NVIDIA GeForce RTX 5070 Ti and RTX 5070: Specifications leaked

Recent leaks have once again revealed the technical specifications of the NVIDIA GeForce RTX 5070 Ti and RTX 5070. The information comes from a well-known leaker and provides insights into the CUDA cores, GPU die, VRAM and power consumption (TGP) of the two graphics cards.

General overview

The RTX-5070 series is part of the upcoming NVIDIA Blackwell RTX 50 series, which is expected to be unveiled at CES. Details about various models in this series have been circulating for weeks. The latest leak confirms some specifications of the two cards from the upper mid-range segment: RTX 5070 Ti and RTX 5070.

GeForce RTX 5070
PG146/147-SKU70
GB205-300-A1
6144FP32
192-bit GDDR7 12G
250W

— kopite7kimi (@kopite7kimi) December 25, 2024

Technical specifications of the GeForce RTX 5070 Ti

• SKU name: GeForce RTX 5070 Ti
• Board: PG147-SKU60
• GPU: GB203-300-A1
• CUDA cores: 8,960
• VRAM: 16 GB GDDR7, 256-bit bus
• Power consumption (TBP): 300 W

The RTX 5070 Ti is equipped with the GB203 GPU die and has 16 GB GDDR7 memory, which is connected via a 256-bit interface. The energy consumption is 300 watts, which represents a reduction of 50 watts compared to previous speculation. Compared to the RTX 4070 Ti, consumption is only 15 watts higher. However, variants with higher consumption of up to 350 watts could appear.

Technical data of the GeForce RTX 5070

• SKU name: GeForce RTX 5070
• Board: PG146-SKU70
• GPU: GB205-300-A1
• CUDA cores: 6,144
• VRAM: 12 GB GDDR7, 192-bit bus
• Power consumption (TBP): 250 W

The RTX 5070 is based on the GB205-300-A1 die and offers 12 GB GDDR7 memory via a 192-bit interface. The number of CUDA cores is 6,144, which is slightly below the previously speculated 6,400. The energy consumption is specified at 250 watts and thus remains at a level that was already suspected in earlier leaks.

Comparison of the specifications:

Classification and release

The RTX 5070 Ti and RTX 5070 use new memory technologies and benefit from the further development of the Blackwell architecture. Both models are expected to be released in the first quarter of 2025, following the launch of the RTX 5080 and RTX 5090. Although prices are not yet known, NVIDIA is expected to set them at a similar level to their predecessors.

This information is based on unofficial sources and is subject to change until the official announcement. However, the high correspondence of the current data with previous leaks indicates that the specifications are likely to be accurate.

Source: Kopite7Kimi, Kopite7Kimi