Nvidia RTX 4090: So hot they're melting power cables GPUs have been a hot commodity but this is ridiculous. Nvidia's use of a 600W power socket and cable with its RTX graphics cards may have sounded like a good idea at the time but some users have reported overheating issues. One unlucky user uploaded photos of an Nv-branded 12VHPWR power cable that had melted where it plugs …
Well, it's got 16K cuda cores, so I can imagine it being power-hungry.
Of course it's only needing the nuclear reactor feed when it's doing a serious game.
My big question is with all this power dissipation, what do the fans sound like? I remember having one card where you started a game, and it sounded like the computer was going to levitate on fan power.
If you look at modern cards, they're 70% heatsink and heatpipes, 2 or 4 fans, and all other sorts of cooling gimcrackery.
Since you can actually buy 500W fan heaters, 600W does seem a bit on the ridiculous side.
Even if they chose to undervolt these things and make the TDP a little less, they'd still blow the previous gen cards out of the water, so why they chose to make them this power-hungry is a bit of a question...
From my perspective the problem is creating a 600W rated connector...and then drawing 600W on it. You should always design an electrical interface with a safety margin, the socket should have been designed for 700W to draw 600W. Drawing the rated load from a connector is pretty much a sure-fire way to assure overheating.
"From my perspective the problem is creating a 600W rated connector"
The connector would be properly rated for how many amps it needs to connect, not watts.
I think that one problem could be that crimped connector pins don't always make good contact and a simple continuity check won't show the problem. Another problem can be cut wire strands which can create a high resistance point at the connector which will heat up under load as the remaining strands can't carry the current.
At 12V, 600W will be 50 amps of current. As the heating effect is proportional to the square of the current (P=I2R, this becomes a major problem.
This is the reason most power delivery is done at high voltages and low currents. Delivering high power at low voltages is a recipe for a heating disaster.
Maybe they should make these things work on 48V or something, rather than 12V.
Does this mean that Nvidia 4090 cards are hot sh?t?
Looking at that cable and connector set (both in the images linked in TFA and in diagrams for the spec) I'm not too certain that someone hasn't dropped a decimal point in a calculation somewhere. But I've not done that level of engineering math in quite some time. Perhaps the materials in use aren't what they used to be?
A connector that is smaller, with less pins than two 8xPCIe connectors, supposed to deliver 600 watt. It felt strange right from the start.
When I see servers which ACTUALLY eat 800 Watt (whole machine), and how the PSU is connected to the mainboard in comparison: That server style connector comes with 16+ connectors for the 12V rail alone. And each connector is wider and often double-spring.
Better would have been to take a look into the past: Expose the PCB on about four to five centimeters with gold plated contacts to plug the power on. Has been used for decades, even way before IBM released their first PC, way before the C64 appeared.
Nor am I surprised either..
600W at 12V is.. 50A. And any fule no: Ohm's law says resistive heating goes with the square of current. If you double the current with the same number of pins & same gauge wire, then you quadruple the power (heat) loss in those pins/wires. If you double the current and double the number of pins/wires, then you still double the heat.
Nvidia have doubled their power consumption, but the supply voltage standard is still 12V, so they are doubling the current. But they certainly haven't quadrupled the number of pins. So it's not surprising that they are getting hotter.
On the face of it, those Molex pins are supposed to be rated to 10A - so 12 of them (6 pairs of + and -) naiively ought to be rated to 60A. But in practice if you put multiple pins in parallel then the slightest bit of dust/grease/oxidation can cause most of the current to go through a few pins. Especially since they don't seem to be gold plated.
Plus you have a bundle of cable attached that is, under that kind of load, as much of a heat source as a heat sink..
Maybe they should switch to XT90 connectors for GPUs. That and/or increase the voltage standard.
Or on the other hand we could always, er, stop wrecking the planet and wasting silicon with all this "AI"/VR/Blockchain bollocks and go back to simpler days when GPUs only needed to draw a few thousand polygons at 30fps to run a perfectly good fun 3D video game, and that was all we ever used them for.
