Today, CES officially announced that Jensen Huang will be delivering the keynote address in three months on January 6 local time, marking the anticipated launch of NVIDIA's RTX 50 series graphics cards.

According to previous disclosures, the RTX 5090 is set to feature a PG144/145-SKU30 PCB design alongside a GB202-300-A1 GPU core. Impressively, the new card will activate 170 Stream Processor Units (SMs) out of a possible 192, resulting in a total core count of 21,760—a reduction of 11.4% from the full core setup of the RTX 4090’s AD102 architecture.

On the memory front, the RTX 5090 will boast 32GB of GDDR7 memory operating on a 512-bit bus interface. The GDDR7 memory is expected to have a bandwidth between 28-32 Gbps, culminating in a potential maximum bandwidth of up to 1.792 or 2.00 TB/s. Furthermore, the card is designed with a total board power (TBP) spec of 600W and includes a dual-slot cooling architecture.

NVIDIA has opted to further widen the gap between the xx80 and xx90 SKUs. Notably, the RTX 5080 has 51% fewer cores compared to the RTX 5090, featuring a GB203-400-A1 GPU core with 84 stream processor units and 10,752 cores.

The RTX 5080 comes equipped with 16GB of GDDR7 memory and runs on a 256-bit bus interface. Its memory bandwidth is anticipated to range from 896 GB/s to 1024 GB/s. This card's total power consumption is rated at 400W, indicating an increase from its predecessor configuration.

For context, the RTX 4090 has 68% more CUDA cores than the RTC 4080, while the speculated RTX 5090 has approximately 102% more CUDA cores than the expected RTX 5080.