Specifications for Nvidia’s RTX 3080 and 3070 have leaked – A 10/20GB graphics card for gamers?

Specifications for Nvidia's RTX 3080 and 3070 have leaked 0 A 10/20GB graphics card for gamers?

Specifications for Nvidia’s RTX 3080 and 3070 have leaked – A 10/20GB graphics card for gamers?

Nvidia’s RTX series graphics cards have been with us since Q3 2018, making it likely that 2020 will see the release of a Turing successor and the start of Nvidia’s next-generation product stack. 

Specifications for Nvidia’s rumoured Ampere series graphics cards have leaked over at mydrivers, listing the core and memory specifications of the green team’s GA103 and GA104 core designs. These core designs are likely to make up Nvidia’s RTX 3080 and RTX 3070 respectively.  

GA103 can be seen as a midway point between Nvidia’s RTX 2080 Ti and RTX 2080 Super, offering users 10 GB of VRAM over a 320-bit memory bus and up to 3480 CUDA cores. With this 320-bit memory bus, 20GB of VRAM could also be offered, but such a high allotment of VRAM would only be suitable for Nvidia’s Quadro series graphics cards. 

With the RTX 2080 Ti offering 4,352 stream processors and the RTX 2080 Super offering 3072 CUDA cores, it is easy to see why GA103 is a candidate for Nvidia’s RTX 3080, especially when combined with Nvidia’s Ampere architectural improvements. GA104 is said to offer 3072 CUDA cores and 8/16 GB of memory over a 256-bit memory bus, making it an RTX 3070 candidate.  

Both GA103 and GA104 are said to be 7nm products, though the source of this rumour doesn’t know if these parts are sourced from TSMC or Samsung. 

Strangely, Nvidia’s previous core names typically use an even-numbered naming scheme, making GA103 a strange choice. MyDrivers also states that there will definitely be a GA102 chip, a part which will likely be used to make Nvidia’s next RTX Titan and their RTX 3080 Ti. 

Specifications for Nvidia's RTX 3080 and 3070 have leaked 0 A 10/20GB graphics card for gamers?  

Architecture Matters

While CUDA core counts and memory allotments matter, the benefits of Nvidia’s Ampere architecture remain challenging to quantify. While CUDA core counts can be a good metric for comparison, it is ultimately useless unless we know what Nvidia’s Ampere architecture brings to the table. Should we expect higher clock speeds or increased graphics IPC? If so, Nvidia’s CUDA core count numbers may become less relevant, as Nvidia could be getting a lot more work out of Ampere’s SMs.    

You can join the discussion on Nvidia’s rumoured RTX 3080 and RTX 3070 graphics cards on the OC3D Forums.