<ducks half-empty bottle of amphetamine-loaded whisky thrown by some resident investment banker>
Mine's the one with the Quake II logo
">Maybe they should switch to XT90 connectors for GPUs.
Or just put battery terminals on them and we use jump leads"
Anderson connectors. I don't think that XT90's would be good enough. 50 amps takes a big chunk of wire to move and the voltage doesn't enter into it. For chassis wiring with some margin, 8gu is recommended.
Stemming from the time video cards simply drew power from the PCI bus (or VESA or recall VLB VesaLocalBus) to the point where graphics cards demand 50A of juice is astounding. Looking at the reddit photos and the melting that occurred, both the plug-in (male) and base connector (female) suffered real damage.
The tiny size of the connector pins really sticks out. They look no bigger than normal 1mm bread-board header-pins. While that maybe fine for Milli-Amps, 50 Amps across those slender pins are a recipe for a very hot connection (not to mention 50A is enough of a boost to start a 7.3L diesel V8 when the batteries are low). That's a hell of a lot of current.
Looking at the female connector of the RTX 4090 in the photo, the keyed slots where literally destroyed with what remained of the melted male parts of the connector still lodged in base connector. Two (or three) 8-pin Molex connectors on prior models look far more robust by comparison. The male part of the adapter simply looked too fragile to reliably carry that current - especially given the variances that can occur in the pinouts during manufacture.
Good article. Definitely made me appreciate the 2070 cards put in my daughters gaming rig.
This post has been deleted by its author
This post has been deleted by its author
Nice selective quote there, cutting off the pertinent bit of the sentence (as I read it), the cores of all graphics cards.
Yes, a consumer-grade RTX 4090 might not be rated to run continuously for 24 hours a day at max TDP, but the only difference between the chips in that, and those in a card found in a render-farm is going to be, at most, a slight difference in the number of CUDA cores, and memory bandwidth. There might also be some production tolerances there, where the chips that benchmark the best in a wafer might get diverted to high-end cards, but I don't know if the manufacturer (Nvidia) would even bother with that. The difference between a gaming card and a professional card is everything else that goes onto the board, from the quality and tolerance of the caps to the type and reliability of the cooling system. Those vary wildly. One would hope that professional cards also have better power connectors.
If you look at the prices of those high-end cards, it's pretty obvious that they are not going to be destined for a Joe-average desktop gaming PC. The non-gaming cards are likely to have 5-figure prices.
So you say: ATI/AMD is the same as NVIDIA the same as INTEL the same as MATROX is the same as S3, the same as TsengLabs the same as...
That is the issue I am pointing out. Not the difference between games graphics cards and roughly the same as workstation card for rendering, as you point out NOW, but not in your original posting. Apart from that there are quite some differences even withing one maker of GFX cards, sometimes new generations really change a LOT, making old games actually slower, which is only compensated by either adding more cores or by a higher frequency. But a specific new feature is a lot faster with the newer card. Around 2009 it was Tessellation for games, for example (Tessellation in hardware existed long before).
This post has been deleted by its author
This post has been deleted by its author
Having dealt with high frequency, high current connections (induction heaters, 20+kW), the best advice is - sand and scrub the connection, apply deoxidizer paste, scrub again and assemble it sloppy-drippy, then clean up and torque. Wash with solvents if necessary, but then re-clean and re-torque. Cleanliness is next to [diety]ness and torque is key to a tight connection.
I've seen em but I didn't know the series name. Seems exactly the thing for the kind of bananas power draw these things are laying down.
Yeah, Molex Mini-fit pins are NOT good for 10A of sustained draw. They will melt the shell and then spot weld themselves. Looking at they way they crimp on, it should be obvious. Just made from a bent up thin coated brass sheet.
