NVIDIA RTX 4070 , the SUPER
A recent benchmark revelation on Geekbench has shed light on the specifications for the RTX 4070 SUPER graphics card. The data indicates that the GPU is equipped with 56 “Compute Units,” which, in NVIDIA’s architecture, correspond to Streaming Multiprocessors. This suggests the GPU boasts 7168 CUDA cores, assuming 128 CUDA cores per Compute Unit. This specification corresponds with what has been previously rumored. Additionally, the graphics card is said to operate with a base clock speed of 2475 MHz and features a memory speed of 10.5 GHz, which translates to a bandwidth of 21 Gbps. The
In the latest performance evaluations using Geekbench 5.5.1, the graphics card achieved a score of 195,384 in the OpenCL assessment and a significant 219,237 points in a benchmark that assesses CUDA performance. Due to compatibility issues, Radeon graphics cards are excluded from this comparison. The focus is on how this card measures up to the RTX 4070 Ti and RTX 4070 versions.
To ensure the reliability of the benchmarks, extensive research was conducted across various platforms to gather a robust data set. The aim was to collect at least 15-20 comparable performance metrics, and the following are the synthesized results from this meticulous data collection.
The RTX 4070 SUPER is found to have roughly 7% fewer cores than its Ti SUPER counterpart. Test results show a slight decrease in performance, ranging from 4% to 7% slower than the Ti SUPER version. However, when pitted against the standard RTX 4070, the SUPER version has 21.7% more cores, resulting in an improvement in performance between 9% to 12.6%. This gain is achieved with just a modest rise in thermal design power (TDP) by 20W, which is about 10% more.
It is essential to note that while Geekbench GPU benchmarks provide valuable insights, they may not always be entirely precise, and comparing such data can be fraught with challenges. Hence, these benchmarks should be regarded as a preliminary look rather than a conclusive measure of the RTX 4070 SUPER’s capabilities.
More definitive benchmarks are expected to be published in the coming week, and comprehensive independent reviews should be available by January 16, offering a fuller evaluation of the GPU’s performance.
Source: Geekbench (OpenCL), (CUDA)