Basically, if all the information is correct, Nvidia hasn’t chosen a bad approach with the solution of a reference board that is as short as possible, which elegantly allows an air cooler to allow the fans to blow freely over a larger area without the air just landing on the board and then being pushed down towards the motherboard or up towards the side panel. If it weren’t for the extremely high waste heat generation in a confined space, whereby the entire energy converted into heat of up to 320 watts must be dissipated quickly and efficiently.
This may also still work, but it becomes quite sporty when you consider the heat distribution on the possible reference board. Because exactly this is anything but symmetrical, if you draw the right conclusions based on the information, basic knowledge and logical conclusions that have been leaked in the meantime. Even if the GPU is compatible with the approx. 230 watt maximum is still located relatively in the middle, the rest is grouped somewhat top-heavy, as the following, of course very simplified graphic shows:
The majority of the memory on the front side is omitted with a share of approx. 20 Watt of the up to 30 Watt also to the upper half of the board. In addition there are up to 4 voltage converters for the memory and probably the whole blocks of NVDD, i.e. the supply voltage for the CUDA cores and of course at least half the GPU. The rest of the GPU, the significantly lower load for MSVDD, the remaining memory modules and some partial voltages are then allotted to the lower half of the board.
So far, so crooked. Air coolers can of course tap the GPU waste heat directly, but at the latest when it comes to the memory modules it becomes critical. This is because it is located very close to the GPU due to the guaranteed shorter signal paths. Normally, pads are used here directly to contact the heatsink or you use the mounting frame that still encloses the heatsink. Other constructs, such as cooling frames bolted to the PCB without direct reference to the main cooler are probably completely out of the question here for reasons of space.
In addition, there are the very close voltage converter areas to the left and right of the GPU. Here, the supply lines in the multilayer board also lead from the converters directly under the two vertical memory blocks, each consisting of 4 modules, to the GPU. This should also mean a total of about 10 watts, which cannot be absorbed directly by the cooler and cooled directly. Now it would be possible to actively cool the components on the way, such as coils and capacitors, but this would create further problems.
With the coils you have to take care that they don’t get too cool, because they buzz and whimper more, with the capacitors the whole thing only works with SMD versions. Cans cannot be cooled in this way. However, since Nvidia will almost certainly rely on very good power stages, which almost completely combine all elements of a voltage transformer in one package, there will still be very small and yet hot areas that need to be actively cooled. I calculate with up to 18 watts for the upper half of the board, the rest ends up at the bottom. This also heats up the board well, of course.
If you believe the rumors and circulated episodes from the circle of suppliers, then e.g. AVC as one of the potential cooler manufacturers for the Founders Edition complained about major difficulties in the technical implementation. Also some of the AIC partners might wear some more grey hair on their heads, if there were any left. But the see al things on the right track everywhere, probably also because Nvidia is pushing quite hard to keep the dates in the first half of September.
Water cooling manufacturers will also have a lot of fun with the current reference design, because actually you would also need a channel on the top side, where the terminal is almost traditionally located. However, there is not much space left for exotic water features, because it is crowded in the smallest of spaces. It will be interesting to see which manufacturer has best met the requirements, as a secondary use of existing coolers is clearly out of the question this time. You don’t even have to be a prophet.
As a reminder, I have already published an estimate of the waste heat:
|Estimated Power Consumption / Losses|
|Total Graphics Power TGP||350 watts|
|24 GB GDDR6X memory (approx. 2.5 Watts per module)||-60 watts|
|MOSFET, Inductor, Caps NVDD / MSVDD (GPU Voltages)||-26 watts|
|MOSFET, Inductor, Caps FBVDDQ (Framebuffer Voltage)||-6 watts|
|MOSFET, Inductor, Caps PEXVDD (PCIExpress Voltage)||-2 watts|
|Other Voltages, Input Section (AUX)||-4 watts|
|Fans, Other Power||-7 watts|
|PCB Losses||-15 watts|
|GPU power||approx. 230 Watts|