Yeah - sorry, but people should NOT listen to you. A ton of current and past gen cards use vapour chambers - ex AMD 6800XT 6900XT, aib RTX 3000 cards, RTX 2000 FE cards. These are just a few examples - vapour chambers are used A LOT. But they are expensive so manufactures prefer large solid copper cold plates when possible.
EX. aorus master
https://global.aorus.com/blog-detail.php?i=805
RTX 2080Ti FE
https://www.techpowerup.com/review/nvidia-geforce-rtx-2080-ti-founders-edition/4.html
I really don't feel like addressing few of the other plain wrong assumptions in your post. It's kinda funny how 100% sure you sound about what you posted but... it's just not correct.
So...here's the point that I ask you whether you read about any of those linked articles or not...
I'll wait momentarily. I'll also cite a comment I made. I'll bold the bits that matter:
"...Well, let me be practical. Heat chambers are excellent for something with a lot of little heat sources. It evens these sources out, and by virtue of a huge difference in surface area can transfer heat well even if the distance isn't huge. Read: those older amongst us will remember the Sapphire Toxic line using vapor chambers...then suddenly remember that nobody else has in years. That's because we now have 2+ slot GPU cards...and the
much more economical and efficient heat transfer methodology with that space is to use heat pipes and a large fin array..."
Let me also cite what you may be missing, and not understanding. I'll admit that I didn't put it glowing lights...so you may need some help.:
"...Read: those older amongst us will remember the Sapphire Toxic line using vapor chambers...then suddenly remember that nobody else has in years...."
"...
TL;DR:
f you still don't get it, answer me a simple question. If vapor chambers are amazing why then is the last time they were used in the HD4870 line from 2008/2009?..."
It's amazing that when you take an immensely complex issue, and TL;DR it, you might lose definition...but salty people always start with "Too Long, if you can't say something in a sentence it's beyond my attention span."
Let me wrap this by asking you what should be an easy answer. What is the difference between a vapor chamber and a heat pipe? It's stated above, but the difference is only geometry. Both are chamber filled with low pressure fluids, that transfer heat by phase change, which require enough energy to induce that vaporization before the physical motion of the fluid to another surface transfers that energy. It's kind of like angel hair and macaroni are both pasta, but used differently due to their geometry.
I'd love it if you could cite the last time you've seen materials about the benefits of a vapor chamber cooler...as the above quote highlights. I'll admit maybe I should have explicitly stated that those of us remember the selling point for those cards was the vapor chamber coolers, and how they were infinitely better than the then standard gigantic stacks of metal fins...consider my point invalid because either you weren't there, or perhaps I assumed too much of people remembering them being a huge selling point. Also, perhaps the simplification for TL;DR was too much. We'll see soon.
Now the fun bit...you shouldn't trust me.. Cool. I like the "I'm 100% sure that other stuff isn't right...but then again, I'm not going to even bother." It's that fun argument that shows you know what was going on...without ever having to really get it.
Let me end with the startlingly idiotic point that I think needs to be the end here.
The vapor chambers you've cited are on a few cards...did you maybe consider the specifications cited in the rest of those articles? Perhaps the 754mm^2 die size. Perhaps the mechanical fusion of the copper cooler to the enormous aluminum heatsink to make that vapor chamber possible? No... Cool. It's not like the OP has already stated that they think they can just glue a fin block onto the cold side. It's also not like there were regular reports for the 2080 FE of overheating when there was any issues with the thermal paste existing...highlighting that the engineers on those cards had a razor thin margin of performance that is silently supported by automatic downclocking during thermal throttling.
With the 3000 series Aorus cards, did you look at the pictures, or just do a google search for the words? I'm really asking because if you look at the picture the large copper plate is a chamber...but the only fin block is connected to it via a series on copper tubes that extend out into an aluminum fin block. This is reinforced in multiple pictures...but apparently the point was "we use a vapor chamber" and thus this other guy must know nothing.
Let me do some number free math, to demonstrate the marketing is not reality, and you seem to be missing the hype. Let me also state that you're welcome to argue finer and negotiable points. My favorite would by the 70 degrees C, because that's literally something that can be specified...and could theoretically be changed.
The above is a graph of phase of water at a given pressure and temperature. Note that if you are specifying a phase transfer cooler the pressure and volume of water are usually derived...because what you know is the target sustained temperature and energy transfer rate from the heat source. More on that momentarily.
Now, the question is how much energy transfers. That's trivial...you actually have a target temperature and mass of water. The specific heat of water is variable with pressure...so the old engineering toolbox is a fantastic way to figure things out:
Engineering Toolbox - Specific Heat of Water
So...if your TDP is expressed in watts, which is Joules/second, and you've got an amount of energy that can be transferred through a phase change, it's trivial to calculate the mass of water that needs to phase change over time for that energy to be transferred. This needs to be more than doubles, because liquid is vaporizing to gas on the hot side, gas is moving to the cold side, gas is condensing to liquid on the cold side, and liquid is moving back to the hot side via capillary action.
Sounds almost simple...but then you have to account for a bunch of other stuff. The sensors inside of a chip are right next to the heat source, and it has to transfer that energy through three layers (transfer media, IHS, transfer media) before it hits the vapor chamber. This transfer is based upon conduction...so it's based upon the thermal conductivity of each layer, its thickness, and the delta between each layer. After you calculate that, you've got an effective amount of energy that is being dumped into the chamber...which is why the chambers (in my past dealings) were specified at around 70 degrees C when the components ran at 80-90C.
