Now that AMD is completed launching its RDNA2 household of GPUs, hypothesis about what it has in retailer for next-gen is ramping up. As with all rumors this new data needs to be taken with a grain of salt, however the newest particulars are juicy certainly. The large information is AMD is rising each the quantity of reminiscence on its flagship GPU and reminiscence bandwidth too. And that’s earlier than we even start discussing its monstrous quantity of Infinity Cache.
This spherical of rumors comes from alleged insider @Greymon55 on Twitter. This time round they’re tipping the reminiscence bus width for AMD’s upcoming GPUs. They’re generally known as Navi 31, 32, and 33, and go in descending order with Navi 31 being the RX 7900 XT. The highest-shelf chip will reportedly be rocking an extra-wide (for AMD) 384-bit reminiscence bus, which is a major improve from the RX 6900 XT’s 256-bit bus. Within the present era of GPUs solely Nvidia makes use of that extensive of a reminiscence bus. It makes use of it on its RTX 3080 12GB playing cards and people above it within the product stack. AMD has been capable of get by with a narrower 256-bit bus because it has Infinity Cache to enhance it.
Given the usage of a wider reminiscence bus, it’s additionally rumored will probably be outfitting it with 24GB of GDDR6. This is a rise from 16GB within the present RX 6900/50 XT. This might assist AMD compete in opposition to the RTX 4090, and it will announce AMD is finished enjoying round. It by no means matched Nvidia’s RTX 3090/Ti 24GB, nevertheless it appears to be like to be intent on taking over that struggle with RDNA3.
Along with a wider reminiscence bus, and extra reminiscence, AMD can also be rising Navi 31’s Infinity Cache too. As proven within the mockup above, it’s believed AMD will embody 64MB of infinity cache on every of the package deal’s six 4GB reminiscence modules. Whether or not that is 3D stacked like on the Ryzen 7 5800X3D or not stays to be seen. That is assuming the package deal has seven chiplets as beforehand reported. This is able to grant the GPU a stunning 384MB of L3 cache. That’s an enormous improve over to 128MB within the RX 6950 XT. This is able to make the Navi 31 a 4K gaming beast, with ample reminiscence bandwidth for high-resolution, ray-traced gaming.
In fact, there’s nonetheless numerous hypothesis about whether or not this 6+1 chiplet design would be the ultimate Navi 31 package deal. Rumors have various to this point concerning the design of its seven chiplets. What is obvious is AMD has taken a revolutionary method for Navi 31 versus an evolutionary method like Nvidia will probably be doing with Ada Lovelace. Nvidia is sticking with a monolithic die this time round. AMD is taking a unique path with a multi-chip module (MCM) method. Which design will finally show superior is one thing we’re all very curious to find.
Sadly, it may be awhile till all of the playing cards are on the desk. Though Nvidia is tipped to launch its RTX 40-series in mid-July, AMD isn’t anticipated to launch RDNA3 any time quickly. Rumors point out it’ll doubtless launch near its Zen 4 CPUs, that are due within the latter a part of the 12 months.