Taiwan Semiconductor Manufacturing Company ( TSMC ) isreportedlyworking with Nvidia to develop next - generation GPUs use advanced chiplet technology . This quislingism is expected to play an important theatrical role in Nvidia ’s coming “ Rubin ” computer architecture , which is rumour to be the successor to the current Blackwell genesis .
The shift toward a chiplet - base design tag a noteworthy departure from traditional massive GPU structure , pop the question improved performance , scalability , and cost efficiency . Chiplet technology enable producer to assemble multiple smaller semiconductor kick the bucket into a single parcel , allowing for better yields and shorten production cost .
This approach has become increasingly popular in the semiconducting material industry , in particular as chip designs become more complex and traditional grading methods face limit . By leveraging TSMC ’s manufacture processes , Nvidia could potentially enhance the index efficiency and processing capabilities of its future GPUs , attain them well - suited for AI , data centers , and gamy - performance computing .
It is rumour that Nvidia ’s Rubin GPUs will be produced using TSMC ’s advanced N3P process node . The N3P process is an optimized version of TSMC ’s 3 nanometre technology , delivering improved performance , power efficiency , and transistor concentration compared to its predecessors . This node is design to maximize the benefits of chiplet - based architecture , which could provide Nvidia to push the limits of GPU performance while sustain energy efficiency .
To further optimise the performance of its chiplet - based GPUs , Nvidia will also leverage TSMC ’s advanced packaging technology , including SoIC ( System - on - Integrated Chip ) . This applied science enables the vertical stacking of buffalo chip , raise power efficiency and bring down latency between unlike chiplets within the GPU . AMD has been using the same figure for its3D V - CacheCPUs for years .
TSMC is expected to ramp up its output with plans to importantly expatiate its SoIC capacity by the end of 2025 . As perWccftech , Nvidia ’s upcoming Rubin lineup is expected to utilize an SoIC design , leverage the capabilities of HBM4 memory . The Vera Rubin NVL144 chopine is rumored to sport a Rubin GPU with two reticule - sized dies , delivering up to 50 PFLOPs of FP4 carrying into action and 288 GB of next - generation HBM4 memory . Meanwhile , the higher - end NVL576 model is forestall to comprise a Rubin Ultra GPU with four reticle - sized dies , push performance to 100 PFLOPs of FP4 and trapping 1 TB of HBM4e computer storage diffuse across 16 HBM scads .
Nvidia ’s espousal of chiplet technology follows a broader industry drift where starring semiconductor equipment company , including AMD and Intel , have already integrated similar designs into their processors . The modular nature of chiplets allows manufacturers to conflate and match unlike processing unit , optimize performance for specific workloads . With AI and gamy - carrying out computation driving the demand for more hefty ironware , the partnership between TSMC and Nvidia is expected to yield find promotion in GPU design .
