VideoCardz (opens in new tab) has the scoop on Nvidia's forthcoming trio of GeForce RTX 30-series (codename Ampere) graphics cards that will vie for a spot on our list of Best GPUs. Even though the publication has a solid track record, it's prudent to approach the specifications with caution nonetheless.
The report says Ampere will debut with Nvidia's second-generation Ray Tracing (RT) cores and third-generation Tensor cores. The Turing architecture incorporated RT and Tensor cores, so it's natural that Ampere follows suit. There was no way that Nvidia would abandon the RT and Tensor after only one generation, considering that they are two attributes that differentiate current GeForce offerings from AMD's Radeon graphics cards.
The A100 is a product of TSMC's 7nm FinFET process node. VideoCardz claims that the anonymously-obtained information pointed to the usage of the 7nm process node on consumer graphics cards as well, but the news outlet was unable to confirm it. Consumer Ampere will likely leverage PCIe 4.0, which is predictable since the A100 GPU (opens in new tab) supports the snappy interface. In regards to display outputs, Ampere presumably has native support for HDMI 2.1 and DisplayPort 1.4a outputs.
Moving on to the specifications, the GeForce RTX 3090 is rumored to come with the GA102-300 silicon, which brings 5,248 CUDA cores to the table. The graphics card sports 24GB of GDDR6X memory that runs across a 384-bit memory interface. This design amounts to a memory bandwidth of 935.8 GBps. The graphics card's TGP (total graphics power) is estimated at 350W, meaning custom models will likely rely on a pair of 8-pin PCIe power connectors. As Nvidia has revealed recently, the Founders Edition will only need a single 12-pin PCIe power connector (opens in new tab).
Nvidia GeForce RTX 30-Series Specifications*
|Graphics Card||CUDA Cores||Memory Clock||Memory Capacity||Memory Bus||Memory Bandwidth||TGP|
|GeForce RTX 3090||5,248||19 Gbps||24GB GDDR6X||384-bit||935.8 GBps||350W|
|GeForce RTX 3080||4,352||19 Gbps||10GB / 20GB GDDR6X||320-bit||760 GBps||320W|
|GeForce RTX 3070||?||16 Gbps||8GB GDDR6||?||?||220W|
*Specifications are unconfirmed.
Going down the Ampere product stack, the GeForce RTX 3080 could arrive with 4,352 CUDA cores and 10GB of GDDR6X memory. Strangely enough, the GeForce RTX 2080 Ti (opens in new tab) also has 4,352 CUDA cores, turning the GeForce RTX 3080 into the new GeForce RTX 2080 Ti. There's a possibility that Nvidia's board partners may eventually launch a second iteration of the GeForce RTX 3080 with twice the memory.
Assuming that Nvidia retains the same memory layout on the GeForce RTX 3090, then the GeForce RTX 3080's maximum memory bandwidth comes down to around 760 GBps. VideoCardz's sources claim that the GeForce RTX 3080 is rated for a 320W TGP so it should also require two 8-pin PCIe connectors, just like the flagship model.
A bit of mystery still surrounds the GeForce RTX 3070, though. The graphics card may debut at the end of September. The rumored specifications include 8GB of GDDR6 memory and a 220W TGP. The exact speed of the memory chips is unknown, but VideoCardz thinks that they might come clocked at 16 Gbps.
It has become common practice to see specifications leaks right before a looming product launch, especially one as big as Ampere. Nvidia's GeForce announcement (opens in new tab) on September 1, however, will finally shine light on the matter. The wait is almost over.