Nvidia GeForce RTX 50 Series GPUs are rumoured to be using the latest GDDR7 memory. Leakers are pointing at 28Gbps chips but can’t coordinate on the bus width. Regardless, Nvidia’s upcoming top dog is on rails to dwarf the beastly GeForce RTX 4090.
In response to a tweet from Andreas Schilling talking about flagship RTX 5000’s memory bandwidth, serial leaker kopite7kimi indicated that upcoming Nvidia GPUs may use 28Gbps GDDR7 chips. That said, while both leakers seem to be in accord regarding memory generation, Andreas Schilling thinks that a 384-bit wide interface is sufficiently fast for a flagship card. With such a bit-width and VRAM speed, the GPU would have access to a whopping 1,344GB/s of total bandwidth, before considering any optimisations in the pipeline. For reference, the RTX 4090 tops at 1,008GB/s.
Meanwhile, kopite7kimi doubles down on the initial claim from eight months ago talking about a much wider 512-bit bus. HBM aside, such a large bus would allow the GPU to have access to a record-high 1,792GB/s of bandwidth. That said, there is a reason why GPU manufacturers try to avoid going this big – costs and area allocation quickly rack up. As a reminder, the last consumer/non-HBM GPUs using a 512-bit bus were 2009’s GeForce GTX 285 using GDDR3 and 2015’s Radeon R9 390X using GDDR5.
Even if Nvidia chooses the ‘slower’ 384-bit bus, GDDR7 memory has not even debuted yet, meaning there is still a good margin for improving its speed. Moreover, if the rumours concerning AMD’s abandoning the high-end segment are true, there is no reason for Nvidia to go nuclear on VRAM speed. We may even see a Super variant down the line using newer 3GB GDDR7 chips, which would bring the total capacity to 48GB instead of 32GB. Lastly, GDDR7 modules are expected to reach 32Gbps in their lifetime.
Most commentators agree that Nvidia’s GeForce RTX 50 Series GPUs are due early year. Great times ahead.