Tipster Greymon (Twitter) is back with some more info. According to the leaker, NVIDIA will be refreshing its RTX 30 “Ampere” lineup early next year for both the notebook and desktop platforms. It’s unknown whether the Super lineup will retain the same chips as their predecessors, but we’re either going to see new ones or lower-end SKUs will be bumped up.
Samsung’s 8nm process node will be retained for the Super refresh, and we’re expecting better supply to improve the condition of the market. The last time, only the RTX 2060, 2070, and 2080 got Super refreshes, while the 2080 Ti and 2090 were left alone. This time we’re looking at a refresh of the might RTX 3090 as well.
The RTX 3090 Super is reportedly getting the full-fledged GA102 die, with 10,752 FP32 units, 336 TMUs, 112 ROPs, 336 Tensor Cores, 336 RT Cores, all contained across 84 fully-enabled SMs and 7 GPCs. The overall TDP will be over 400W, possibly close to 420-430W. Considering that this is a fully enabled die, the RTX 3090 Super should cost more than the RTX 3090, possibly around $2,000-2,500.
The mobile variants for gaming notebooks should also land around the same time. The core counts are likely to be bumped up for the RTX 3060 and the RTX 3070. Both these GPUs use slightly cut variants of their respective dies, GA106 and GA104. The RTX 3060 packs 3,584 FP32 cores while the RTX 3070 comes with 5,888 FP32 cores. The full-fledged GA106 and GA104 dies, on the other hand, contain 3,840 and 6,144 FP32 cores, respectively. These should be the core counts of the Supers as well. The TMUs, RT and Tensor core counts should also get a small boost as all the SMs and the accompanying fixed-function hardware units are fully enabled
Like last time, the RTX 3080 Super should be a mild upgrade over its non-Super counterpart. We can expect the fully enabled GA102 (or nearly fully enabled for the 3080 Super) die to power the latter.
As for Lovelace, both the GeForce RTX 40 series and AMD’s RDNA 3 lineup are slated for a late 2022 launch, so this is merely a confirmation of sorts from the source. It’s worth noting that NVIDIA and AMD will leverage the same process node (TSMC’s N5) for their next-gen GPU lineups. Therefore, the shortages may not subside even late into next year. Intel, on the other hand, will launch its Xe-HPG graphics cards in early 2022, right alongside the Supers. Fabbed on N6, Alchemist will initially have an advantage, but by the end of the year, it’ll start lagging behind rival GeForce and Radeon parts. Raja’s kids will be competing with the RTX Super and perhaps an RDNA 2 refresh. Things will indeed be tense in the graphics card market next year, a nice change in the aftermath of the pandemic.