NVIDIA is planning to launch its GeForce RTX 40 series graphics cards sometime in the third quarter of the year. Based on TSMC’s 5nm process node, the GeForce RTX 4090 will pack as many as 18,432 FP32 cores along 24GB+ of GDDR6X memory. The other notable upgrades include 96MB of L2 cache and 384-bit bus width. To power this monster, you’ll need at least 600W of power which means a 1000W power supply (at the minimum).
GPU | TU102 | GA102 | AD102 | AD103 | AD104 |
---|---|---|---|---|---|
Arch | Turing | Ampere | Ada Lovelace | Ada Lovelace | Ada Lovelace |
Process | TSMC 12nm | Sam 8nm LPP | TSMC 5nm | TSMC 5nm | TSMC 5nm |
GPC | 6 | 7 | 12 | 7 | 5 |
TPC | 36 | 42 | 72 | 42 | 30 |
SMs | 72 | 84 | 144 | 84 | 60 |
Shaders | 4,608 | 10,752 | 18,432 | 10,752 | 7,680 |
TP | 16.1 | 37.6 | ~90 TFLOPs? | ~50 TFLOPs | ~35 TFLOPs |
Memory | 11GB GDDR6 | 24GB GDDR6X | 24GB GDDR6X | 16GB GDDR6 | 16GB GDDR6 |
L2 Cache | 6MB | 6MB | 96MB | 64MB | 48MB |
Bus Width | 384-bit | 384-bit | 384-bit | 256-bit | 192-bit |
TGP | 250W | 350W | 600W? | 350W? | 250W? |
Launch | Sep 2018 | Sep 2020 | Aug-Sep 2022 | Q4 2022 | Q4 2022 |
The GeForce RTX 4080 should be a cut below with roughly 15,000-16,000 FP32 cores and 16GB of GDDR6X memory paired with a 320-bit bus. With a slightly lower L2 cache, it should have a total board power (TBP) of around 450-500W which translates to an 850W power supply at the minimum.
The lower-end Ada dies should have a power profile similar to the GA100, and below. Expect a significant increase in the shader count, and memory buffer as well as the L2 cache. Even the AD104 which is set to power the GeForce RTX 4060 (?) will include a large 48MB L2 cache reserve.
If the GeForce RTX 4090 will draw up to 600W of power, then the RTX 4090 Ti may increase it to 700W. Some estimates even place it as high as 800W+, but I personally find it highly unlikely. The GeForce RTX 4080/4090 should be announced at either Computex 2022 (late May) or Gamescom 2022 (August). The GH100 may be launched as soon as 20-21st March when NVIDIA CEO Jensen Huang takes the stage to talk about the next generation of machine learning hardware and neural network acceleration.