This is why you can't just look at the box when designing something. It's not like there aren't papers on this stuff. What they did is like installing 15A of sustained load on a 15A breaker. That teachable moment where someone says "Yes, the numbers match, no it won't work, because you don't understand what the ratings mean. Go grab a 20A breaker, that 15A part is for ten amp overhead lighting loops that rarely pull 3."
This is literally why we have a National Electrical Code. Those rules about conduit fills are there for a reason to.
Right. And you'd never see a building installation using separately-insulated narrow-gauge wires with low-current connectors in parallel to connect something like an electric cooker or shower.
But "Computer Standards" were written by IBM in the 1980s, and they obviously had some sort of deal going with Molex. They never thought anything would draw more than 1A through the 12V wire (probably only needed for erasing the EEPROM and maybe spinning a disk platter).
(at that time they didn't have "point-of-load" switching regulators on the motherboard itself, so they were more worried about the current on the 5V and 3.3V wires, which are barely used these days)
But for a PC to be "ATX Compatible" it has to use those shitty crimped Molex connectors that were designed in the days before CNC machining was a thing, and only take wire up to 16AWG or so.
Don't blame IBM. They were very good at providing what was required, and at the time, the first couple of generations of PCs were relatively low powered devices. What happened after the PC/AT had very little to do with IBM.
If you looked at the power systems of mainframes of a similar vintage, there would have been thick wires and large connectors for power distribution.
The IBM HPC installation I used to work at a few years back had extremely substantial, very solid bus bars to take the higher currents inside the drawer of each node, and they were distributing power around the racks from rack-top PDUs at something like 48V, and then having local DC-DC converters close to where the power was needed to keep the current down.
The problem we have here is that the ATX form factor for systems requiring high powered GPUs is no longer really fit for purpose, and someone should really come up with something better, rather than shoe-horning GPUs and CPUs with high power draw and huge active cooling into a form that is 25+ years old.
That does not look like a high amp connector. Those exist, but this is certainly seems way past what the pins are rated for. 600w should either have a tea kettle cord or multiple ~18ga wires on it. Switching to a new connector would mean co-ordination motherboard and PSU companies, but would probably save money, space, and sanity by cleaning up the connections and cutting down the number of wires. I don't know why they would go with something this tiny looking.
I realize that these new cards are pushing into Anderson power pole territory(google it, or look on the back of a rack mount ups where the booster batteries attach), but why the hell aren't they at least using using Phoenix/euroblock sized connectors for these things? The larger offset versions of those would still only take half the space of the old ATX/Molex and leave plenty of power overhead. I realize that the answer is probably that their supply chain agreements with Molex are probably mapped years out, and need thin margins on parts like that.
(for those that aren't EE types/Electicians, each pin and wire in those 40+ coming out of a modern ATX power supply can only handle so much power. The ATX connector came from the pre-millenium AT connector, so as power draw exceeded the output of a single pin that easily handled 80s era computing, they just kept adding more wires and pins to spread the load. Many of those wires are connected to the same rail inside a PSU. If you pull to much on a single wire the pin will spot weld itself and start melting plastic.)
I'm actually somewhat surprised that Case/PSU/Motherboard standards haven't evolved to include busbars in the design as opposed to cable. The losses involved in dumping 600W down 30cm of thin copper wires are severe, and the airflow constraints of cables are what they are.
The PowerMac G5 had similarly crazy power draw; and used bars from PSU to mobo, probably for this reason.
Though the bonkers power consumption for a few FPS that your monitor can't handle, and you can't see even if your monitor could handle, is a use case for expensive kilowatts that I cannot justify (no matter how able to afford).
I feel like that is where the modular PSU market is hurtling towards. And regardless of if it settles on Phoenix type connectors or the XC90s the other guy pointed out, or something similar, The PSU makers can drive early adoption as they can provide the adapters and such to hook up the existing stuff. But with less wires and less melting.
