It seems the leaks never end for gamers. Sure, The Last of Us Part II was ruined for a lot of people but the important leak is here, the next line of graphics cards from NVIDIA (my condolences to people whose excitement for The Last of US Part II has been soured by the aforementioned leaks). The current RTX line is kind…useless. Virtually no games use the RTX feature and the ones that do end up coming off second best. Ray-tracing has been undercooked in everything that isn’t Minecraft RTX, DLSS is just not great and similar performance results can be achieved by increasing the resolution scale in games, and RTX voice doesn’t actually require an RTX card to work. People have struggled to justify the absurd price of the RTX 2080 Ti (with a price of entry around AUD $1750) for such a little performance gain and tech that is half-baked. It’s been well over a year since the launch of the RTX line and the time has come for rumours regarding NVIDIA’s next line for GPUs to circulate.
Sourced from Twitter user @chiakokhua, whose tweet has since been deleted, this graph detailing the sepcs of the next line of GPUs reveals quite a lot. Chiakokhua has a pretty solid history with tech, like predicting that AMD would use chiplet design for its newer Zen line. So let’s take a look at the NVIDIA leak.
Clocking in at 1750MHz with up to a whopping 8192 CUDA cores, the GA100 (presumably the 3080 Ti) looks to blow the current flagship RTX card out of the water. It also appears to feature substantially more RT cores (increasing from the current 68 to, at most, 256), which almost leads me to believe they couldn’t combat the speed issue of these cores and have instead opted for simultaneous compute. Tensor cores have been beefed from 544 to a max of 1024.
|Model||Architecture||CUDA Cores||RT Cores||Boost Clock (MHz)||VRAM Speed|
|RTX 3080 Ti (GA100)||Ampere||<8192||<256||<1750 MHz||16 Gbps|
|RTX 2080 Ti||Turing||4352||68||1545 MHz||14 Gbps|
All the other cards in the potential line up look pretty decent too, with the GA103 (3080) featuring a higher CUDA core count than the 2080 Super and the GA104 (3060) potentially boasting the same CUDA core count as the 2070 Super.
Obviously, this should be taken with a grain of salt but there are some changes here which seem pretty consistent. The shift from a 12nm lithography to 7nm is pretty big (or small, technically) and so it would make sense that NVIDIA would be cramming more into their next line of graphics cards. Where this leak kind of looses me is the boost clock…or “boots clock”. It is a pretty minor thing to point out but it is still a leak, after all.
Knowing NVIDIA, cards this powerful mean there are going to be new pricing tiers, as evidenced with their pricing behaviour with the Pascal and Turing architectures. I won’t be surprised to see an insane price tag slapped on cards this powerful, especially the 3080 Ti (GA100). However, only time will tell.