NVIDIA doesn't seem to be stopping at the GeForce RTX 3090 as its flagship Ampere graphics card and preparing a SUPER refresh of the card with the full GA102 GPU.
NVIDIA GeForce RTX 3090 SUPER Rumored To Feature 10752 CUDA Cores & Over 400W TGP
Currently, the GeForce RTX 3090 serves as NVIDIA's flagship gaming offering but the green team wants to take things up a notch with the upcoming SUPER refresh. The NVIDIA GeForce RTX 3090 SUPER is expected to feature the full-fat GA102 GPU core, faster clocks, and a huge requirement in terms of power, as rumored by Greymon55.
400W+
— Greymon55 (@greymon55) August 25, 2021
Maybe,
RTX 3090 Super
"GA102-350-A1
10752FP32, no NVLINK
Original 21Gbps MEM
TGP>=450W"
Launch in 2021.
🤫🤫🤫— kopite7kimi (@kopite7kimi) August 26, 2021
NVIDIA GeForce RTX 3090 SUPER Graphics Card Rumored Specifications
At the heart of the NVIDIA GeForce RTX 3090 SUPER graphics card will be the GA102 GPU. The GPU is based on Samsung's 8nm custom process node designed specifically for NVIDIA and features a total of 28 Billion transistors. It measures at 628mm2 which makes it the 2nd biggest gaming GPU ever produced right below the Turing TU102 GPU.
The NVIDIA GeForce RTX 3090 will be the first gaming graphics card to feature the full GA102 GPU with a total of 84 SM units that would result in a total of 10752 CUDA cores. This would be a 2.5% increase in core count but the rumor also states that the graphics card will feature a TGP of over 400W. That's 50W more than the existing GeForce RTX 3090 which means that we could be looking at faster clocks on GPU and VRAM.
The GeForce RTX 3090 rocks a 24 GB GDDR6X memory interface which should be retained but we can expect NVIDIA to boost memory clocks to 20 Gbps (19.5 Gbps on RTX 3090) for 960 GB/s bandwidth or they could take things further to hit that sweet 1 TB/s spot with 21 Gbps DRAM chips.
We could expect a 5% performance increase which would be in line with previous 'SUPER' launches from NVIDIA. It is likely that NVIDIA could cut MSRPs on existing cards and replace them with these newer SUPER refreshes at the same MSRP early next year. With that said, NVIDIA's GeForce RTX 30 SUPER Series are reportedly going into production this quarter and we can expect a launch within the first half of 2022.
NVIDIA GeForce RTX 30 Series Graphics Card Specifications
| Graphics Card Name | NVIDIA GeForce RTX 3090 Ti | NVIDIA GeForce RTX 3090 | NVIDIA GeForce RTX 3080 Ti | NVIDIA GeForce RTX 3080 12 GB | NVIDIA GeForce RTX 3080 | NVIDIA GeForce RTX 3070 Ti | NVIDIA GeForce RTX 3070 | NVIDIA GeForce RTX 3060 Ti | NVIDIA GeForce RTX 3060 Ti | NVIDIA GeForce RTX 3060 | NVIDIA GeForce RTX 3060 | NVIDIA GeForce RTX 3050 | NVIDIA GeForce RTX 3050 6 GB |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GPU Name | Ampere GA102-350? | Ampere GA102-300 | Ampere GA102-225 | Ampere GA102-220 | Ampere GA102-200 | Ampere GA104-400 | Ampere GA104-300 | Ampere GA104 | Ampere GA104-200 Ampere GA103-200 | Ampere GA106-300 | Ampere GA106 | Ampere GA106-150 | Ampere GA107-325 |