Servers will also be a beneficiary, as the ability to cram yet more power into a 2u box is always welcome, and most server chassis are more tailored as far as mobo footprints, PSUs etc.
That or these GPUs are gonna need their own mains plug.
If you look closely at your tea kettle cord, you should find various national approval body markings (DIN, BSI, UL...) showing that it won't set your house on fire, at least in normal use. I wonder how long before insurance companies start looking at this and requiring the same sort of certification for high power PC accessories?
Obvious icon!
I made the mistake of putting a dual supply rail PSU in a PC with a then nearly current Intel Core 2 Quad Extreme and Nvidia GeForce 9800, and not paying enough attention to the power draw from each rail.
Even though the overall rating of the PSU should have been sufficient at 550W, the amount of power available on each rail was not, and the power supply literally went bang when my son fired up whatever was the latest-and-greatest game at the time (was it BioShock maybe?)
Only ever bought over-spec'd single supply rail 80 Bronze+ (or better) PSUs from reputable suppliers since that time.
Those bonus power connectors went from 4 pins to 6 to 8 already. So one, just doubling the number of pins every two years has left us with more and more cable sprawl. Also the GPUs are already huge and board real estate is already an issue.
So some cleanup was in order but it looks like this is going to be one of those engineering screw ups that people make fun of for years.
There are legitimate uses for this level of processing power but they're few and far between. In this world of energy scarcity, I don't believe making them freely available is morally, ethically or logically justifiable. People that want a GPU that consumes an excessive amount of power should have to make their case and demonstrate that all that carbon will be emitted for the greater good.
I haven't had the chance to play with a RTX4000 series but typically by changing power targets and undervolting you can reduce a GPUs powerdraw by 50% with only a 5-10% loss of performance. But the vendors want that extra 5% performance for the benchmarks so that it what they make the default.
It's just close to the election, and they feel the need to project an imaginary police state on everything. The AC is most likely shilling for them.
Really, if you need to offset massive GPU power drains, put an import duty on them to cover an suitable amount of wind/solar/whatever and dump it back on the grid. Anything more intrusive is going to waste more in paperwork than it will save, and now that GPU mining is pretty much dead, none of the cards on the home market are likely to be driven full tilt for sustained periods.
But that's how spot the trolls and shills. The pass over the simple and obvious in favor of the impractical and and outrageous.
Don't worry about a liberal police states taking over the world. It's a constructed fantasy, and the few liberal politicians that are actually trying have been pretty east to stop. Right now we have totalitarian dictator problems. Once that is unfashionable again, we can move back to smaller problems, assuming we don't go up in the big flash shortly.
https://homeguides.sfgate.com/federal-regulations-toilet-gallons-88640.html
> The Energy Policy Act of 1992, which became law in 1994, mandates a maximum flush volume of 1.6 gallons for toilets manufactured and installed after this date
(the US regs don't seem to be as amenable to searching and quoting as the UK ones, so haven't quoted the specific text)
And there is a healthy market for toilets you don't have to flush thrice to sink a turd here stateside. Also good for those who's sewer service needs a little more flow to make it past the curb. After that it's the cities problem.
One of those above raised a valid point though, why the heck is the loo still hooked to a potable water supply? In reality, in most bathrooms you could swap the tub for one with an impound tank in the base, and recycle enough shower water to cover your toilet biz, especially with a two stage flush. That would have been better to mandate that the 1.6 gal hard limit. Most of the time the bowl just needs a water change, which will save more water even if the full flush causes tidal surge downstream and could sink the Bismark.
The extra plumbing needed is pretty straightforward too. Install a parasitic pump on the incoming water line to the shower/tub to boost stuff from the impound tank out, since there is usually some free energy due to the relatively high service line pressure. You'd need two extra positions for the spigot/shower selector nozzle, a bypass for the impound and a drain setting. Not a huge engineering feat at you can hang a second disk valve on the common spindle with minimum redesign. You could also put a bigger tank on the toilet, one that holds multiple flushes of reclaimed water, and you'd want to hang a secondary float so that the toilet would still work if the shower/bath hadn't run in a while. For bonus points it could have a second drain to a reclamation tank for watering the landscaping, which can probably tolerate a little tooth paste, soap, and shampoo.
