Unofficial Specifications of the Nvidia RTX 5000 Series

Explore the latest unofficial specifications surrounding Nvidia's RTX 5000 series, delving into intriguing rumors and potential revelations

by Sededin Dedovic
Unofficial Specifications of the Nvidia RTX 5000 Series
© PNY Pro / YOutube channel

The next generation of Nvidia GeForce graphics cards remains a hot topic for rumors, and a new post on X has attracted a lot of attention as it reveals some potential specifications of the Blackwell GPUs. It comes from the regular "information supplier" Kopite7kimi (via VideoCardz), who discloses details about the number of Streaming Multiprocessors (SMs, which directly reflect the total number of cores), as well as the memory bus and type of memory across the entire product lineup.

Kopite7kimi states that GB202 – the most powerful chip for Blackwell that will power the RTX 5090 – purportedly will offer a maximum of 192 SMs. This is calculated using the sum of Graphics Processing Clusters (GPCs) multiplied by Texture Processing Clusters (TPCs), i.e., 12 x 8, resulting in 96, then doubled (assuming Nvidia sticks to 2 SMs for each TPC) to obtain a total of 192 SMs.

The memory bus of the RTX 5090 could reach 512 bits, as previously informed, although Nvidia doesn't have to push to the maximum capacities of the GB202 specifications – we'll get back to that in a moment. Next is GB203 (RTX 5080) equipped with (up to) 84 SMs and a 256-bit bus, followed by GB205 (likely RTX 5070) which has 50 SMs and 192 bits.

There's concern regarding the power of Nvidia's RTX 5070 GPU. With GB206 (perhaps RTX 5060 Ti?), Nvidia opts for 36 SMs and a 128-bit bus, and finally, GB207 (RTX 5060?) is equipped with 20 SMs. As a last detail for differentiation, this chip may stick with GDDR6 instead of transitioning to GDDR7 video memory like the rest of the Blackwell lineup.

Performance here isn't actually the central issue – it will more likely be the price of the next generation, but there's still no information on that. But let's get back to the details. Although the GPC and TPC mathematics might seem complex and confusing, it's not as important as comparing the number of SMs with the current generation of Lovelace graphics cards.

These figures are roughly in the same range, with significant exceptions at the top with GB202 (RTX 5090), which has a significantly higher number of SMs (192 for Blackwell compared to 144 for Lovelace) and what could be the RTX 5070 or GB205 (50 for Blackwell compared to 60 for Lovelace).

However, we must emphasize that the SM numbers given here are the total available for a given chip, so while GB202, for example, could be equipped with 192 SMs, the RTX 5090 may use a reduced version of that chip. In fact, this is precisely what's rumored – that the memory bus will be somewhat reduced as well (to 448 bits – and likely 28 GB VRAM, not the full 32 GB Nvidia could put if they wanted, at least theoretically).

The ‪@NVIDIA‬ RTX 5000 Ada Generation combines the latest-generation RT Cores, Tensor Cores, and CUDA cores with 32GB of gra© PNY Pro / YOutube channel

Nvidia might not use the full configuration of the GB202 processor on GeForce RTX 50 models?

Why reduce GB202 for RTX 5090? Because Nvidia doesn't have to use the full configuration, and that most powerful GB202 chip will likely be used without any restrictions in professional AI graphics cards – or perhaps reserved for a future RTX 5090 Ti (although that seems a bit unlikely, but you never know).

As for the potential performance of the RTX 5090, previous rumors certainly suggested a very powerful graphics card – around 50% faster than the RTX 4090, or even more, so there's no worry in that regard. Still, this rumor has caused some concern for some that the RTX 5070 seems somewhat weak, although even the RTX 4070 uses a greatly reduced chip (46 out of 60 enabled SMs).

Additionally, we're not aware of all the improvements in terms of GPU architecture that Blackwell brings, so speculating about performance at this stage, based on mere specifications, isn't particularly useful, as reported by Techradar.

This doesn't only apply to the RTX 5070, but also to the RTX 5080 which seems quite distant from the flagship model. Indeed, the lower-end Blackwell graphics cards on paper don't look very enticing, but let's wait for more details and reliable information before we start drawing conclusions.

The next generation of Nvidia GeForce GPUs, codenamed Blackwell, continues to fuel speculation with purported details about its architecture and performance. From the high-end RTX 5090 to the mid-range RTX 5060, each model offers varying specifications, raising questions about their power and price competitiveness.

While concerns linger about the balance between performance and cost, particularly with the RTX 5070, concrete conclusions remain elusive until official announcements unveil more comprehensive insights. Until then, enthusiasts and industry observers eagerly await further details to gauge the true potential of Nvidia's upcoming lineup in reshaping the landscape of gaming and professional graphics processing. There is a lot of excitement, and we hope that maybe we will get something better than announced.