
Nvidia GeForce RTX 5050 and RTX 5060 series GPUs leaks, via renowned industry insider Kopite, reveal minimal changes across the board, especially in the VRAM department. Earlier today, the leaker spilled the beans surrounding Nvidia's upcoming budget GPU offerings in a series of tweets. From the looks of it, after three long years we're finally getting another 50-class GPU from Nvidia, but there isn't much to show for it.
As shown by the Steam Hardware Survey, Nvidia's 60-class GPUs are extremely popular among gamers, with flagship-grade SKUs nowhere in sight. While it's hard to estimate how much Nvidia rakes in from the budget segment, it undoubtedly plays a key role in shaping mind share and consumer sentiment. The cheapest GPU in the top-five list goes for $300. This speaks volumes about how much the average user is willing to spend on their graphics card.
As per their usual leaks, Kopite has detailed specifications of Nvidia's soon-to-be-announced budget GPUs, likely at GTC in a few days. The RTX 5060 Ti reportedly carries the GB206-300-A1 GPU core, with 4,608 CUDA cores (36 SMs) and a 128-bit memory interface. Going by rumored data from the same leaker, this should be a fully enabled GB206 die, coming in 8GB and 16GB (clamshell) flavors with a 180W TGP (Total Graphics Power). With 75W from the PCIe slot, the 5060 Ti falls comfortably within range of a single 8-pin connector (150W), even for some custom variants.
The vanilla RTX 5060 only purportedly offers an 8GB configuration, but that's expected since all these GPUs are limited to a 128-bit bus. With 3,840 CUDA cores (30 SMs), it is expected to be built using binned GB206 dies (GB206-250-A1) that otherwise did not qualify for the more powerful RTX 5060 Ti. The TGP has been upped to 150W from 115W, though you could tune that with a pinch of undervolting if needed. Nvidia might be able to extract a decent uplift versus the RTX 4060 considering the 25% more CUDA cores, 30% higher TGP, and other architectural refinements.
Lastly, the budget RTX 5050 drops to the entry-level GB207, with a fully functional die provided Kopite's data is solid. The GB207 design limits it to just 20 SMs or 2,560 CUDA cores. Assuming a 10% architectural uplift from Ada Lovelace to Blackwell, normalized for core counts and frequencies, the RTX 5050 might struggle against the RTX 4060 in certain scenarios. In any case, we'd love to see this GPU hit shelves for under $200, but let's not get ahead of ourselves.
Nvidia's GTC runs from March 17 to 21, where we can expect to learn more about budget Blackwell among other AI, data science, and robotics related developments.