| Process Node | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm |
| Die Size | 628.4mm2 | 628.4mm2 | 628.4mm2 | 628.4mm2 | 628.4mm2 | 395.2mm2 | 395.2mm2 | 395.2mm2 | 395.2mm2 (GA104) | 276mm2 | 276mm2 | 276mm2 (GA106) | 200mm2 (GA107) |
| Transistors | 28 Billion | 28 Billion | 28 Billion | 28 Billion | 28 Billion | 17.4 Billion | 17.4 Billion | 17.4 Billion | 17.4 Billion (GA104) | 13.2 Billion | 13.2 Billion | 13.2 Billion (GA106) | 8.7 Billion (GA107) |
| CUDA Cores | 10752 | 10496 | 10240 | 8960 | 8704 | 6144 | 5888 | 4864 | 4864 | 3584 | 3584 | 2560 | 2048 |
| TMUs / ROPs | 336 / 112 | 328 / 112 | 320 / 112 | 280 / 104 | 272 / 96 | 184 / 96 | 184 / 96 | 152 / 80 | 152 / 80 | 112 / 64 | 112 / 64 | 80 / 32 | TBD |
| Tensor / RT Cores | 336 / 84 | 328 / 82 | 320 / 80 | 280 / 70 | 272 / 68 | 184 / 46 | 184 / 46 | 152 / 38 | 152 / 38 | 112 / 28 | 112 / 28 | 80 / 20 | TBD |
| Base Clock | 1560 MHz | 1400 MHz | 1365 MHz | TBA | 1440 MHz | 1575 MHz | 1500 MHz | 1410 MHz | 1410 MHz | 1320 MHz | 1320 MHz | 1552 MHz | TBD |
| Boost Clock | 1860 MHz | 1700 MHz | 1665 MHz | TBA | 1710 MHz | 1770 MHz | 1730 MHz | 1665 MHz | 1665 MHz | 1780 MHz | 1780 MHz | 1777 MHz | 1470 MHz |
| FP32 Compute | 40 TFLOPs | 36 TFLOPs | 34 TFLOPs | TBA | 30 TFLOPs | 22 TFLOPs | 20 TFLOPs | 16 TFLOPs | 16 TFLOPs | 13 TFLOPs | 13 TFLOPs | 9.1 TFLOPs | TBD |
| RT TFLOPs | 74 RFLOPs | 69 TFLOPs | 67 TFLOPs | TBA | 58 TFLOPs | 44 TFLOPs | 40 TFLOPs | 32 TFLOPs | 32 TFLOPs | 25 TFLOPs | 25 TFLOPs | 18.2 TFLOPs | TBD |
| Tensor-TOPs | TBA | 285 TOPs | 273 TOPs | TBA | 238 TOPs | 183 TOPs | 163 TOPs | 192 TOPs | 192 TOPs | 101 TOPs | 101 TOPs | 72.8 TOPs | TBD |
| Memory Capacity | 24 GB GDDR6X | 24 GB GDDR6X | 12 GB GDDR6X | 12 GB GDDR6X | 10 GB GDDR6X | 8 GB GDDR6X | 8 GB GDDR6 | 8 GB GDDR6X | 8 GB GDDR6 | 12 GB GDDR6 | 8 GB GDDR6 | 8 GB GDDR6 | 6 GB GDDR6 |
| Memory Bus | 384-bit | 384-bit | 384-bit | 384-bit | 320-bit | 256-bit | 256-bit | 256-bit | 256-bit | 192-bit | 128-bit | 128-bit | 96-bit |
| Memory Speed | 21 Gbps | 19.5 Gbps | 19 Gbps | 19 Gbps | 19 Gbps | 19 Gbps | 14 Gbps | 19 Gbps | 14 Gbps | 16 Gbps | 15 Gbps | 14 Gbps | 14 Gbps |
| Bandwidth | 1008 GB/s | 936 GB/s | 912 Gbps | 912 Gbps | 760 GB/s | 608 GB/s | 448 GB/s | 608 GB/s | 448 GB/s | 384 GB/s | 240 GB/s | 224 GB/s | 168 GB/s |
| TGP | 450W | 350W | 350W | 350W | 320W | 290W | 220W | 175W | 175W | 170W | 150W | 130W (GA106) 115W (GA107) | 70W |
| Price (MSRP / FE) | TBD | $1499 US | $1199 | $999 US? | $699 US | $599 US | $499 US | $399 US | $399 US | $329 US | TBD | $249 US | TBD |
| Launch (Availability) | 29th March 2022? | 24th September 2020 | 3rd June 2021 | 11th January 2022 | 17th September 2020 | 10th June, 2021 | 29th October 2020 | October 2022 | 2nd December 2020 | 25th February 2021 | October 2022 | 27th January 2022 | 2024 |
