NVIDIA plans to introduce two brand new ones GPU architectures, Ada Lovelace e Hopper (from the surname of the American computer scientist Grace Hopper), next year. According to reports from DigiTimes colleagues, the new graphics processors will target virtually all segments of the market and various sources believe they will be built using the TSMC’s N5 process technology.
Right now, NVIDIA uses the Samsung’s 8LPP manufacturing process to build its GPUs GeForce RTX serie 30 based onAmpere architecture, as well as the nodo N7 at TSMC to produce its GPU for data center GA100 dedicated toartificial intelligence (TO THE) and al high performance computing (HPC). However, it appears that next year the company will exclusively use TSMC for all of its GPUs, specifically those based on the Hopper architecture for datacenter, AI and HPC applications, as well as Ada Lovelace ones for GeForce GPUs for gaming.
NVIDIA Hopper H100 it should be there the company’s first multi-die processing GPU to maximize per-socket performance. It is expected that it will be paired with memorie HBM and will use the tecnologia di packaging chip-on-wafer-on-substrate (CoWoS) di TSMC. By contrast, the GeForce RTX 40 series Ada Lovelace GPUs will be aimed at gamers and likely use traditional memory and FC-BGA packaging. DigiTimes says ASE technology was used to test and package these GPUs.
What isn’t completely clear is when exactly NVIDIA plans to introduce its next generation GPUs. Based on the rumors, the company intends to renew its GeForce 30 series family once again early next year with new Ti and Super models. Since there is no point in updating a line-up if a brand new series is planned to launch soon, we could speculate that Ada Lovelace graphics cards will only be available towards the end of 2022.
Keeping in mind that the A100 compute GPU will have to contend with AMD’s Instinct MI200 series accelerators, NVIDIA is certainly interested in launching its Hopper architecture as soon as possible, most likely anticipating Ada Lovelace products.