The specifications of NVIDIA’s top-end RTX 40 series graphics card have braced the web yet again in more detail than ever (courtesy: Videocardz). Turns out that only the RTX 4090 and the two variants of the RTX 4080 will be released this year. The announcement should come on the 20th of September during the chipmaker’s GTC 2022 keynote.
The GeForce RTX 4090 will be based on AD102-300 GPU and feature 16,384 FP32 CUDA cores across 128 Steaming Multiprocessors (SMs) and 11 Graphics Processing Clusters (GPCs). The GPU core will peak out at 2,520MHz alongside 24GB of 21Gbps GDDR6X memory paired with a 384-bit bus. The memory subsystem will be complemented by a large 72MB L2 cache, with a TGP of 450W and an upper limit of 660W. The Founders Edition (and base AIB models) should have a power cap of 450-500W while the factory overclocked AIO models might reach the 660W mark.
| GPU | GA102 | AD102 | RTX 4090 | AD103 | RTX 4080 | RTX 4070 Ti (AD104) | RTX 4070 |
|---|---|---|---|---|---|---|---|
| Arch | Ampere | Ada Lovelace | Ada Lovelace | Ada Lovelace | |||
| Process | Sam 8nm LPP | TSMC 5nm | TSMC 5nm | TSMC 5nm | |||
| GPC | 7 | 12 | 11 | 7 | 7 | 5 | 5 |
| TPC | 42 | 72 | 64 | 42 | 40 | 30 | 30 |
| SMs | 84 | 144 | 128 | 84 | 80 | 60 | 60 |
| Shaders | 10,752 | 18,432 | 16,384 | 10,752 | 9,728 | 7,680 | 7,680 |
| TP | 37.6 | ~100 TFLOPs? | 83 TFLOPs | ~50 TFLOPs | 47 TFLOPs? | ~35 TFLOPs | 35 TFLOPs? |
| Memory | 24GB GDDR6X | 48GB GDDR6X | 24GB GDDR6X | 16GB GDDR6X | 12GB GDDR6X | ||
| L2 Cache | 6MB | 96MB | 72MB | 64MB | 48MB | ||
| Bus Width | 384-bit | 384-bit | 256-bit | 192-bit | |||
| TGP | 350W | 600W | 450W | 450W | 285-340W | 300W | 285W |
| Launch | Sep 2020 | Sept 22? | Sept 22? | Q1 2023? |
The GeForce RTX 4080 will be the more accessible Lovelace SKU coming in two variants: 12GB and 16GB. The former will be based on the AD104-400 core with 7,680 shaders across 60 SMs and 5 GPCs. The bus width is supposedly going to be set at 192 or 256-bit. The graphics core will peak out at 2,610MHz and the memory will run at 21Gbps GDDR6X. The default TGP is said to be 285W with a maximum of 366W on custom AIB variants. This may very well be the RTX 4070 or the 4070 Ti as we’re looking at an entirely different die.
The 16GB variant of the RTX 4080 will leverage the AD103-300 GPU with 9,728 cores pulsating at 2,505MHz alongside 23Gbps GDDR6X memory modules. Unlike its 12GB sibling, we’re looking at a bus width of 256-bit or higher. It’ll have a default TGP of 340W, raisable to 516W on the over-engineered board partner designs.
The GeForce RTX 4070 is somewhere in the air for the time being and is unlikely to be launched this year. We might see a sneak peek or reveal later this year but the official announcement should come at CES 2023 in January.
