Since NVIDIA released RTX 3090 in September 2020 until the last launch of the company, the RTX 3070 Ti The availability of which began this month of June has passed no more and no less than nine months, and it is expected that the company still has a handful more RTX 30 Series GPUs in the bedroom pending to see the light. Typically 18 months to two years pass between generations of graphics cards, but it seems that given the chip shortage situation we won’t see the RTX 40 Series until 2023… at least.
When will we see the NVIDIA RTX 40 Series?
During a business meeting that took place shortly before the launch of the ARM NVIDIA Grace processors For servers, NVIDIA CFO Colette Kress dropped that we won’t see NVIDIA ARM CPUs until at least year 2023, saying at the same time that it will be when we will see the next generation and next GPU architecture of the brand of green.
The ARM NVIDIA Grace processor has been announced at the latest GTC, and will be primarily targeted for large-scale Artificial Intelligence systems and high-performance HPC computing. NVIDIA has not announced specifications, it has only emphasized the large amount of bandwidth you will have using NVLink cache coherence with its own bus for GPUs, surpassing 900 GB / s with the GPU and reaching 600 GB / s between CPUs in multi-processor systems.
Curiously, Grace takes her name from Grace Hopper, one of the first programmers in the history of computing since she was the one who led the development of the Cobol programming language, and we say that it is curious because precisely the next GPU architecture of the brand will be call to Hopper.
The NVIDIA CFO said that we could be sure that Grace would be made with 5 nanometer lithography, and this code name is also shared by their upcoming GPU architecture as we have already mentioned; They have not said it directly but all this suggests that Hopper, the architecture on which the upcoming NVIDIA RTX 40 Series graphics will be based, will also be made with this same 5 nanometer lithography.
Based on previous exposures, the main number of the Ada Lovelace architecture will be AD102 and it will have up to 12 GPCs (Graphics Processing Groups), 72 TPCs (Texture Processing Groups), and 144 SMs (Streaming Multiprocessors). The SM count would remain at 128 stream processors, so we would be talking about no less than 18,432 Stream Processors, which compared to the 10,752 that the current Ampere architecture GA102 chip has, represents an increase of 71%.