The upcoming iteration of NVIDIA’s entry-level GPU might be the company’s strategic response to the evolving memory market landscape.
NVIDIA is reportedly preparing a revised version of its GeForce RTX 5050 graphics card, featuring an updated memory configuration. According to well-known insider kopite7kimi, this model is expected to come with 9GB of GDDR7 memory while retaining its current power consumption level of approximately 130W.
The existing RTX 5050 model uses 8GB of GDDR6 memory clocked at 20 Gbps, operating on a 128-bit bus, which provides a total bandwidth of around 320 GB/s. The new modification from NVIDIA is rumored to switch to three GDDR7 memory modules, each with 3GB capacity. This change would boost the total memory to 9GB, though it would simultaneously narrow the memory bus to 96 bits.
Despite the narrower bus, the faster GDDR7 memory is anticipated to largely offset this reduction. GDDR7 is expected to operate at speeds of approximately 28 Gbps, which would increase the overall memory bandwidth to about 336 GB/s. This represents a modest 5% improvement over the current version. Consequently, the video memory capacity would increase by 12.5%, accompanied by a slight boost in total bandwidth.
The graphics processor itself is rumored to remain unchanged, according to the leak. The graphics card will reportedly continue to utilize the GB206 chip, featuring 2560 CUDA cores.
The insider’s report confirms key specifications for the rumored model, identified as PG152-SKU40, based on the GB206-150-Kx-A1 chip, with 2560 FP32 cores, 96-bit GDDR7 9G memory, and a 130W TDP.
Media speculation suggests that the transition to GDDR7 might be driven by current conditions in the memory market. Supplies of GDDR6 for the mainstream segment are reportedly becoming less stable, whereas GDDR7 production is steadily increasing. Utilizing 3GB modules also allows for a reduction in the total number of modules, from four to three.
Rumors indicate that this new version of the GeForce RTX 5050 could be unveiled closer to Computex 2026, scheduled to take place from June 2nd to 5th in Taipei. NVIDIA has not yet officially confirmed the existence of such a modification.