cards on an almost yearly basis
that saw significant speed improvements over the previous year's
models, without major cost or thermal/power increases. This gave users incentive to buy new cards every 2-3 years. But new releases are now
released two years (or more) apart, and the performance improvements
are much less, all the while costing more to develop, more to build,
and having much higher thermal and power draws.

End result: the per-card margin is much less than it used to be even
as there is less incentive for end-users to buy the damn things.
Nvidia is spending more to make the cards, making less money from the
sales, and the end-product is less attractive to users.

Is it any wonder they've turned towards AI, a bubble with too much
money and too little sense in how to spend it, to make up the deficit?
Or that they'll spend trillions to prop up that industry for as long
as it'll last?

The video itself ends rather pessimistically, saying its unlikely
we'll ever return to the salad days of the 2000s and 2010s, when we
saw rapid performance gains without significant cost increases. I'm
not sure I'd go that far, but I do agree Nvidia has backed itself into
a corner for the time being with the technology it has.

Nvidia's Pascal architecture was an amazing leap back when the 10xx
line was released, but it hasn't scaled well and nvidia has been
brute-forcing the tech , or relying on software tricks (DLSS and such)
for any performance gains since then. It needs a refresh and new ideas
for nvidia to reach the next level. And maybe one day they'll manage
this, but in the immediate future it doesn't seem likely.

But on the other hand... nvidia isn't the only company out there. AMD,
ARM and even ARC are increasingly competitive (even if nvidia remains
the 800lb gorilla in the room). ARM especially may be the way of the
future, as it rushes forward in performance gains while maintaining
its reputation for low thermal/power load. As much as I am an x86
fanboy, I think it's only a matter of time before PC gaming makes the switch-over to ARM.

But besides all that... there's increasingly less need and desire
amongst gamers for super-high-end GPUs. Partly because of sour grapes
("It's too expensive / I can't find a card so I don't want one
anyway!") but also because games already are so fucking gorgeous.
Increasingly we seeing developers move towards more stylized visuals
because graphics are Good Enough that competing for the Most Realistic
Visuals just isn't a winning move anymore. And while there will always
be a few games that push the envelope, it's not like you really NEED
that 5090RTX. And if that means that we can get buy with less powerful
(and less expensive!) cards, I'm all for it.

Still, despite all these problems, it's hard for me to have any
sympathy for nvidia. Their refusal to lower their margins per card and
focus on markets outside of their core gamer are why they're in the
place they are. If this means their competitors start getting a leg up
or they have to invest more in gaming GPUs to keep profitable, it's
only advantageous to us.





--- Synchronet 3.22a-Linux NewsLink 1.2