It’s the time of year when there are lots of rumors. One of the most trusted NVIDIA hardware leakers, Kopite7kimi, has started to share some fresh details, although there appears to be a bit of a twist in the narrative.

Going back to the 512-bit memory bus interface rumors. Kopite7kimi initially claimed that the flagship gaming GPU from the upcoming series would feature a 512-bit memory bus. This information was supposedly confirmed by multiple sources. However, a few months down the line, the tune changed, and it was suggested that it’s not 512-bit but rather 384-bit. Last week, the same Leaker hinted, that the RTX 50 and 40 series might share very similar memory bus widths, hinting at no significant changes for the flagship GPU known as GB202.

Now, Kopite7kimi is once again putting the 512-bit option on the table without delving into the details. It’s unclear whether NVIDIA is exploring different memory specs, or if the leaker is merely making educated guesses. Given the track record, one might lean towards the former option. What is clear though is that the details remain a bit elusive for now.

More importantly, Kopite is now claiming that the new series will stick to GDDR7 28 Gbps memory, outpacing GDDR6X specs by a large margin. NVIDIA will still have a lot of headroom for future upgrades, as GDDR7 memory is typically advertised at 32+Gbps speeds.

The 3GB GDDR7 modules are most likely not to be expected in the first RTX 50 lineup, but we know they are in the works and should launch next year, just as NVIDIA might launch their mid-range and entry-level GPUs. This opens the possibility of new memory capacity options, which shouldn’t be ruled out just yet.

At worst, NVIDIA will stick to the same memory bus widths on their next-gen gaming GPUs, which will still result in much higher bandwidth thanks to the faster GDDR7 memory. However, should NVIDIA for some reason explore the 512-bit option, that means the GPU will support even greater memory capacity and offer nearly 80% higher bandwidth than AD102.

Finally, the same leaker also mentioned that GB203 GPU might be a half of GB202, while not confirming or denying that GB202 uses a multi-chip package. The current AD102 flagship boasts 80% more cores compared to AD103. This has propelled RTX 4090(D) GPUs to an entirely different performance level. If NVIDIA takes a similar approach with Blackwell, we might witness a significant performance gap once more between the RTX 5090 and 5080 series, especially with the circulating rumors about a 512-bit bus.