SihastruThe two smaller ones are reinforcement pins for a partial metal plate.

Eh? Where did you see that?

Did you watch the Nvidia video from a few days ago?

www.techpowerup.com/271410/nvidia-shares-details-about-ampere-founders-edition-cooling-power-design-12-pin-confirmed

GungarShort PCB means HBM, that card is going to be expensive!

There's clearly room for 2x12 DDR6 dies on that PCB, so no HBM.

But it's still going to be very expensive, and availability will be abysmal in the beginning.

Even though with all the cool kid's technology, DLSS 3.0, NEW Raytracing tech, I am unsure of the immense heat and power consumptions would be great, in addition to that the price would be very very hefty.
I wouldn't count on AMD launching the top of the cream products, but if they launch something on the lines of 3080 and 3090 and lands on a good price (as in just slightly higher than 3080) with great technology (DirectML, raytracing) I will buy AMD. I hope Nvidia don't push me away

DuxCroVery short PCB. That gigantic cooler makes the card long.

Yep. Water cooling this would make for a dreamy SFF GPU as long as your case has sufficient room for radiators.

I have to admit I'm surprised that the 12th memory channel (channel 6 IIRC from the previous pics) is actually located down by the PCIe slot, even if the spacing is a bit less tight than what it looked like from those leaks. That would mean that a) Nvidia has memory controllers on all four sides of the die, b) there are GDDR6X traces in between the PCIe 4.0 traces (that is going to need some serious crosstalk mitigation!), and c) Nvidia knew they had to make it this way quite early on, given that placing a new memory controller would be anything but trivial once the design was taped out. I would guess that there simply wasn't room for a fourth channel up top, though one has to wonder if that is due to space constraints or signalling issues with the top channels being squeezed in between the side ones already.

Nonetheless, this looks like a beast of a GPU, but with that VRM and cooling solution it's also sounding like a power hog. Looking forward to later today to see how this plays out on paper.

steenThey are double stacked pads.

TheLostSwedeUhm, only one allows for either HDMI or DP, the other three are DP, although there is something going on with the top most port, as there's space for a chip underneath the connector, which is highly unusual.
However, none of them are double stacked, if you compare with this. It also makes it clear that the 3090 doesn't have a USB-C port for VirtualLink.

I'm guessing the bottom one could be either DP or HDMI with the top one DP or VirtualLink, if my guess that those pads are for a USB-C controller is correct. It lacks the Turing-era mounting holes from the USB-C port, but that wouldn't work anyway with the controller beneath the port housing, they'd need some sort of bracket or a custom port. Both of which are feasible, if not cheap. For a card like this I doubt they care about another $1 on the BOM though.

GungarShort PCB means HBM, that card is going to be expensive!

Did you somehow overlook the 12 GDDR6X pads on both sides of the PCB?

kiriakostHmm about 20 VRM's stacked closely to its other.
Next thing we need to know this is : How many layers and thickness of the PCB.

A large number of VRM's this translates that the GPU has the ability to turn off engines that the software can not use them.
Its like an active power saving mechanism.

Yeah, inductors are right next to the GDDR6 chips.

TheLostSwedeUhm, only one allows for either HDMI or DP, the other three are DP, although there is something going on with the top most port, as there's space for a chip underneath the connector, which is highly unusual.
However, none of them are double stacked, if you compare with this. It also makes it clear that the 3090 doesn't have a USB-C port for VirtualLink.

iO4x DP with the option of swapping the bottom one for a single HDMI port.

Yeah, quick edit on my part. Optional DP/HDMI, also means you can stack. FE is single I/O row with full width vent. AIB versions available with 3xDP + 2xHDMI stacked at the bottom.

TheLostSwedeSadly it's a bit too blurry to see, but it seems to be a 12 layer PCB if you look at the rear picture of the PCB and up top there's a yellow-ish square where I'm guessing the PCB layer number is present and it looks like a 12 to me.

It's probably batch/QC, but not enough res to be definitive about many things. Previous Titan/Quadro as well as higher end AIB 2080ti used 12 layer stackup. The rumor was for 12 layer, odd RAM placement, & no USB-C.

steenIt's probably batch/QC, but not enough res to be definitive about many things. Previous Titan/Quadro as well as higher end AIB 2080ti used 12 layer stackup. The rumor was for 12 layer, odd RAM placement, & no USB-C.

Uhm, you're aware those things are often for showing all the PCB layers right? Most motherboards have a little "window" that shows the layers. It might not be the case here though, as it's on the break-off part that's just there to guide the PCB through the production line machines. Even so, I think we're looking at a 10-12 layer PCB for the 3090, simply due to the super complex design.

TheLostSwedeEh? Where did you see that?
Did you watch the Nvidia video from a few days ago?

Yes, that's where I saw it. The two small circles near the connector tower correspond to the extra two smaller holes you see on the PCB. I believe it's a variant micro-fit connector that is "rail mounted" (if I remember correctly), each row (GND, +12V) is mated to a single vertical rail (in this case). Each rail has two holes on the PCB for added rigidity (plugging and unplugging can put a lot of pressure on these soldered mount points). It is not clear if they'll use the metal reinforcement column that corresponds to the smaller mount points). I guess we'll find out in an hour or so :)

The two larger round holes near the edge of the PCB correspond to two clamps that are part of the plastic molding.

too close to GPU , I think memory temp will rise to 90'c.look at Navi card.

SihastruYes, that's where I saw it. The two small circles near the connector tower correspond to the extra two smaller holes you see on the PCB. I believe it's a variant micro-fit connector that is "rail mounted" (if I remember correctly), each row (GND, +12V) is mated to a single vertical rail (in this case). Each rail has two holes on the PCB for added rigidity (plugging and unplugging can put a lot of pressure on these soldered mount points). It is not clear if they'll use the metal reinforcement column that corresponds to the smaller mount points). I guess we'll find out in an hour or so :)

The two larger round holes near the edge of the PCB correspond to two clamps that are part of the plastic molding.

It would appear you were mistaken. The connector uses two solid pieces of metal to connect to the PCB from the looks of it and the ground pins seem to have an additional stabiliser connected to the two additional, smaller holes. No metal shroud/plate whatsoever.

Wow this NVIDIA GeForce RTX 3090 Ampere Bare PCB looks outstanding and I think Nvidia did a really good job on this graphics card model. I am a designer of pcb and also an engineer and I created similar pcb in my prototype circuit board project. I can not even imagine how good would be this graphics card in games and video rendering. Also, it is very interesting how would Radeon graphics card look like as I made a prototype for a new Radeon card as well. Hope it will be better in terms of performance as Nvidias cards are most of the time much more expensive than Radeon cars.