At CES 2025, NVIDIA officially introduced the new RTX 50 series graphics cards, unveiling detailed specifications. Jensen Huang, in his keynote, announced that the RTX 5070 has achieved performance levels equivalent to the RTX 4090.
The RTX 5090 is powered by the Blackwell GB202-300 core, with a base clock of 2.01GHz and a boost clock reaching 2.41GHz, supported by 21,760 CUDA cores and 170 SM units. This powerhouse includes a 28Gbps 32GB GDDR7 memory configuration, a 512-bit memory interface, and a staggering memory bandwidth of 1792GB/s. It utilizes a PCIe 5.0x16 interface and requires a TBP power consumption of 600W, powered by a single 16-pin connection.
The RTX 5080 features the Blackwell GB202-400 core, operating at a base frequency of 2.3GHz and a boosted frequency of 2.62GHz. It houses 10,752 CUDA cores and 84 SM units. Unique among the RTX 50 lineup, it sports 32Gbps 16GB GDDR7 memory, a 256-bit memory bus, and a memory bandwidth of 1024GB/s. This card follows a PCIe 5.0x16 standard, with a total power draw of 400W and also uses a single 16-pin power connector.
The RTX 5070 Ti utilizes the Blackwell GB203-300 core, with its base and boost frequencies set at 2.3GHz and 2.45GHz, respectively. It offers 8960 CUDA cores and 70 SM units and contains 28Gbps 16GB GDDR7 memory. Its 256-bit memory interface enables a bandwidth of 896GB/s. The PCIe 5.0x16 interface supports a TBP of 300W, with power supplied through a 16-pin connector.
The RTX 5070 is built on the Blackwell GB205-300 core, featuring a base clock of 2.16GHz and a boost clock of 2.51GHz. It includes 6144 CUDA cores across 48 SM units. Its 28Gbps 12GB GDDR7 memory is supported by a 192-bit interface and a bandwidth of 672GB/s. This model also uses the PCIe 5.0x16 interface, with a TBP power requirement of 250W, powered through a single 16-pin power connector.