AMD’s RDNA3 7900XT Flagship to Offer 384-bit Memory Bus, 24GB of VRAM

Now that AMD is finished launching its RDNA2 family of GPUs, speculation about what it has in store for next-gen is ramping up. As with all rumors this new info has to be taken with a grain of salt, but the latest details are juicy indeed. The big news is AMD is increasing both the amount of memory on its flagship GPU and memory bandwidth too. And that’s before we even begin discussing its monstrous amount of Infinity Cache.

This round of rumors comes from alleged insider @Greymon55 on Twitter. This time around they are tipping the memory bus width for AMD’s upcoming GPUs. They are known as Navi 31, 32, and 33, and go in descending order with Navi 31 being the RX 7900 XT. The top-shelf chip will reportedly be rocking an extra-wide (for AMD) 384-bit memory bus, which is a significant upgrade from the RX 6900 XT’s 256-bit bus. In the current generation of GPUs only Nvidia uses that wide of a memory bus. It uses it on its RTX 3080 12GB cards and those above it in the product stack. AMD has been able to get by with a narrower 256-bit bus since it has Infinity Cache to augment it.

Given the use of a wider memory bus, it’s also rumored it will be outfitting it with 24GB of GDDR6. This is an increase from 16GB in the current RX 6900/50 XT. This could help AMD compete against the RTX 4090, and it would announce AMD is done playing around. It never matched Nvidia’s RTX 3090/Ti 24GB, but it looks to be intent on taking up that fight with RDNA3.

In addition to a wider memory bus, and more memory, AMD is also increasing Navi 31’s Infinity Cache too. As shown in the mockup above, it’s believed AMD will include 64MB of infinity cache on each of the package’s six 4GB memory modules. Whether this is 3D stacked like on the Ryzen 7 5800X3D or not remains to be seen. This is assuming the package has seven chiplets as previously reported. This would grant the GPU a surprising 384MB of L3 cache. That’s a huge upgrade over to 128MB in the RX 6950 XT. This would make the Navi 31 a 4K gaming beast, with ample memory bandwidth for high-resolution, ray-traced gaming.

Of course, there’s still a lot of speculation about whether this 6+1 chiplet design will be the final Navi 31 package. Rumors have varied thus far about the design of its seven chiplets. What is clear is AMD has taken a revolutionary approach for Navi 31 versus an evolutionary approach like Nvidia will be doing with Ada Lovelace. Nvidia is sticking with a monolithic die this time around. AMD is taking a different path with a multi-chip module (MCM) approach. Which design will ultimately prove superior is something we’re all very curious to discover.

Unfortunately, it might be awhile until all the cards are on the table. Although Nvidia is tipped to launch its RTX 40-series in mid-July, AMD isn’t expected to launch RDNA3 any time soon. Rumors indicate it’ll likely launch close to its Zen 4 CPUs, which are due in the latter part of the year.

Continue reading

Nvidia Will Mimic AMD’s Smart Access Memory on Ampere: Report

AMD's Smart Access Memory hasn't even shipped yet, but Nvidia claims it can duplicate the feature.

Nvidia Unveils Ampere A100 80GB GPU With 2TB/s of Memory Bandwidth

Nvidia announced an 80GB Ampere A100 GPU this week, for AI software developers who really need some room to stretch their legs.

AMD Will Bring Smart Access Memory Support to Intel, Nvidia Hardware

AMD is reportedly working with Nvidia and Intel to bring hardware support for Smart Access Memory to other GPU and CPU platforms.

Tesla Ordered to Recall 150K+ Vehicles to Repair Memory Failures

Tesla has been asked — or "asked" — to recall some 159,000 vehicles to repair a NAND memory issue that will eventually cause failures on every affected vehicle.