The Ada generation Titan graphics card is real. Photos have surfaced confirming that it has a quad-slot heatsink and a flip-PCB design that has never been seen before in graphics, but could make cooling the extreme power GPUs quite a bit easier.
Even before the release of the GeForce RTX 4000 graphics, information was circulating that Nvidia was planning to a “monster” card that would carry a fully active AD102 GPU with a consumption (or at least a power limit) of somewhere around 800W. This will be the high-end Titan model of the Ada generation, and now photos of it have appeared that clearly show a four-slot cooler and confirm a crazy design using possibly two PCBs, one of which is turned perpendicular to the card.
The photos, which are very interesting, were published on Twitter by a leaker with the nickname MEGAsizeGPU already known from before. It is about two pictures, on which there is a blank of the upcoming graphic – first itself and then a direct view of the entire graphic. According to MEGAsizeGPU, it is either a GeForce RTX 4090 Ti graphics sample or a Titan model.
According to leaker Kopite7kimi, who usually has the best news about Nvidia’s plans, the latter of the two is true, and we’re looking at a Titan card. That is, a “premium” (which, in marketing parlance, means high-priced) model, which could be the absolute pinnacle of Nvidia’s consumer and gaming graphics offerings.
The blind itself is normally not interesting, but here it is the opposite. The card is clearly almost four full slots thick (minus the few millimeters missing from the last occupied slot, so in theory it could be possible to use the fifth slot on the board – or at least for some low-PCB card). Through the middle two positions, the card has a free space through which part of the waste heat will be ejected, similar to the GeForce RTX 4080 and 4090 “Founders Edition” cards.
4090Ti/Titan exists pic.twitter.com/ceiD89KL0o
— MEGAsizeGPU (@Zed__Wang) January 28, 2023
But the biggest candy (bomb?) are the image outputs. As you can see, the four ports lie vertically instead of being in a row in the space of the top blank. This indicates that the card has the PCB stored in a 90° rotated position. That is, at least part of the PCB, since the card can have two printed circuit boards, with the other stored in the traditional position. It is possible that on one of the boards there are separate power supply components (VRM) and on the other GPU and memory.
Anyway, at least one of the PCBs is placed so that when the card is inserted into the PCIe x16 slot, it lies parallel to the motherboard PCB. It is possible that the card is precisely measured so that its side, or actually the backplate under that rotated PCB, rests on the other PCIe slots on the board below the one in which the card is installed. This would certainly give a very heavy creature better stability and limit the tendency of the card end to bend downwards.
The card in the photo seems to fit exactly to these photos of the passive itself, which have already appeared before. Already on them you could see the “coldplate” located on the side instead of on the bottom of the passive. At the time, we didn’t really believe that the card could really be built in this style, but the reality is evidently like that.
It is not clear if this PCB turned “on top” will support the GPU itself. But it wouldn’t be out of the question. This placement would limit the area available for components on the PCB. But on the other hand, this could make almost the entire passive, which is usually “plugged” by the graphics PCB, be open at the top, allowing air to pass through it freely above the card into the CPU space. From there, it will be drawn out by the system fans on the back of the cabinet and in its ceiling (with a card with similar consumption, the full upper side will probably not be completely suitable). Such a graphics card solution would be much more suitable for cooling than the current one, where the traditionally stored PCB makes passage through the cooler impossible, except for possible “windows” at the end of the card.
Model PG137?
The photo of the cover hides one more nutritious piece of information. There is a label on it that mentions that it is “PG137”, which is the code that Nvidia uses internally to mark various card designs. Kopite7kimi shared some information about a card internally labeled “PG137-SKU0” a while back. According to him, it has an AD104-450 chip in which 18,176 shaders are active, which is 142 of the total 144 SM blocks of the AD102 chip.
This card was supposed to have 48 GB of GDDR6X memory, which might fit with its “Titan” moniker. And it was also supposed to have a maximum card consumption of up to 800 W. However, this is data from July 2022, so it is possible that the specifications have changed in the meantime. For example, it is possible that Nvidia managed to reduce consumption. The originally reported 800 W may not have been the value that Nvidia states as the default TDP, but the maximum Power Limit. For example, the GeForce RTX 4090 is set at 600 W, while its TDP is 450 W. If we were to speculate that the same ratio would remain between the values, then the Titan could have a TDP of 600 W and the maximum Power Limit would be the 800 W. If not to change in specifications…
Resources: MEGAsizeGPU, Kopite7kimi, VideoCardz, Chippell
