Additional details regarding NVIDIA’s next-gen RTX 4080 graphics card have surfaced. From what I can tell, the core specifications are now more or less set in stone. Based on the AD103-300-A1 GPU core, we’re looking at 9,728 FP32 cores (ALUs) spread across 76 SMs, paired with the PG136/139-SKU360 board. This is a step up from the 8,704 shaders on the RTX 3080 and core clocks going from 1.7GHz to nearly 3GHz.
The memory bus has been trimmed, going from 320-bit on Ampere to 256-bit on the Lovelace RTX x080. The GDDR6X memory is substantially faster, running at a remarkable 23Gbps for an overall bandwidth of 736GB/s, a smidge less than the RTX 3080’s 760GB/s.
|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|
|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|
|Launch||Sep 2020||Sept 22?||Sept 22?||Q1 2023?|
Then there’s the L2 cache. We’re most likely still looking at 48MB of LLC, a massive upgrade from the 5MB found on the RTX 3080. Finally, there’s the matter of the Total Board Bower (TBP). NVIDIA has further optimized the power consumption of the RTX 4080, bringing it down to just 340W. That’s just 20W more than the RTX 3080, a massive improvement over the initially rumored 400-450W target.
It’s unclear whether NVIDIA will launch the RTX 4080 later this year or postpone it to early 2023. Rumors claim that the RTX 4090 will land sometime next month but the remaining Lovelace SKUs might get pushed to next year.