All of which you could retrofit into an average bathroom in an afternoon if you had the parts. As they aren't available as an off the shelf kit, the only person I know with that kind of setup was my great uncle, who was a machinist, on his ranch, 30 years ago. The well only put out so much water on dry years, and it saved them rationing toilet flushes and probably saved his marriage.
Those of you that are really into the efficiency gain could also try thermal recovery off of it, but that's getting crazy.
> there is a healthy market for toilets you don't have to flush thrice to sink a turd here stateside
When the low-water-usage loos came over to the UK and sofa dollies got their knickers in a twist about it, engineers pointed out that it isn't the volume, it's the rate it swirls around the bowl that's important. If you're having to flush thrice, you'll be feeling the need, the need for speed (and maybe look at the diet, eh?).
As the old adage goes, it ain't what you got, it's how you use it that's important.
> could also try thermal recovery off of it, but that's getting crazy.
Now you're just trying to encourage someone to try!
I want a GPU that allows me to enjoy my hobbies.
Anybody that wants to complain about the power I use had better sell their car, cancel their mobile contract and stop buying food at the supermarket, because as soon as we start dictating others' energy use they're going to find out that being a hypocrite will be a very unhealthy position to hold.
I'd put my washing machine on at night, except for two things: I have downstairs neighbours who almost certainly wouldn't appreciate it, and I don't fancy having my clothes go all smelly when I leave them in there for six hours before hanging them out.
Smart meters? Well, they don't do anything for energy efficiency (they don't magically make your fridge draw less power), and whilst they might allow you to see how much energy you are using, you can also do the same with cheap plug-in meters for your sockets. If you’re serious about actually saving energy, rather than just seeing the display (as a gimmick), just turn things off when you're not using them, and when it comes to replacing your appliances at the end of their life, look at the energy efficiency ratings of the new ones when deciding which to buy.
All the smart meters actually do is allow the energy companies to take readings (which takes five minutes to do yourself) and turn your power off remotely when it comes to winter brownouts. I may be being selfish, but I'd rather see my power stay on, so I don't end up having to throw away the milk in my fridge when the power goes off for four hours every night.
I'm basically deeply suspicious of smart meters when those promoting them claim they save you energy, or somehow make the energy you are using "greener", rather than being honest about what they do and are for.
Not that you couldn't build something closer to to that idea, but that was never the plan.
You covered the broad stokes pretty well, and I feel for those in the UK that had to endure the boondoggle that was the UK smart meter rollout.
The main play wasn't just to improve the utilities ability to collect data and inflict power though(/s nice bonus though right? /s) it was to saddle ratepayers with a device that cost hundreds or thousands that was slighly less capable than a Rasberry Pi and a couple of relays.
Then the first generation gear was(surprise, surprise) unfit for purpose, insecure, and impossible to update and had to be ripped out.
A smart meter should be able to communicate with devices to provide actual load managemet, not just service disconnects, and could have provided a service that Europeans could use this winter, which is to split rates for essential services like your fridge, thermal management, and important low power stuff like lights, internet routers, and phones.
Oh, and as an aside, why are washer/dryers two machines most of the time? Yeah it makes sense for big families, but for singles and couples you don't do that much laundry. The ventless ones caught on, but take forever to dry stuff as a result rough even for one person. There are fully vented in ones for RVs that run on LP gas that are awesome, but none of the majors sell them for home use. They both literally use a rotating drum. They already make them for RV's. They make ventless ones. Why no vented option?
I want the space back, and I don't like laundromats.
Sometimes using your tumble dryer off peak hours is the worst idea. Using your tumble dryer at night also means that said dryer is depressurizing the house in the coldest time of day. Good for summer, bad for winter.
