Built around the GB207 die, the RTX 5050 has 2560 CUDA cores, 8 GB of GDDR6 memory and a 130W TDP. That puts it on paper exactly where the RTX 3050 left off, with the same core count, power draw and VRAM. But the devil’s in the speed bumps.

The memory runs at 20 Gbps, which, coupled with a 128-bit interface, gives it a bandwidth of 320 GB/s. That’s 43 per cent more than its predecessor and the same uplift over the RTX 4050’s GDDR6. The laptop version might get fancier GDDR7, but you’re stuck with GDDR6 on the desktop.

NVIDIA is pushing this as a proper Blackwell upgrade, touting 5th gen Tensor Cores, 4th gen ray tracing units and up to 421 TOPs of AI horsepower. It claims this is nearly six times more than the RTX 3050. Actual gaming numbers show more modest gains, with 50 to 60 per cent improvements in eSports staples like Apex Legends and CS2 and about four times the performance when DLSS 4 gets involved.

Clock speeds sit at 2.31 GHz base and 2.57 GHz boost, while power comes via a single 8-pin connector, with a 550W PSU recommended. There’s no Founders Edition on offer but AIB partners are expected to slap together their usual assortment of RGB-stuffed variants when the card hits shelves in the second half of July.

The RTX 5050 will retail at $249, lining it up against Intel’s Arc B580 and B570 GPUs, which offer more memory for similar money. It is also being positioned just beneath the RTX 5060, which carries a $299 tag.

NVIDIA is hoping that DLSS 4 and the faster memory will help the RTX 5050 stand out in what’s already a pretty crowded and cut-throat price segment. Whether gamers bite on yet another 8 GB card in 2025 remains to be seen.

NVIDIA GeForce RTX 50 GPU Specs