Details about NVIDIA’s future Blackwell architecture are slowly coming together and becoming clear. We’ve known about the name for a while, seen the GPU nomenclature leak, and heard rumours that NVIDIA is in talks with TSMC to develop Blackwell on the TSMC’s 3nm node, but this is the first (pseudo) official confirmation that the architecture will, in fact, arrive in 2024 as expected.

In response to a question at the Arete Tech Conference 2022, NVIDIA’s Ian Buck underlined that the business is completely committed to releasing a significant GPGPU architecture every two years, with Hopper being delivered this year. This was in reaction to probable new advancements in 2023 and indicates that the Blackwell GPU architecture will be released in 2024 (as we already know it is the architecture that would successor Hopper), which was the predicted timeframe based on NVIDIA’s historical cadence.

At the time of its release, NVIDIA Hopper was the world’s fastest 4nm GPU and the first to use HBM3 memory.

It has more CUDA cores than the NVIDIA RTX 4090 (which has 16,384 CUDA cores), for a total of 18,432 CUDA cores. Blackwell will be a substantial generational upgrade above Hopper (as has always been the case). In a previous leak, four NVIDIA Blackwell GPUs were confirmed.

There isn’t much else known about the Blackwell GPU series at this moment (obviously), because it’s just a string of text. All we know is that this series will most likely follow the Hopper GPUs (which was released in parallel with the mainstream Ada Lovelace GPUs). It is now known that this name relates to the Hopper replacement rather than consumer GPUs.

Also Read:

Source