I try to tumble dry in the most energy efficient time of day, not the 'peak' or 'off-peak'. Sucking in -5C air in the middle of the night into the house, because my dryer is running, is a silly idea IMHO.
I wish I could use one, I really do. I investigated them when I purchased the house / installed [my] new dryer and they won't fit into the space given to the dryer, which is in a dedicated closet sitting besides the washer. When the doors are closed there is only around 6 inches of space between the equipment and said doors, not enough room for the diverter install.
Thank you for the thought though!
Having looked at the connector and the ludicrous 'installation advice' the problem isn't just the small pins but actually the heatshrink/sleeving on the cable.
Where the individual wires exit the connector they are bundled tightly and sleeved with heat shrink which is actually preventing relative movement of them when the cable is bent to fit in cases. Compare this cable with the old 4pin molex on 5.25 HDDs; the wires can move individually thus allowing a lot of flex in the system without pulling some of the pins out, now shrink the pins and add 8 more wires, bundle the whole lot tightly and is it any wonder that bending the cable to make it fit pulls some of the pins out of the housing?
Also the damn thing is only rated for about 40 insertions........ Who specs a connector for that few uses?
"Also the damn thing is only rated for about 40 insertions........ Who specs a connector for that few uses?"
Anyone tasked with designing a connector into a product which, in reality, is only expected to see maybe half a dozen (*) mating cycles, perhaps?
* 1 in the factory during post-assembly test, 1 when the card gets installed by its first owner, maybe 1 more when they decide to swap to a different power cable to improve cable routing, 1 when they do a partial upgrade of their PC, and then 1-2 more once it's served it's purpose with them and they sell it onto a second owner who'll most likely keep it until it dies or gets junked...
I've done similar on stuff I've designed, where the connectors (one of which IIRC, is rated for just 20 cycles) are being used to aid assembly by being a quicker and cheaper alternative to soldering PCBs together and where, therefore, there's zero expectation that the boards will ever be separated again during the lifetime of the product.
As any ful kno, adding more wires in parallel will help carry a larger current - but electricity will always follow the path of least resistance so the wires and connectors between them won't automatically spread the load evenly. To get THAT to happen, any engineer with high current experience will use swamping resistance. This deliberately added load acts as the largest resistance in the line and so negates the differences between each conducting path - it's a common trick we all learnt back in the day: traditional high current PSU's used swamping loads to spread the current out evenly over all the pass transistors.
In short (no pun intended), you can't simply parallel a bunch of wires and hope for the best when you're working with high current because Ohms Law will bite you in the backside. I see no signs of the PSU industry adding swamping to their devices - and there's visibly no indication of swamping on the card (hint: it'd most likely look like a U-shaped bit of wire). The outcome is, I'm afraid, entirely predictable.
I've no idea why you got downvoted, it's obvious that using a lot of small wires is risky, if some of them enconter higher resistance for whatever reason, more load will be on the pins with a better connection, overloading them.
I think two giant pins and maybe a pair of small ones for signalling would be a better solution.
Downvotes? Well, one will be from the person who designed the connector ;)
The other two? Most likely people who missed the physics lesson when resistance was covered. For their benefit, let's have a refresher:
600W at 12V is 50 Amps. Let's assume it's connected via two pairs of wires and hope we get 25A in each.
Now, assume you have 0.01 Ohms resistance in one connector. Not much, is it? A tiny bit of corrosion or even the oil from human hands can add this kind of resistance, never mind the (relatively) high resistance we get with push-on fittings. Ohms law says P=I^2R so power loss in the connector will be 25^2*0.01. That's 6Watts of heat that our one itty bitty connector needs to dissipate somehow. Good luck.
