Well-reputed tipster @kopite7kimi has provided an update on the specifications of NVIDIA’s upcoming RTX 40 series graphics cards. It looks like the higher-end SKUs will be getting a healthy memory upgrade albeit still lower than AMD’s Radeon RX 6800 XT/6900 XT. The RTX 4090 will come with 24GB of 21Gbps GDDR6X memory paired with a 384-bit bus, much like the 3090 Ti. The shader count is expected to be somewhere between 15,000 and 18,000. This compute monster will draw up to 600W of power under load (TGP).
The RTX 4080 which will be sought after by most enthusiasts should feature 16GB of GDDR6X (20 Gbps?) or 18Gbps GDDR6 memory. In comparison, the RTX 3080 comes with a 10GB/12GB GDDR6X memory buffer. It’s important to remember that the faster GDDR6 modules from Micron are almost on par with GDDR6X so there’s a good chance that the latter will be limited to the 4090 and the 4080 Ti. The RTX 4080 should offer 10,752 FP32 cores, a 256-bit bus, and a TGP of 600W. Expect high core clocks to make up for the lower bandwidth.
|GPU||TU102||GA102||AD102 (RTX 4090)||AD102 (RTX 4080)||AD104 (RTX 4070)|
|Arch||Turing||Ampere||Ada Lovelace||Ada Lovelace||Ada Lovelace|
|Process||TSMC 12nm||Sam 8nm LPP||TSMC 5nm||TSMC 5nm||TSMC 5nm|
|TP||16.1||37.6||~90 TFLOPs?||~50 TFLOPs||~35 TFLOPs|
|Memory||11GB GDDR6||24GB GDDR6X||24GB GDDR6X||16GB GDDR6||16GB GDDR6|
|Launch||Sep 2018||Sep 2020||Aug-Sep 2022||Q4 2022||Q4 2022|
Finally, we have the RTX 4070 which will be based on the AD104 die. This particular SKU along with the RTX 4060 will be the bestsellers of Lovelace. NVIDIA has paired the 4070 with 12GB of 18Gbps GDDR6 memory across a 256-bit bus. The shader count is pegged at 7,680, and the TGP will likely max out at 300W. The RTX 3070 features 8GB of GDDR6 memory across a 256-bit bus along with 5,888 shaders.