Leaker reveals possible memory configurations of the upcoming NVIDIA GeForce RTX 5000

According to colleagues at Tom’s Hardware, leaker Kopite7kimi hinted on X that Nvidia’s upcoming RTX-50 series “Blackwell” GB203 and new GB205 chips will feature the same memory bus widths as Nvidia’s discontinued RTX-40 series AD103 and AD104 GPU chips – found in graphics cards like the RTX 4080 Super and RTX 4070 Super, some of the best graphics cards currently available.

Kopite7kimi’s latest post seems to be a confirmation of what he said a few days ago about Blackwell’s memory bus sizes. Rather than widening or shortening the bus width of its RTX-50 series GPUs, Kopite believes that Nvidia is happy with the memory bus width configurations it uses on its RTX-40 series Ada Lovelace GPUs and will carry them over to its new GPUs based on the Blackwell architecture. Kopite specifically states that at least two of the Blackwell GPUs, GB203 and GB205, will have the same memory bus widths as their predecessors. GB203 will be the direct successor to AD103, which is used in the RTX-4080 series and RTX-4070 Ti Super with a 256-bit interface.

Although I still have fantasies about 512 bit, the memory interface configuration of GB20x is not much different from that of AD10x.

— kopite7kimi (@kopite7kimi) March 9, 2024

GB205 is rather special as there were no “205” class GPUs in the past, but it is rumored to be a brand new GPU model that will not directly replace AD104. Still, Kopite believes it will have the same memory configuration as AD104 in the RTX-4070 series and the RTX-4070 Ti with a 192-bit interface. There is no mention of a GB204 chip, which would logically fall between the GB203 and GB205 in terms of specs. It’s possible Nvidia will skip this potential GPU chip entirely, just as there were no “AD105” or “GA105” GPUs in the last two generations.

I think GB203 is half of GB202, just like GB102 and GB100. But I don't know if GB202 has a multi chip package.

— kopite7kimi (@kopite7kimi) March 11, 2024

All of this is rumor and speculation, of course, so it should all be taken with a healthy dose of skepticism. What does this mean in terms of real-life performance? Not much on its own. It’s thought that Nvidia will be switching to the GDDR7 graphics memory standard with the RTX-50 series. GDDR7 is a brand new standard that was formalized by JEDEC a week ago and offers double the bandwidth potential of GDDR6X (let alone GDDR6). The immense bandwidth potential of GDDR7 could allow Nvidia to reduce/reuse its outbound memory interfaces on its next-generation GPUs while significantly increasing bandwidth.

I think my persistence is correct. So the difference is that GB202 is 512-bit and AD102 is 384-bit.

— kopite7kimi (@kopite7kimi) March 11, 2024

Combined with larger L2/L3 cache sizes, this is one reason why memory bus widths have shrunk in both Nvidia and AMD GPUs in recent years. Larger caches and newer memory technologies will allow Nvidia to reduce the bus width while increasing the effective memory bandwidth. It’s also important to note that GDDR7 opens the door to several other potential upgrades. While GDDR6 (and GDDR6X) theoretically reached a capacity of 32Gb per chip (4GB), GDDR7 will support up to 64Gb (8GB) memory sizes.

That’s probably still a long way off, and most GDDR6 solutions still use 16Gb (2GB) devices. Perhaps more importantly, GDDR7 will also support non-two-potential capacities such as 24Gb and 48Gb, and Samsung and Micron have already indicated that their first GDDR7 devices will reach up to 16Gb/24Gb. The use of 24Gb (3GB) devices would mean that GB203 could support up to 24GB and GB205 up to 18GB, with GB206/207 supporting 12GB. This would help alleviate VRAM memory concerns, even if the memory bus widths remain the same. In conclusion, Nvidia’s upcoming RTX-50 series promises interesting developments in memory bus widths and technologies. It remains to be seen how these changes will affect the actual performance and functionality of the graphics cards.

Source: Kopite7kimi via X via Toms Hardware