In the Geekbench database, two unannounced NVIDIA graphics cards, which are related either to the new Ampere line or to Hopper, were immediately lit up. In any case, 7552 CUDA cores with 118 stream processors showed record performance, exceeding the performance of the flagship Titan line.
The Geekbench test dispelled the myths of 128 stream multiprocessors, since the test revealed that there were only 64 of them. At the same time, the processor worked at a frequency of 1.11 GHz in the new card, demonstrating a performance of 16.7 teraflops. If you believe past leaks, the flagship model will be another video card: with 8291 CUDA core and 24 GB of memory.
As for the tested model, it scored an impressive 184,096 points, which is 40% higher than the pumped Titan RTX video card. The second leak was the results of testing GPUs with 6912 CUDA cores and 108 SM. With low consumption and a standard frequency of 1.01 GHz, its performance was 13.9 teraflops, which corresponds to the level of RTX 2080 Ti. The memory value of 47 GB, according to insiders, is most likely a Geekbench error in determining the amount of memory. The test result was 141,654 points, which is slightly higher than the performance of Titan RTX.
Given the low value of the clock frequencies, experts concluded that the new generation of NVIDIA graphics cards will be manufactured using the 7-nanometer process technology, since it has significantly lower power consumption. At the same time, the performance of the GPU grows significantly.