We reported that leaks revolving around the GeForce graphics cards from Nvidia show that Nvidia is looking towards improving the traditional rasterization as well as the new Ray Tracing rendering. The new Ampere architecture which will be manufactured on the 10nm node by Samsung foundries. As reported earlier, the HPC graphics cards and the GeForce graphics card may feature different manufacturing processes and eventually different architectures.
The GeForce graphics cards will feature one of the 5 GPUs based on the Ampere architecture. The leak specifically features the specifications of the flagship graphics cards. Since Nvidia announces the top-tier hardware first and the mid-end and the low-end graphics cards are released later, it adds to the credibility of the leak. The “side SKUs” (SUPER family) are released depending upon the competition.
1 Surprisingly, the next gen chip will be based on Samsung 10nm node 2 Not very solid about what GA102 really got, may struggle to be 40% above 2080ti, but bound to be under that number if compared against full bore TU102 3 SLi will be available to GA102 ONLY 4 RTX for everyone pic.twitter.com/b5YAKmOmgf
— Yuko Yoshida (@KittyYYuko) March 11, 2020
Though it is still unclear the presence of GA 103 in the traditional GA 102, GA 104, GA 106, and GA 107 GPU lineup, it is too early to assume that Nvidia may end up utilizing two different GPUs in different SKUs of the xx80 graphics cards. Nvidia will also make structural improvements in the RT cores and Tensor cores design so that even low-end GPUs could deliver on the RTX in every graphics card of the family promise. Ampere architecture will also improve shading performances.
It is still unclear how the low-end GPUs like the GA 107 will handle the intense Real-time ray tracing scenarios and how will the additional cores will be implemented in the graphics cards since these graphics cards get the required power from the motherboard. Additionally, the ampere architecture will feature PCIe 4.0 interface across the board and support for SLI through NVLINK only in the graphics cards based on the GA 102 GPU
GA 102 inside the GeForce RTX 3080Ti
Nvidia GeForce RTX 3080Ti will be the flagship graphics card of the Ampere series. It will be replacing the RTX 2080Ti. It will feature a trimmed-down version of the GA 102 that will have 84 of the SMs enabled.
It should be noted that GA 102 could only have 84 SMs. Nvidia may end up releasing another GPU for the TITAN graphics cards. Traditionally, the XX 102 GPU is fully enabled in the TITAN graphics cards. Since these are only the early rumors and Nvidia is keeping the information secured, please take the specifications with a grain of salt.
Nvidia will retain 64 CUDA cores per SM regime, and hence the RTX 3080Ti will have 5376 CUDA cores. It means it will have 16% more CUDA cores than the $2500 TITAN RTX graphics. The GPU is said to be 40% faster than its predecessor. Such a generational leap seems a little far considering the current specifications. On the other hand, we do have architectural improvements, so it is still very early to jump towards the conclusion.
It will feature 12GB of GDDR6 memory over a 384-bit bus. The frequency of the GDDR6 module is rumored to be equal of 16Gbps making the total bandwidth equal to 768GB/s.
GA 103 inside the GeForce RTX 3080
The mysterious GA 103 GPU will reportedly be featured in the RTX 3080 graphics cards. Instead of using the flagship GPU in the xx80 graphics card, Nvidia will release a separate GPU. The decision aligns with the future release of SUPER SKUs since Nvidia needs GPU leverage to release more and more SKUs.
It will feature 60SMs, which makes the total CUDA core count will be equal to 3840. It is a 25% increase in the core count compared to its predecessor. The GPU has support for up to 20GB of GDDR6 memory with a 320-bit interface. It is highly unlikely for Nvidia to put 20GB memory in a consumer-grade graphics card. No gaming titles requires 20GB of VRAM, and we can safely say that no game in a foreseeable future will.
So, an educated guess would be 10GB of GDDR6 memory clocked at 15/16 Gbps with a 320-bit interface would be featured in the GeForce RTX graphics card.
GA 104 inside the GeForce RTX 3070
The GeForce RTX 2070 graphics card will feature the GA 104 GPU. It will feature the same number of CUDA cores as the RTX 2080 offers i.e., 3072 CUDA cores spread equally across 48 SMs. The graphics card will feature 8GB of GRRD6 memory clocked at 15Gbps with a 256-bit interface.
The GPU is said to deliver almost the same performance as the vanilla version of current RTX 2080Ti. It means Ampere architecture will bring $1200 worth of Turing architecture performance into a $500 category. Speaking in economic terms, it is a huge leap forward.
GA 106 and GA 107
The mid-range and low-range graphics cards from Nvidia will feature configurations of either GA 106 or GA 107 GPU. The GA 106 GPU would feature 1920 CUDA cores while the GA 107 will reportedly feature 1280 CUDA cores. The GA 106 GPU will support 6GB of GDDR6 memory with a 192-bit interface. On the other hand, its lower-end sibling will feature 4GB of memory with a 128-bit interface.
Lastly, these leaks suggest that Ampere GPUs truly have a magnificent leap over the Turing GPUs. However, these are just rumors, Nvidia had to push its GTC conference due to coronavirus, and we still do not have any release date of the architecture. We hope that the company finds a way around the global phenomenon and release the graphics cards as soon as possible.