If the vast majority of the power is taken from a 5V rail (and I suspect it is) then we're looking at having to dispose of 36W of heat, all other things being equal. Sure, some will radiate back up the connecting cable - but by bundling all the wire together and then stuffing them into pretty tube the PSU guys aren't helping much there. Going back to our 12V circuit though: the voltage on our bad connection has now dropped so our circuit will demand the majority of its current from the other feed wire, unbalancing the 25A we hoped to share out equally and so causing the other connector to heat up (no connection is without some resistance).
End result: a muted 'crack' sound - some magic smoke if you're lucky - and an engineer going back to first principles. I speak as someone who's built - and blown - a high current PSU because I forgot to swamp the resistive imbalances in the regulation stage.
It may be worse than that. A lot of high powered PCs are put together by custom build outfits who, as part of the service they charge for, look to make the inside of the PC extremely neat and tidy, as this helps airflow (and aesthetics).
So they bend all the cables around sharp corners to neatly tie them out of sight or at least out of the way. I suspect there are a few custom builders sharing this article with their teams right now.
This post has been deleted by its author
"any bending of the wires within 35mm of the connector "may lead to an uneven load across the other wires, increasing the risk of overheating damage.""
Means that the pins aren't sufficiently well located in the housings (to allow for sloppy moulding tolerances) and don't latch on to the receptacles well enough.
So too thin, too flimsy, not big enough.
What's to bet that the engineers originally designed a chonker of a connector, but it was shot down for being too big, and containing too much copper.
OK, the 12V power connector has 12 pins, thus half of them should be power and the other half ground.
So we have 600W/12V = 50 A, over just 6 pins. That's 50/6 = 8.333A per pin.
Those pins do look rather small for something that has to take 8A each. So it's not surprising it's melting.
The best solution would be to up the voltage to 24V or 48V, this will require a new power supply specification.
50A down multiple parallel pins?
Make sure you're wearing a hair net (and beard net!) before plugging this lot together. Muck on the connectors (hairs, dandruff) is going to get carbonised (aka burnt) and/or just create a hight-resistance path. The results - well, guess.
https://corrosion-doctors.org/PrimBatt/li-hazards.htm
" In general, battery manufacturers do not like to use the word "explode" and have invented other terms. One interesting euphemism is "spontaneous disassembly" or "decrimpling". "
Here's an example of a Lithium Thionyl Chloride battery "spontaneous disassembly".
Out of all the lithium battery chemistries, it is thionyl chloride that makes the biggest bang when misused, by the way. It's like a small grenade going off.
https://ntrs.nasa.gov/api/citations/19830026959/downloads/19830026959.pdf
On page 219: "Well, first of all I don't care what kind of vent you put in, the vent you saw was a true detonation. We have other cells where actually the vents did open up and materials were coming out and subsequently they detonated *after* venting." ... "We're talking about complete schrapnelization of the can. For example, when there is nothing left of the can, when you can't even find a tear in it."
That was a NASA guy in a conversation with battery manufacturer employees, who were trying to minimise and deny the severity of the battery detontations. In fact the Duracell guy there *really* doesn't like talking about the safety hazards of such batteries.
And here is what happens when the recycling plant that deals with lithium thionyl chloride batteries goes up in flames.
Honestly, 600W for a video card? Pretty much everyone else in the tech sector is trying to move to more power efficient designs, like ARM in data centers, but along comes nVidia putting out a video card that uses more power on its own than some entire computers. To use this thing at all you probably need at least a 1MW PSU. Which, I guess, makes it easy to figure out how much it's costing you to run your computer. If you're being billed X amount per MWh, you can say your computer is costing you exactly that much every hour it's switched on.
Sounds like the cables aren't beefy enough to handle the power draw demanded of the card. Need to re-design or scale back if one graphics card requires that much juice, a 600 watt power supply easily powers my whole i9 11700k, 32 gig, Geforce 3060 video card, two 1 tb SSD drives, a 1 tb M.2 drive and a mechanical 2 tb harddrive. Ridiculous graphics card power needs!!
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
