Intel Falcon Shores AI GPUs Reportedly Utilize TSMC’s 3nm & CoWoS Packaging

wccftech.com:

Intel's next-gen AI GPU, Falcon Shores, is expected to utilize TSMC's 3nm process, and may not use its own process technologies.

Intel's Next-Gen "Falcon Shores" AI GPUs To Be Made At TSMC Using 3nm Process Tech & CoWoS Packaging

The markets expected Intel's move, not only because the IFS is underperforming but also because the firm hasn't made strides in the AI industry through its Gaudi AI offerings.

Related Story Intel Core Ultra 9 285K “Arrow Lake” Flagship Desktop CPU Spotted: ES Chip With 24 Cores, 24 Threads, 76 MB Cache & 5 GHz Clocks

Intel's Falcon Shores GPUs will be the first true GPU IP for the AI segment based on the next-gen Xe graphics architecture, and the firm has hyped up the architecture immensely, claiming that it is far superior to Gaudi offerings, probably being the product that will potentially bring the "redemption" of Intel in the markets. Well, that's something that remains to be seen as Intel's Ponte Vecchio chip, despite being an engineering marvel, failed to amass a sizable customer base, and the one product where it was going to be a leadership design was delayed immensely (Aurora Supercomputer).

Taiwanese media outlet Cnyes reports that Intel has yet again opted for the "outsourcing" route with its upcoming Falcon Shores AI GPU generation, citing that Team Blue is eager to create a capable product, ready to compete with the likes of NVIDIA's Blackwell products. Intel plans on adopting an aggressive product portfolio with Falcon Shores, and rather than just "specializing" on a set of products, Intel plans on creating high, mid-range, and low-end variants, catering to every type of consumer in the AI segment. Having several variants will mean that Intel will be able to attract a diverse range of customers and also let them enter markets that are regulated heavily by the US, such as China.

Additionally, Intel plans to utilize TSMC's CoWoS-R packaging, which is known as one of the best solutions for making HPC/Datacenter chips on the market