NVIDIA 3nm GPUs Not Launching Till 2025, Expected To Utilize TSMC For Next-Gen Chips

NVIDIA Unlocks More Video Encoding Capabilities On Its GeForce GPUs Dating Back To Maxwell 1

NVIDIA's next-gen GPUs based on TSMC's 3nm process node might not be coming any time soon as reported by DigiTimes.

NVIDIA Pushes Back Next-Gen 3nm GPU Production At TSMC, Now Scheduled For 2025 Launch

According to the tech outlet quoting its own sources, NVIDIA is likely to push back its 3nm GPUs back until 2025. The reason for this pushback is said to be due to a slowed-down PC market and economic tensions which have gripped the globe.

NVIDIA won't be introducing any new GeForce GPUs this year and possibly even the next since they are yet to complete the transition to their brand new Ada Lovelace architecture. The GPU maker is going to unveil its mainstream parts in the coming month and we can expect slight refreshes down the road such as the RTX 4090 Ti and RTX 4080 Ti. The laptop segment can also see a refresh at the CES 2024 show floor.

The company however is reportedly stuck with lots of inventory and doesn't intend to produce lots of gaming GPUs till the supply starts moving. Recent price hikes on GeForce RTX GPUs per segment have made consumers unwilling to buy the mainstream cards as much as they did the last-gen. The existing stock of older GeForce RTX 30 series GPUs also doesn't make it easy since those cards offer better value and can be found in the used market for way lower prices.

NVIDIA estimates that it will not enter the 3nm generation until 2025.

Due to the slowdown of AMD, NVIDIA, which holds the dominant position in the GPU market, is still there. Currently, gaming GPUs use the 5/4nm family, while AI GPUs are 7nm and 5/4nm families. It is estimated that by 2025 Years will enter the 3nm generation.

Machine Translated DigiTimes

This year at GTC 2023, NVIDIA unveiled no new roadmap or plans to introduce a brand-new GPU. The company announced brand new Hopper products and volume production of its 4N powered Data Center chips which are now shipping to supercomputers, HPC customers, and the AI segment too. Considering that there is a healthy demand for Hopper which costs 10-20 times more than the average consumer GPU, the company is most likely going to focus its efforts on producing Hopper and delivering it to customers before moving to a brand-new GPU architecture.

With that said, NVIDIA's next-gen GPUs are already deep into development. We know that Blackwell is one of the many potential architecture codenames for the next-gen chips. The GPU is likely going to be announced at next year's GTC but won't see production until close to 2024 or even early 2025 which would be similar to Hopper which saw first volume shipments several quarters after its first reveal. According to leaker, Kopite7kimi, Blackwell GPUs may not utilize the 3nm process node and will be a monolithic design.

So there's a possibility that Blackwell could launch on TSMC's 4nm design which will still be an upgrade over the TSMC 4N which is an optimized 5nm node. The generation after Blackwell is possibly going to utilize a 3nm or sub 3nm process from TSMC.

The GPU giant and chipmaker are also going to leverage its own AI prowess to develop the next-gen GPU. With the help of Machine Learning, NVIDIA will be capable of making chips "Better Than Humans" and along with cuLitho, the same chips will be designed and manufactured 40x faster than traditional designs. NVIDIA is working with ASML and TSMC to help with the adoption of this technology.

Written by Hassan Mujtaba

Post a Comment

0 Comments