NVIDIA’s GeForce RTX 4080 will feature 16GB of GDDR6 memory and 10,240 FP32 cores while the RTX 4090 will pack 16,384 cores alongside 24GB of GDDR6X memory, much like the RTX 3090 Ti. The GeForce RTX 4070, on the other hand, will feature a total of 7,168 FP32 shaders paired with 10GB of GDDR6 memory via a 160-bit bus. The RTX 4090, like its predecessor, will leverage a 384-bit bus alongside super-fast 21Gbps GDDR6X memory on the AD102-300 die.
The GeForce RTX 4080 will cut the bus down to 256-bit alongside 16GB of 18Gbps memory on the AD103-300 cores. Finally, the RTX 4070 will be a product of the AD104-275 die with a 160-bit bus and 10GB of 18Gbps memory. AMD will almost certainly pair its Radeon RX 7700 XT with 12GB of GDDR6 memory to offer a clear advantage at higher resolutions.
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? |
You can go through the configuration of the full-fledged AD10x dies in the above table. NVIDIA is clearly leaving enough room for a GeForce RTX 4070 Ti, RTX 4080 Ti, and RTX 4090 Ti by offering partially disabled dies with its first wave of Ada Lovelace dies.
The AD102 core which is set to power the RTX 4090 and the RTX 4090 Ti will have a peak power draw of 800W, although as Kimi states, the maximum limit will likely not be utilized. Instead, we’re likely to get a peak TGP of 450-550W for the RTX 4090 and 650-700W for the RTX 4090 Ti. Meanwhile, the RTX 4080 will get a TGP of 400-450W on the desktop and 150-175W on the mobility variant.
The midrange RTX 4070 leveraging the AD104 GPU core will have a peak TGP of 400W, indicating a power spec of 350W. The mobile variant will have the same power figures as the RTX 4080 mobile at 150-175W. Finally, we have the RTX 4060 which will be based on the AD106 die. The desktop variant will have a TGP of 200-260W while the notebook part will top out under 140W.
As already stated in the beginning, these are the maximum supported power draws by these dies, and NVIDIA is unlikely to push them to the max, at least on the reference models. The extreme OC liquid-cooled variants may come close but the air-cooled variants should run a fair bit under these limits. The higher-end RTX 40 series cards are slated for a Q4 2022 launch, right alongside the Radeon RX 7000 GPUs. The delay is a result of the flooding of the retail channel due to the sudden crash in cryptocurrency prices which has caused miners to dump their hardware into the reseller market.