Hello.
I have a patent pending idea on how to solve the cooling problem of GPU, specifically when one GPU overheats in SLI/CrossFire array, while other GPUs are relatively cool.
Here I attach a file with the short description of this idea. Disclosed information is not confidential.
I would like to develop a non-provisional patent and license it together with AMD.
Please, feel free to contact me if you are interested:
First of all, thanks for your critic, because pointing the at the problem is the first step to solve the problem.
Now, answering to your critic. I totally agree that there are better GPU cooling systems available on the market.
But, for some reason most of the reference cards are made with blower style cooling system. And believe me, engineers who design reference coolers are not stupid and fully aware about blower design issue, but there are certain limitation that make them use blower design.
I belie reference cards use blower style coolers, because it should work in worst case scenario, such as small computer case with very bad or no air flow organized through the case. Reference card should work in this conditions.
Lets imagine you have such computer case and you bought board partner GPU with nice cooling system, where several axial fans blow on the heat sink from the top (it takes air from inside the case and bring hot air back to the case). In case there are no sufficient air flow organized through the case, your GPU very quickly will heat up air inside computer case, overheat itself and also other components of computer. This is the reason why blower style cooler exist and used as reference card, it is safe.
Blower type cooler provide certain air flow through the computer case by itself (takes air from the case and throw it outside), even with bad computer case cooling it will work (still will overheat, but will not effect other computer components).
Compare blower style cooler (work in any case) and cooler with axial fans (works if you have good cooling in the computer case) is same as compare HUMMER (drives anywhere but slow) and Ferrari( (very fast, but only on good roads).
I believe this post will be better at AMD Forum General Discussion since it doesn't involve any particular issues with Graphics.
It sounds like a good idea except all the GPU cards must be in the Motherboard's PCIe x 16 slots next to each other to join them together. Many motherboards have the PCIe x 16 slots stagnated with other PCIe non x 16 slots in between.
Also you will need to make an enclosure that would fit all the different GPU cards together for achieving a united cooling effect for all the GPU cards at the same time.
But it does sound like a very good idea if you are able to actually make it work in the future.
Here is the PDF File that the OPS attached. It is a small so I was able to show it with two images:
Hello elstaci. It is very nice to see that I am not only one who think this good idea.
And, also you pointed at very good problem with graphic card spacing.
I made provisional patent on this idea, where I address this issue, but post here whole patent document of 20 pages is not best idea. I decided to keep original post short, to help ignite the interest and hopefully help me to find partner to help me develop non-provisional patent and license it.
For the case with motherboard spacing, where two graphic cards in x16 slots are not attached to each other. One can use additional adapter to thermally join them together. For example this adapter can be U-shaped bent vapor chamber, where straight "branches" of the named vapor chamber attach to the graphic cards vapor chambers. It can be fixed directly to the graphic cards or occupy non x16 slot in between graphic cards.
It is not optimal solution, but I think issue with spacing you pointed is not very common. Also, question with this adapter distribution to the customer, but this is question to sales guys and I am engineer.
Regrading enclose that would fit all different graphic cards, I am not fully understand what you mean here. Why you use different graphic cards in your CrossFire array? And I don't see why one need such enclosure. It would be nice if you could describe it more, there are might be some issue that I didn't thought before.
I think you mean that proposed idea cooler is slightly bigger, due to the bend around PCB. But it didn't take too much space and should fit in most of the computer cases, if that is what you mean.
Both AMD and nVidia GPU cards are made by various Manufacturers. So an MSI RX 580 GPU card will not be the same size and shape as a Sapphire RX 580. So you would need to have a connecting enclosure to be able to fit and adapt to various GPU card sizes and Cowls. Some GPUs have a small Cowl with one fan while others may have three fans and a large Cowl cover.
As for PCIe slot spacing it is very common on motherboards to have the x16 slots stagnated. Possibly on a Server or mining motherboard it may have all the PCI x16 slots together.
