With the increased threat from Intel in the form of their upcoming Xe-based ARC Alchemist gaming graphics cards, and AMD's upcoming RDNA 3-powered Radeon RX 7000 series GPUs... rumor has it NVIDIA is moving their next-gen Ada Lovelace GPU from Samsung to TSMC.
Originally, NVIDIA had planned to use Samsung's new 5nm node for their upcoming Ada Lovelace GPUs which will arrive as the GeForce RTX 40 series graphics cards -- but now leaker kopite7kimi has said that the new Ada Lovelace GPUs will "come out a little bit earlier" and that they will be made on TSMC's new 5nm node.
NVIDIA is making a monolithic GPU with its Ada Lovelace GPU architecture, with the flagship AD102 GPU powering the new flagship GeForce RTX 4090 graphics card.
According to the latest rumors, the new GeForce RTX 4090 would pack a huge 18432 CUDA cores, consume upwards of 450W of power, and be twice as fast as the GA102 -- making it twice as fast as the GeForce RTX 3090.
More NVIDIA Ada Lovelace + Hopper GPU content:
- Read more: NVIDIA might launch Hopper GPU instead of Ada Lovelace to beat RDNA 3
- Read more: NVIDIA's next-gen Ada Lovelace GPU: for Nintendo Super Switch, not PC?
- Read more: NVIDIA's next-gen Hopper GPU taping out soon: multi-chip module (MCM)
- Read more: NVIDIA rumored GPU refresh: GeForce RTX 30 SUPER coming soon
- Read more: NVIDIA Hopper GPU reportedly delayed, Lovelace GPU coming soon on 5nm
- Read more: NVIDIA's next-gen AD102 GPU: RTX 4090 twice as fast as the RTX 3090?!
The bigger note here is that NVIDIA will be using Micron's new GDDR6X memory that is clocked at a huge 24Gbps, up from the 19.5Gbps on the RTX 3090. The new AD102 GPU would have a 384-bit memory bus, so we should expect 1152GB/sec of memory bandwidth -- an increase of 23% over the 19.5Gbps on the GDDR6X memory used on the RTX 3090 (and its 936GB/sec of memory bandwidth).
NVIDIA has finished the Ada Lovelace GPU project, which is why we're hearing so much about it right now -- with the next phase in Ada Lovelace being the final design and tape out. We won't see it in our hands until the end of 2022 -- with estimates of Q4 2022, while performance-wise we can expect a 100%+ improvement over the GeForce RTX 3080 Ti and GeForce RTX 3090.
This should ensure 4K 120FPS+ gaming in the very next-gen games with ray tracing, and I'm sure we're going to be introduced to next-gen black magik DLSS 3.0 technology with Ada Lovelace which will only make things better again.
A year away from now is a long time, so we should expect refreshed Ampere GPUs in the form of the GeForce RTX 30 SUPER series graphics cards in between now and Q4 2022.
As for Micron's much-faster GDDR6X memory clocked at 24Gbps, that's something I reported on a year ago now, with Micron announcing that it would have 16Gb GDDR6X memory in 2021 that would be reaching 24Gbps. Micron explained in August 2020: "Micron's roadmap also highlights the potential for a 16Gb GDDR6X in 2021 with the ability to reach up to 24Gb/s. GDDR6X is powered by a revolutionary new PAM4 modulation technology for Ultra-Bandwidth Solutions. PAM4 has the potential to drive even more improvements in data rate".