Alleged GeForce RTX 3090, 3080 and 3070 Specifications Leak Just Ahead Of Launch

GeForce RTX
The big day is just around the corner—on Tuesday, September 1, NVIDIA will hold a GeForce Special Event in which it is expected to reveal all the juicy details about its upcoming GeForce RTX graphics cards, based on Ampere. We have questions, and NVIDIA presumably has answer. Perhaps spoiling some of the surprises, however, supposed specifications of the next-gen stack have leaked out.

Of course, there have been a million and one leaks and rumors up to this point already, and not all of them align. In theory, though, the closer we get to launch, the more accurate the information should be. But there's also been chatter that NVIDIA CEO Jensen Huang makes decisions at the final hour, so take this all with a dose of skepticism.

That said, Videocardz heard from its unnamed sources that NVIDIA will launch three SKUs on Tuesday: GeForce RTX 3090, GeForce RTX 3080, and GeForce RTX 3070. No big surprise there, but there is a mild twist.

GeForced RTX 3090
Source: @GarnetSunset

According to the leaked info, the GeForce RTX 3090 will arrive with 24GB of memory, the GeForce RTX 3080 will arrive with 10GB of memory, and the GeForce RTX 3070 will debut with 8GB of memory. The memory arrangement is also not surprising, except it is said NVIDIA's add-in board (AIB) partners will eventually be launching a second version of the GeForce RTX 3080 with 20GB of memory.

Here is a breakdown of the reported specifications (and bear in mind that none of this is official):
  • GeForce RTX 3090: 5,248 CUDA cores, 1,695MHz boost clock, 24GB GDDR6X @ 19.5Gbps, 384-bit bus, 936GB/s memory bandwidth, 350W TDP
  • GeForce RTX 3080: 4,352 CUDA cores, 1,710MHz boost clock, 10GB GDDR6X @ 19.0Gbps, 320-bit bus, 760GB/s memory bandwidth, 320W TDP
  • GeForce RTX 3070: ?,??? CUDA cores, ?,???MHz boost clock, 8GB GDDR6 memory @ 16Gbps, 256-bit bus, 512GB/s memory bandwidth, 220W TDP
According the leaked information, the GPUs on these cards are in fact based on a 7-nanometer manufacturing process. So is Ampere as it already exists in the machine learning sector (A100), but there were rumors that Ampere in the consumer space might not be.

To put those specifications into perspective, the GeForce RTX 2080 Ti features 4,352 CUDA cores with a 1,545MHz boost clock, along with 11GB of GDDR6 memory @ 14Gbps on a 352-bit bus, giving it 616GB/s of memory bandwidth. It also has a 260W TDP.

Specs alone do not tell the whole store. Architectural improvements play a big role too, including whatever improvements NVIDIA has cooked up for its RT cores. In short, we could be looking a massive uptick in performance with NVIDIA's next-gen cards. Fingers crossed that pricing does not shoot up as well.