MSI MOTHERBOARD EXAMPLE SERIES:
GIGABYTES GAMING MOTHERBOARD SERIES EXAMPLE:
I was thinking maybe installing a watercooling GPU Chip cover like the ones used for CPUs and then having only two small rubber water hoses going from one GPU card to another as in a daisy chain.
Radiator --------- to first GPU Chip cooler --------- exhaust watercooler hose going to second intake of GPU Chip --------- etc and then exhaust from last GPU Chip water hose going back to the Oversize Radiator and fan.
Don't know if it is possible to install a Watercooler GPU Chip cover apparatus on a GPU Chip with heatsink and fan cowl. But if you have small hoses going between GPU cards it will be less intrusive and Space saving inside the Computer Case.
Also with the adapted Coverings for the various GPU installed, wouldn't it obstruct air flow to other parts of the motherboard components installed?
Thanks for the detailed answer.
First of all, my goal is to license this idea to GPU manufacturer. And this idea is a replacement for reference blower cooler.
I sincerely don't think this is a good idea for a replacement cooling system. But it is definitely better than cooler of the reference cards.
Probably this forum is not a place where I can achieve my goal, but it is extremely useful to see an opinion of the potential future buyer.
So let me try to win your heart for this idea.
Regarding spacing, as you showed, it might be an issue. I think this motherboard spacing is made to address this issue of one GPU overheat in SLI/CrossFire array.
Here is the picture of the solution I have described in the previous post.
The red part is one more U-shaped vapor chamber that thermally joins two GPUs together if motherboard spacing doesn't match.
There might be various sizes of this vapor chamber for different motherboards, but it might be difficult to arrange distribution of such an adapter.
Regarding your idea of water cooling, let me give you some critics. As I understand you want to install GPU water coolers in line, so water from GPU 1 go to GPU 2.
This idea might have the same problem of not equal cooling. GPU 1 will have a nice cold water, GPU 2 will use water after GPU 1(so it is not so cold) and its temperature will be higher. If you have GPU 3 and/or GPU 4 further in line, the situation will get worse. But water has very high heat capacity, compared to air, so this temperature rise of the water will not be so high and might be not an issue.
To address this issue you will need to increase water flow rate, so water doesn't worm up so much at each GPU cooler, but this effect will be there anyway.
To increase flow rate, you will need to use a bigger water pump. Also, it will result in a higher water pressure in the pipe, which increases the risk of leakages. To address it you will need to make your pipes diameter bigger.
It might be better to make GPU coolers in parallel. But also, it needs a higher water flow rate and a lot of pipes.
As you can see it is not so easy and need to be carefully calculated.
Now, I am back to my idea and would like to show you one more cool feature
You can install the additional cooling system, liquid or air cooled, to the free surface of GPU vapor chamber. The picture above shows a simplified liquid-cooled cold plate. This cold plate is connected to the chip via the vapor chamber of the proposed cooling system, so lost of efficiency will not be so much, also vapor chamber would help to distribute cooling effect to other components such as memory and voltage regulators. And you don't need to remove the original cooling system and brake the warranty.
So here is the question for you. If AMD or NVIDIA would sell reference GPU with such a cooling system, would you prefer it to GPU of other manufacturers with a not blower cooling system design?
I think the idea is Good. Nice man!
Thanks, man. By the way, does anybody know how I can submit this idea to AMD for review officially?
This Forum is moderated by AMD Employees:
Ray_M is Technical Support Engineer 2
AMDMATT is Technical Support Engineer
Or you can ask through AMD EMAIL SUPPORT directly and see what they say or suggest: Email Form
Already done AMD email support. Thanks for advice
That's decidedly a bad idea. Not only are the coolers absolutely terribly inefficient blower coolers, airflow is terribly restricted meaning they'd be as loud as a 747, and this would only work with reference edition coolers as board partners all have their own much more efficient designs which custom fit to their PCBs. Not only that, Crossfire and SLI are all but limited to two cards with interest even in two diminishing greatly.