GeForce RTX 4000, the Ada Lovelace GPUs manufactured at 5 nanometers by TSMC?

NVIDIA could embrace the TSMC’s production process N5 (5 nanometers) for next year’s GPUs, Ada Lovelace e Hopper. It is not the first time that there are rumors of this development, but this time it is the site that talks about it Digitimes, closely linked to the Taiwanese production chain.

The Ampere GPUs that we find on the GeForce RTX 3000 are all produced by Samsung with a custom 8 nanometer process, while the GA100 processor that moves the NVIDIA A100 accelerator is made by TSMC at 7 nanometers (N7). While having two manufacturers gives NVIDIA greater flexibility, it may not mean that both offer an adequate process to meet the performance, consumption, volume and cost targets set for next generation projects.

Perhaps Samsung is unable to meet these requirements and for this NVIDIA would have decided to return to TSMC with all its projects. The hope is that the Taiwanese foundry will first of all be able to guarantee a production of chips in adequate volumes to cover the demand, the real problem of this generation of video cards, even if the TSMC N5 process is at the center of the production roadmap of many other companies. then NVIDIA will have to elbow.

Hopper is the code name of an architecture intended for datacenters, to artificial intelligence and the supercomputer sector, while Ada Lovelace will respond to the needs of gamers.

As for the server world, we expect a called GPU H100 based on a project with more die on the package, a bit along the lines of what has recently been seen at AMD with the Instinct accelerators of the 200 series. The Hopper GPU should be paired from memory HBM latest generation (HBM2E or HBM3) thanks to TSMC’s CoWoS (chip-on-wafer-on-substrate) packaging technology.

As for the GPUs Ada Lovelace, NVIDIA should continue on the path of “monolithic die“flanked by the more traditional, and less expensive, type memory GDDR.

About the respective exit period, while for Hopper we expect an announcement at the GTC 2022 to be held between 21 and 24 March next year, on the Ada Lovelace front an arrival is expected in the third – fourth quarter: at the beginning of 2022 NVIDIA is in fact expected to refresh the Ampere range with new models, so it is conceivable a greater expectation for gamers.

Expectation that we hope will be rewarded by the performance. According to previous rumors, the flagship gaming GPU should be called AD102 and could boast 18432 CUDA cores, returning double performance compared to the current GA102. A graphics chip of that type should occupy around 600mm2 and consume 400-450W despite the more advanced manufacturing process. Consequently, the card will need a new power connector, a 12-pin already indirectly revealed by ASUS.

Gift ideas, why waste time and risk making mistakes?
GIVE A GOOD AMAZON!