Where all of this is further complicated is that any decent engineer will tell you that a vapor chamber relies of fluid flows...and calculating the complex nature of energy transfer in a flow is...a nightmare. It's most often done on a computer using iterative calculations to model, and where we get those nice color coded pictures of relative temperature that appear in so much of the media. So...yeah.
Now that you've somehow gotten an engineer to provide you with the thermal transfer properties of the entire chip, and another to model the non-trivial energy transfer, you can design a vapor chamber.
---If it isn't sinking in, this is why companies take years to design things, have their own thermal management engineers, and don't just slap on a cooler rated for "xxx TDP" anymore... unless they're the 5xxx series from AMD. Good lord, those cards sucked...and the updated drivers were a saving grace.---
Now, what exactly is the wrong statement from before?
Well, if the vapor chamber overheats it is an insulator...so not that. Instead of soaking energy by phase change at a constant temperature (cause phase changes do occur at a fixed temperature...and if you don't believe that ask how a glass of ice water is all at 0C or 32F until all the ice melts), the phase change literally doesn't occur and energy transfer via convection inside the low pressure chamber is negligible as after phase change temperature you only have heat capacity and temperature increases...with the nearest analog being the vacuum mugs that keep beverages cold or hot for hours by having a low pressure region between an inner and outer lining. There is some transfer by conduction via the chamber walls...but it's nearly trivial given almost no cross sectional area.
You could argue the vapor chamber usage...but I'd then argue that vapor chambers are not advertised as premium solutions and are only used as an expensive and less efficient option...as demonstrated by virtually all third party boards using fans blowing over fins with heat pipes embedded in them...including those options that bastardize the use of "vapor chamber" by actually using heat pipes feeding into a chamber. Gotta love Nvidia and AMD default blower designs...which is the first thing to go when you purchase a more premium card and they can spend money on increasing clocks and decreasing noise...
You could argue that 70 degrees is wrong because you know better. I'd then retort that 70 degrees is what I've seen in similar cases where an NDA is involved, and you need power transfer to not break down. It's arguable and probably somewhat of a "special sauce." That said, items with 99.9995% up-time generally are specified well.
I'll poke fun at marketing one last time. Let me explain this in words...that the tech community can understand. Is CISC worse than RISC? Theoretically, because RISC is more efficient, the x86-64 processor inside the desktop or laptop you are currently viewing this on is a fraction as good as your cell phone. There's marketing material on this website that says the same...that China will soon release processors that are 50% more efficient than anything Intel or AMD can release. They fail to mention that CISC vs. RISC instruction set. Likewise the average circular tube fitted into a plate makes direct contact with a surface only fractionally, meaning that with the same effective surface area the average plate makes contact with a much larger than the area that a pipe can contact, facilitating much better thermal transfer... It's almost like looking at a brochure to sell you things might not be an objective source. It's also pretty goofy to think that memory chips might need a cooler specified like the 215 watts of an entire card...when they've already got one installed and it's almost entirely dedicated to the GPU instead of all of those chop packages around it...when a crappy piece of thermal pad is sufficient to their design...and thermal pads have 25% or less the thermal conductivity of pastes.
All of the above said, you win. I implore you to contact an engineer at the cited company, explain what you want, and get a quotation to have a custom designed solution that actually meets your need. Let me state that you're welcome to tell the OP that this will do something more than a trivial change... That said, I will stake about $100 today. That is, I will pay $100 if once you talk to the representative, you get something designed, fabricate it, and come in under $40.
Why exactly do I pull that number squarely from the ether? Well...it isn't. It's about $18 for self stick GPU copper cooling fins...and I'm giving you the money for a set of fins, and more than doubling that...so that you can see exactly how much engineers will laugh when you tell them the order quantity is 1...and the budget is $100. It'll either be a laugh, or a dumbfounded dismissal. Meanwhile I'll have ordered the fins, and be waiting for them to arrive...so that we can decrease the temperature on a component that is designed to run that hot...
Amazon - Copper GPU memory heat sinks
I'll at last close with a bit of fun. Why not just slap a cooler for a 3080 onto a 3060ti? Surely a cooler rated for 300 watts is more than capable of 180 watts of TDP. Well, not really. The much higher TDP cards usually do two things special. They've got enough thermal mass to take momentary temperature spikes, and because they never have a substantive delta temperature between components and environment the coolers do very little. As such, that cooler is an expensive way to increase costs without delivering any real potential improvements. Most people don't get this...and it's how a company can design one 3070 cooler, reuse it on the 3060ti, and change a $100 or more premium. While seeing a memory chip hit "high" temperatures isn't ideal, it is what they tolerate.. Of course, what would Nvidia know about designing GPUs after 2+ decades anyways.
Anecdotally, I've got an x79 motherboard with one component hitting 90 degrees daily. It's done so for 10 years. Sometimes what looks bad is a design meant for it.
), because even the Arctic (ex Arctic cooling) MX-2 is head and shoulder (and cheaper) above AS5 and MX-4 also cost less for more.
)
)
erformance ratio, and most frustratingly requires real math to make work. Vapor chambers are the beta of the technology. Heat pipes are VHS. Using more modern examples HD DVD versus Blu-ray. Either way, it's not a reasonable expenditure unless you can do real math...and if you're asking about this on a forum without being aware of how to spec the "fin side" you should really consider an alternative solution.
