The Nvidia GeForce RTX 4070 (opens in new tab) is about to be launched. So AMD is trying to pull the first punch by criticizing the VRAM allocation of this new GPU. Not necessarily overtly, but implicitly. Also included are graphs.

In a new blog post (opens in new tab), AMD shows how 12GB of VRAM vs. 16GB of VRAM affects gaming at 4K. Needless to say, AMD's announced 12GB figure is ugly (via Tom's Hardware (opens in new tab)).

The GPUs AMD used for comparison were its own RX 6800 XT (open in new tab) and Nvidia's RTX 3070 Ti (open in new tab). The former has 16 GB and the latter 12 GB. However, with the RTX 4070 launch just a few days away, and with all the leaks predicting that it will offer 12GB of graphics memory, the timing of this important coup from AMD is unlikely to be a coincidence.

Interestingly, and as is becoming increasingly apparent, AMD has shown that there is often a large jump in VRAM usage when ray tracing is enabled. The ironic result of this is that with what is traditionally Nvidia's competitive advantage enabled, VRAM usage goes beyond 12 GB and reverses, with AMD GPUs outperforming Nvidia GPUs with ray tracing enabled.

AMD also absolutely underscores this point with the new 20GB RX 7900 XT (open in new tab) and 12GB RTX 4070 Ti (open in new tab), and several other GPU combinations where the AMD board provides more VRAM for the amount The numbers are favorable for the following.

All these joker cards are of course upscaling; Nvidia's DLSS scaling technology could be very important in the near future, and not just for the RTX 4070.

Current rumors suggest that the upcoming RTX 4060 and RTX 4060 Ti (open in new tab) will offer only 8GB of VRAM. This could be a limitation for native resolution games, even at 1080p for some titles. In that case, the competitiveness of these GPUs may depend on whether they use DLSS by default.

If VRAM limitations are an issue for Nvidia, the bottleneck is bus width; Nvidia has proven to be particularly aggressive this time around regarding bus width for its RTX 40 series GPUs, which are rumored to step down to 128 bits for the xx60 series cards The issue is that the bus width of the RTX 40 series GPUs will be 128 bits.

The question is how memory allocation will be determined by existing memory chip specifications. For example, the RTX 4070 Ti's 192-bit bus effectively limits Nvidia to a choice of 12GB or 24GB; 16GB is impossible with a 192-bit bus.

Similarly, the RTX 4060's rumored 128-bit bus is a choice between 8GB and 16GB; the latter is tricky for Nvidia, both in terms of cost and the potential for awkward comparisons within its own GPU lineup. Can they really offer more VRAM on the RTX 4060 than on the RTX 4070?

It remains to be seen exactly what specs the RTX 4070 and RTX 4060 will have on the desktop and, indeed, how impactful VRAM size will be for this generation of GPUs.

At this point, however, it is more likely than any other recent GPU generation from Nvidia and AMD that the major point of contention will be how much VRAM will be on the GPU. This is a point to keep an eye on.