NVIDIA Titan ADA: prototype would have 48 GB DDR6 and 18,432 CUDA cores

A leak from user FluxRBLX on Reddit shows a prototype NVIDIA card with 48 GB DDR6 and 18,432 CUDA cores. He did not provide a link, only images and, from the information, it is not possible to know whether the card is intended for games or workstations.

The images show a prototype card with an NVIDIA GPU on an RTX 5090 Founders Edition PCB. The card comes with 12 GDDR6 modules on one side and another set of 12 memory chips on the other side, totaling a combined capacity of 48GB when adding the 24 modules as indicated in the GPU-Z screenshot.

Related news:

Also depending on the leak, the board features 18,432 CUDA coresa configuration far superior to the RTX 4090 (which has 16,384) and even superior to the RTX 6000 ADA (a workstation model), which comes with 18,176 CUDA cores.

Based on the specifications, the card would have a base clock of 735 MHz and a boost clock of 2,490 MHzwhile the memory clock would also be lower than that of the 4090: 2,250 MHz, offering a speed of 18 Gbps.

The bandwidth would be lower than that of the 4090, with 864 GB/scompared to the RTX 40 flagship model, which exceeds 1 TB/s of bandwidth.

TITAN Model

The model is undoubtedly a prototype, featuring several measurement points, DIP switches and secondary boards for engineering purposes.

And the same user responsible for the leak reportedly has a GTX 2080 Ti prototype for sale. Yes, GTX, because the card uses a pre-production GTX cooler from before NVIDIA switched to the RTX naming convention. And the information reinforces that he has access to prototypes.

It’s possible that NVIDIA initially planned to use a four-slot prototype cooler from the TITAN ADA for something more advanced, but the prototype only offered 4090-level specs.

In technical terms, by combining the leak prototype with the cooler from the TITAN ADA model, it would be possible to build a robust card.

NVIDIA

Currently, NVIDIA does not need new cards to deliver market-leading performance. And Jensen Huang himself said that NVIDIA’s AI chips advance much faster than Moore’s Law.

During this year’s CES, NVIDIA presented DLSS 4 technology, which brings AI benefits to cards with Blackwell architecture: RTX 5090, 5080, 5070 Ti and 5070. Highlights include multiple frame generation, which can generate up to three additional frames for each traditional frame.

Font: Reddit.