Hardware leaker MEGAsizeGPU (opens in new tab) has shared a photograph of a mysterious quad-slot cooler for one of Nvidia's upcoming GeForce RTX gaming graphics cards. However, the jury is still out whether it's for the rumored GeForce RTX 4090 Ti or Titan RTX Ada, which will eventually arrive to compete with the best graphics cards.
The cooler boasts a gold theme, implying it's probably the Titan RTX Ada. The I/O bracket features four display outputs stacked vertically. It's the first time we've seen such a design from Nvidia. It implies that the chipmaker is likely placing the PCB in a vertical orientation for the mysterious graphics card, which has been christened "The Beast." Therefore, the PCB won't be perpendicular to the motherboard like your conventional graphics card. Instead, the PCB will be in a parallel position to the motherboard. Given the vertical design, it'll be interesting to see how cooling manufacturers develop water blocks for the graphics card.
The small sticker on the graphics card's I/O bracket carries the "PG137" PCB board number, coinciding with the information from reputable leaker kopite7kimi (opens in new tab) from July last year. The PCB board number for the GeForce RTX 4090 was PG136. The display connectors on the Titan RTX Ada include one HDMI port and three DisplayPort outputs. The exhaust is significant in size and covers up to two PCI slots.
We can't see the entire cooler, but kopite7kimi has previously stated that the Titan RTX Ada will leverage three cooling fans as opposed to the two-fan configuration we've seen on current Founders Edition models. However, he didn't reveal the placement of the third cooling fan.
"the beast" PG137-SKU0AD102-450-A118176FP3248G 24Gbps GDDR6Xtotal board power ~800WJuly 25, 2022
"The Beast" reportedly features the AD102 silicon that Nvidia currently uses in the GeForce RTX 4090 and RTX 6000 Ada. Nvidia hasn't maximized the AD102 die yet. The GeForce RTX 4090 comes with 128 SMs out of 144, whereas the RTX 6000 Ada sports 142 out of 144. There's a little headroom for Nvidia to squeeze in a GeForce RTX 4090 Ti or Titan RTX Ada, for sure.
The GeForce RTX 4090 Ti will likely share the same configuration as the RTX 6000 Ada (142 SMs or 18,176 CUDA cores). If Nvidia plans to release a Titan RTX Ada, we can picture it with the full AD102 die (144 SMs or 18,432 CUDA cores). Kopite7kimi believes that "The Beast" will sport 48GB of lighting-fast 24 Gbps GDDR6X memory, which will likely be for the Titan RTX Ada. A GeForce RTX 4090 Ti may have a 24GB GDDR6X configuration, like the GeForce RTX 4090.
According to kopite7kimi, the graphics card has a total board power (TBP) of around 800W, which explains the robust quad-slot cooling system. For comparison, the GeForce RTX 4090 has 600W TBP, meaning the Titan RTX Ada has a 33.3% higher TBP. With an 800W TBP, the test board for the Titan RTX Ada allegedly draws its power from dual 16-pin PCIe power connectors (12VHPWR). However, consumers will have to be extra careful when plugging in the 12VHPWR connectors, as we've seen that an improper connection can melt your GeForce RTX 4090.
Despite all the leaks, there's no firm date on when we could see the Titan RTX Ada. However, the recently leaked photographs suggest that the graphics card has been in development and testing in Nvidia's labs for some time. So the chipmaker is probably just waiting for the right time to unleash the beast.
PC need to go to a unified memory architecture with a unified cooling system, like ARM (Apple), Consoles, Mobile and Servers have already done.
Do the performance implications outweigh the upgrade-ability/vendor lock-in implications?
No, I think.
"Shut up and take my money!"
8K gaming?, or very fast workstation?
I really hope anyone who buy this make sure to plug the 2x 16 pin connector all the way in.....
Do you need two PSUs for this ?, or Are there any desktop PSU out there in the market with 2 x 16 pins or 1 x 16 pin + 4 x 8pins?
I prefer the unlimited approach..
who the f thinks this is a good idea? what are you going to need a 220V wall plug?