AMD has taken an unexpected dig at Nvidia, well, the story goes that While Nvidia’s GeForce RTX 4070 has yet to be released, there have been numerous leaks about it, including the possibility that it will feature 12GB of GDDR6X VRAM. However, with some of the most recent games consuming memory (especially at 4K), AMD is using the opportunity ahead of the 4070’s release to poke fun at Nvidia and highlight the more generous amounts of VRAM it provides.
In the blog post “Are YOU an Enthusiast?” AMD product marketing manager Matthew Hummel emphasises the massive amounts of VRAM used at top specs at 4K for games such as the Resident Evil 4 remake, the recent port of The Last of Us Part I, and Hogwarts Legacy.
The short version is that AMD recommends its 6600-series with 8GB of RAM for each of these games at 1080p, its 6700-series with 12GB of RAM for 1440p, and the 16GB 6800/69xx-series for 4K.
AMD provided benchmark comparisons versus Nvidia chips, pitting its RX 6800 XT against Nvidia’s RTX 3070 Ti, the RX 6950 XT against the RTX 3080, the 7900 XT against the RTX 4070 Ti, and the RX 7900 XTX against the RTX 4080.
There are a lot of testing notes in this piece of marketing, as there are in any benchmark heavy piece of marketing. Those are reproduced in the gallery below. Depending on the claim, some interesting notes include different CPUs, motherboards, changes in RAM speeds, and more.
Although the majority of the gaming tests use the same specifications, other claims differ and some use older test data. The games aren’t the same in each GPU showdown, and it doesn’t appear that FSR or DLSS were used in these tests, for those who are interested in those technologies.
It’s also worth noting that VRAM capacity options are directly related to the size of the memory bus, or the number of channels preferred. Similarly to AMD’s Navi 22, which has a 192-bit 6-channel interface, Nvidia’s AD104 (RTX 4070 Ti and the upcoming RTX 4070) has a 192-bit bus.
With 2GB chips, Nvidia can only do up to 12GB — doubling that to 24GB is also possible, with memory in “clamshell” mode on both sides of the PCB, but that’s an expensive approach that’s typically only used in halo products and professional GPUs.
Also Read:
