NVIDIA Takes No Chances Against AMD: Introducing RTX 5090 with 512-bit, RTX 5080 with 256-bit, and RTX 5060 with 192-bit – Will We Finally See 32GB, 16GB, and 8GB Versions?

A fresh and exciting leak has reached us today, as leaker Kopite7kimi, who has been inactive for some time, returns with data on the new RTX 50 series. Furthermore, almost everything is revealed except for the new technologies and specific details of these future graphics cards. For now, we can say that NVIDIA will postpone the 384-bit memory bus, leaving their GPUs as follows: RTX 5090 with 512 bits, RTX 5080 with 256 bits, and RTX 5060 with 192 bits. Can AMD compete with what’s coming?

To be completely honest, AMD will have a difficult time competing in this segment, not so much because of the memory bus configurations, but because of the Shader configurations and potential performance being discussed behind the scenes for the RTX 50 series.

NVIDIA RTX 5090: No 384 bits, but instead 512 bits and 32 GB of GDDR7

The rumor surfaced yesterday, but it wasn’t clear if it was accurate, as some informants were still saying that 384 bits (possibly for an RTX 5080 Ti or SUPER?) were the correct choice for Huang. Today, with the clout that Kopite7kimi has in the industry and his degree of accuracy, we must give more credibility to this leak.

Not only has he leaked the memory bus configurations, but also the GPC and SM configurations for each GPU. Let’s start with the RTX 5090, which will feature the GB202 chip no longer the GB1xx, which is reserved for AI and HPC GPUs. This chip will support 12 GPCs and 192 SMs, which means 16 SMs per GPC (8 TPC per GPC) (RTX 4090 has 12 SMs per GPC).

This implies that the RTX 5090 with its 512-bit bus will have 32 GB of GDDR7 at 28 Gbps and an impressive count of 24,576 shaders, dwarfing the RTX 4090 with its 16,384 shaders, confirming a 50% increase in shaders.

Therefore, this is the approximate improvement in performance between the two GPUs in traditional rasterization, as the architecture’s scalability in games is not as significant as it usually is.

NVIDIA RTX 5080, the mid-high range to compete with AMD, will have 256 bits

Nothing changes here. NVIDIA appears to be reserving the 384 bits for a hypothetical RTX 5080 Ti that could have 3 GB VRAM chips if AMD finally launches its RDNA 4 GPUs with MCM. Either way, the leaker has unveiled the maximum configuration of the GB203, which means 6 GPCs with 96 SMs.

The ratios are maintained again, with 16 SMs per GPC (again 8 TPC per GPC), making it quite dense. With that said, its bus will be 256 bits, suggesting that the RTX 5080 will feature 16 GB of GDDR7 at 28 Gbps, resulting in another interesting scenario as it will have 12,288 shaders beating the RTX 4080 SUPER by about 20%, if not more.

Additionally, a hypothetical RTX 5070 could arrive with fewer shaders on a trimmed GB203, with the same bus, same VRAM capacity, and speed, competing with the lower-end RX 8700 XT. But for now, these are just assumptions. Lastly, before closing this section, the rumors are confirmed once again, where this GPU has 50% fewer shaders than the RTX 5090, leaving a massive gap for the 384 bits in the RTX 5080 Ti.

NVIDIA RTX 5060, 192 bits on an unprecedented chip by naming convention: GB205

This is curiously the strangest information we can find. There will be a GB205 chip, being the first of the odd series in the latest-generation desktop GPUs. In Ada Lovelace, we had the AD107 for the RTX 4050 (not released in the PC version, but existed) and RTX 4060, but never a series 5, and this has been verified through all the chips since the nomenclatures changed.

In any case, all we know about this new GB205 guest is that its memory bus will be 192 bits, suggesting that in its full version, it will power the RTX 5060 and come with a maximum of 12 GB of VRAM at 28 Gbps (or lower, which would be strange).

Looking at all of this, you might be thinking that some games are poorly optimized and consume a lot of VRAM, leading to more of the same except for the RTX 5090. However, several factors need to be considered when discussing the final specifications, such as the increased bandwidth, increased performance, and increased L0, L1, and L2.

In short, it’s about taking the best features from the RTX 30 series and the RTX 40 series, then combining them into the RTX 50 series concept with a larger memory bus, larger caches, and faster memory. With that said, we are still waiting for information on the GB204, GB206, and GB207 chips, which will likely be scaled-down versions of the previously mentioned three.

The entry NVIDIA leaves no room for AMD: RTX 5090 with 512 bits, RTX 5080 with 256 bits, and RTX 5060 with 192 bits. Will we finally see 32 GB, 16 GB, and 8 GB options? first appeared on El Chapuzas Informático.