Nvidia Blackwell graphics card is expected to launch soon, with RTX 5090 by the end of 2024
Nvidia plans to launch specialized GPUs that process language models or high-end AI algorithms, based on the Blackwell chip architecture, named after mathematician David Blackwell, a pioneer in developing probability theory and information theory. As early as April last year, the first rumors about Blackwell were revealed. Then in September, the configuration of the GB 202 GPU, expected to be equipped on the RTX 5090, was also released by Moore’s Law is Dead channel:
The two most important factors for Nvidia to make a decision will be the sales of RTX 40 series graphics cards, which will be launched at the end of 2022, and are expected to have Super versions with slight performance upgrades this year, and Second is the power and competitiveness of AMD’s new generation of graphics cards, the Radeon RX 8000 series, applying the new generation RDNA 4 architecture GPU.
Rumors say that the RTX 50 series, with the most advanced product being the RTX 5090, will not create a significant power difference like the transition between the RTX 3090 Ti and RTX 4090, at least when combined. with technologies such as Frame Generation and Ray Reconstruction of the DLSS 3 solution. However, the nature of the AD102 GPU on the RTX 4090 has also been reduced by Nvidia by about 10% of its power compared to its inherent capabilities.
In addition to increasing the number of processing cores by 50%, Nvidia also set a goal of increasing memory bandwidth by 52%, thanks to the application of 32Gb/s GDDR7 memory technology, compared to 21Gb/s GDDR6X bandwidth on the RTX 4090. Present. In terms of clock speed, the number can increase by 15%, which means reaching the threshold of 2.9 GHz boost clock when playing video games. Then when combined with GPU Boost technology, that number can completely reach the 3 GHz mark. Currently the most powerful graphics card in the world, RTX 4090 has a base clock of 2,520 MHz.
Finally, the cache on the Blackwell GPU equipped with the RTX 5090 will dramatically increase L2 cache capacity, to increase memory bandwidth, exactly like how AMD does with Infinity Cache. On the Radeon RX 7900 XTX, L2 cache reaches 96MB, while Nvidia wants the RTX 5090 to have up to 128MB of L2 cache.
