Nvidia RTX 5070 Series Details Highlight Blackwell Architecture and Performance Gains

Kopite, a well-known hardware leaker, has provided detailed specifications for Nvidia’s upcoming RTX 5070 series, which includes the RTX 5070 Ti and the base RTX 5070. These GPUs are expected to launch next month during CES 2024, along with the RTX 5090 and RTX 5080, all part of Nvidia’s new Blackwell architecture. Kopite has a solid track record for accurate leaks, having previously revealed information about the RTX 5090 and RTX 5080. The new RTX 5070 series shows both improvements and trade-offs compared to Nvidia’s Ada Lovelace architecture.

The RTX 5070 Ti features the GB203-300-A1 die, the same as the RTX 5080, with 8960 CUDA cores (70 SMs), representing a 16% increase over the RTX 4070 Ti. It is paired with 16GB of GDDR7 memory on a 256-bit bus, with a memory speed of 28 Gbps and a total bandwidth of 896 GB/s. Despite having fewer cores than the RTX 5080, the RTX 5070 Ti’s specs put it in close competition with that card. The power consumption of the RTX 5070 Ti is set at 300W, which is 15W higher than its predecessor, the RTX 4070 Ti.

The base RTX 5070, on the other hand, is powered by the GB205-300-A1 die, a smaller variant compared to the die in the RTX 4070. This results in a reduction in core count, with 6144 CUDA cores, which is still 4% more than the RTX 4070. It comes with 12GB of GDDR7 memory on a 192-bit interface, offering 672 GB/s of memory bandwidth. The power draw of the RTX 5070 is expected to be 250W, 25% more than the RTX 4070, reflecting the increased performance.

In comparison to the RTX 4080 and RTX 4070 Ti, which had a noticeable spec gap, the Blackwell-based RTX 5070 series may not have as large a disparity, potentially resulting in a more competitive pricing structure. The power consumption increases with each tier in the RTX 5070 series, with the 5070 Ti drawing 300W and the 5070 drawing 250W, compared to the 200W of the base RTX 4070. The anticipated spec changes suggest a well-balanced performance-to-power ratio for the new GPUs.

One key takeaway is the increased memory bandwidth and CUDA core count in the RTX 5070 Ti, which could allow for better performance in demanding workloads. However, Nvidia’s RTX 5070 series might face competition from both Intel and AMD, who are currently absent from the high-end market. Additionally, Nvidia may explore future GPUs with higher VRAM capacities using the latest GDDR7 memory, although these are still speculative. Users looking for greater VRAM might need to turn to Nvidia’s Blackwell data center or workstation GPUs, which are likely to be priced significantly higher.