Is this Nvidia's plan for their RTX 3090 Ti/Super? Thoughts on GDDR6X
Could Nvidia have a faster/cheaper RTX 3090-like product in the works?
Published: 4th September 2020 | Source: OC3D Opinion |
Is this Nvidia's plan for their RTX 3090 Ti/Super? Some thoughts on GDDR6X
What makes this 24GB graphics card extra crazy is its use of 1GB GDDR6X memory chips from Micron. 2GB/16Gb GRRD6X modules aren't due to be available from Micron until early 2021, forcing Nvidia to use 24 1GB/8Gb GDDR6 chips across their RTX 3090's 384-bit memory bus. This move means that all RTX 3090 series products must use GDDR6 memory chips on both sides of their PCB, as a 384-bit memory bus is designed to connect 12 32-bit GDDR6X chips in parallel.
With 24 GDDR6X DRAM chips, Nvidia's RTX 3090 has to dedicate a much larger portion of its power budget to VRAM than graphics cards like the RTX 3080. For comparison, the RTX 3080 only uses ten 1Gb GDDR6X memory chips to create its 10GB frame buffer, giving the RTX 3090 140% more DRAM chips to power.
An easy upgrade path for an RTX 3090 Ti/Super?
Micron has already confirmed that they have 21 Gbps GDDR6X chips and that the company has 2GB/16Gb GDDR6X memory in the works. This fact alone has gotten us to consider the future of Nvidia's Ampere architecture.
What if Nvidia decided to create an RTX 3090 Ti or RTX 3090 Super graphics card? What would Nvidia need to do to increase the performance of its upcoming RTX 30 series flagship? To us, the answer seems simple, change the RTX 3090's memory configuration.
Our hypothesis is that Nvidia will want to move to Micron's 2Gb memory modules whenever they become available in large quantities. This will allow Nvidia to create a 24GB RTX 3090 class product with half as many memory chips. This change should enable Nvidia to maintain the high capacities any future RTX 3090 class offering while powering fewer memory modules, which will allow more power to be pushed to other areas of the graphics card.
If Nvidia moved to 2GB/16Gb GDDR6X memory with 21Gbps speeds, the company would be able to offer over 1TB/s of memory bandwidth across 12 GDDR6X memory chips. If this also allows Nvidia to shift more power to their Ampere GA-102 core, this should also allow Nvidia to boost the clock speeds of newer Ampere offerings.
A side benefit of 2GB GDDR6X memory chips is that they will reduce the manufacturing complexity of future RTX 3090-class products, as fewer memory chips will need to be precision mounted during the manufacturing process. This alteration will slightly minimise manufacturing time and lowers the number of steps that are required to produce these graphics cards. These changes will reduce production costs and potentially lower product defect rates.
With an RTX 3090 Ti/Super, it is always possible that Nvidia will be able to enable more SMs to increase the CUDA core counts of refined GA-102 variants. Add that to Nvidia's potential use of faster GDDR6X memory and the prospect of lower VRAM power consumption, and Nvidia has the potential to release a much more powerful configuration of their RTX 3090.
With 2Gb GDDR6X memory chips being unavailable, Nvidia was forced to utilise twenty-four 1GB GDDR6X memory chips to create the RTX 3090's huge frame buffer. The addition of 2GB GDDR6X chips to the market could also allow manufacturers to develop 20GB versions of Nvidia's RTX 3080 in 2021.
While Nvidia RTX 3090 graphics card is already an incredibly powerful product offering, there is a change that the company has something more powerful in the works. That said, such a graphics card relies on GDDR6X memory chips that currently do not exist, at least in mass manufacturing quantities.
You can join the discussion on Nvidia's potential plans for an RTX 3090 Ti/Super on the OC3D Forums.